feat: add core domain types for Photo Gallery SDK including media items, albums, and annotations

This commit is contained in:
2026-07-23 07:39:01 +05:30
parent 7bcd1a2a2d
commit 2613c767a6
107 changed files with 20699 additions and 7 deletions
+61
View File
@@ -0,0 +1,61 @@
/**
* Shared entry gate for every /api/gallery/ai/* route: authenticate, then
* throttle. Lives in a `_`-prefixed file so the App Router never treats it as a
* route (only `route.ts` defines an endpoint).
*
* Order matters: we authenticate FIRST so the rate limit can be keyed by
* principal rather than by a spoofable `x-forwarded-for` hop wherever possible.
* The session check is one cheap BFF round trip; the work it guards is a GPU
* call, so paying it before throttling is the right trade.
*
* SERVER-ONLY.
*/
import { NextResponse } from "next/server";
import { limit } from "@/lib/server/rate-limit";
import { rateLimitKey, requireGallerySession } from "@/lib/server/session";
/** Per-minute budgets, per the Smart Gallery route contract. */
export const RATE_LIMITS = {
classify: 30,
edit: 12,
tilt: 30,
transcribe: 20,
denoise: 20,
} as const;
const WINDOW_MS = 60_000;
export type GuardResult =
| { ok: true; principalId?: string }
/** Ready-to-return error response — the route should return it unchanged. */
| { ok: false; response: NextResponse };
/**
* @param route Which budget to apply (also namespaces the limiter key so a
* caller's `edit` spend does not consume their `classify` budget).
*/
export async function guard(req: Request, route: keyof typeof RATE_LIMITS): Promise<GuardResult> {
const session = await requireGallerySession(req);
if (!session.ok) {
return {
ok: false,
response: NextResponse.json({ error: session.error }, { status: session.status }),
};
}
const key = `${route}:${rateLimitKey(req, session.principalId)}`;
const { ok, retryAfter } = limit(key, RATE_LIMITS[route], WINDOW_MS);
if (!ok) {
return {
ok: false,
response: NextResponse.json(
{ error: "Too many requests — slow down." },
{ status: 429, headers: { "Retry-After": String(retryAfter) } },
),
};
}
return { ok: true, principalId: session.principalId };
}
+71
View File
@@ -0,0 +1,71 @@
import { type NextRequest, NextResponse } from "next/server";
import { RunpodError } from "@/lib/server/runpod/client";
import { rpDetect } from "@/lib/server/runpod/endpoints";
import { guard } from "../_guard";
export const runtime = "nodejs";
export const maxDuration = 60;
export const dynamic = "force-dynamic";
/**
* Object-detection proxy for the RunPod YOLO construction-material classifier (#1).
* The key + endpoint URL stay server-side. The client calls this only when
* NEXT_PUBLIC_APG_RUNPOD_DETECT is on; otherwise detection runs fully in-browser
* (COCO-SSD) with no server round-trip. Returns the SDK's DetectedObject[] shape
* (box as 0..1 fractions) so it drops straight into the Objects browser / smart
* albums / search.
*
* POST { imageBase64, width, height } -> { objects: [{ label, confidence, box }] }
* Auth: session-gated (see lib/server/session.ts). Rate limit: 30/min.
*/
const MAX_BASE64 = 4_000_000; // ~3 MB decoded — under serverless body limits
export async function POST(req: NextRequest) {
const gate = await guard(req, "classify");
if (!gate.ok) return gate.response;
if (!process.env.RUNPOD_API_KEY || !process.env.RUNPOD_YOLO_URL) {
return NextResponse.json(
{ error: "RunPod detection is not configured (set RUNPOD_API_KEY + RUNPOD_YOLO_URL)." },
{ status: 503 },
);
}
let body: unknown;
try {
body = await req.json();
} catch {
return NextResponse.json({ error: "Invalid request body." }, { status: 400 });
}
const { imageBase64, width, height } = (body ?? {}) as {
imageBase64?: unknown;
width?: unknown;
height?: unknown;
};
if (
typeof imageBase64 !== "string" ||
imageBase64.length === 0 ||
imageBase64.length > MAX_BASE64
) {
return NextResponse.json({ error: "Invalid or oversized image." }, { status: 400 });
}
const w = Number(width);
const h = Number(height);
try {
const objects = await rpDetect(
imageBase64,
Number.isFinite(w) && w > 0 ? w : 1,
Number.isFinite(h) && h > 0 ? h : 1,
);
return NextResponse.json({ objects });
} catch (err) {
const message = err instanceof Error ? err.message : "Detection failed.";
const status = err instanceof RunpodError ? err.status : 502;
return NextResponse.json({ error: message }, { status });
}
}
+48
View File
@@ -0,0 +1,48 @@
import { type NextRequest, NextResponse } from "next/server";
import { RunpodError } from "@/lib/server/runpod/client";
import { rpDenoiseAudio } from "@/lib/server/runpod/endpoints";
import { guard } from "../_guard";
export const runtime = "nodejs";
export const maxDuration = 60; // cold-start denoise worker can take a while
export const dynamic = "force-dynamic";
/**
* Audio noise-removal proxy. Accepts base64 WAV (48 kHz mono PCM16, produced
* in-browser) and returns a cleaned base64 WAV. Calls the RunPod audio-denoise
* endpoint (RUNPOD_AUDIO_DENOISE_URL) — key stays server-side. Used before
* transcription on noisy sites.
*
* POST { audio } -> { audio }
* Auth: session-gated. Rate limit: 20/min.
*/
const MAX_BASE64 = 12_000_000; // ~9 MB decoded WAV
export async function POST(req: NextRequest) {
const gate = await guard(req, "denoise");
if (!gate.ok) return gate.response;
let body: { audio?: unknown };
try {
body = await req.json();
} catch {
return NextResponse.json({ error: "Invalid JSON body." }, { status: 400 });
}
const audio = typeof body.audio === "string" ? body.audio : "";
if (!audio) return NextResponse.json({ error: "Missing audio." }, { status: 400 });
if (audio.length > MAX_BASE64) {
return NextResponse.json({ error: "Audio too long — keep it under ~30s." }, { status: 413 });
}
try {
const { audioB64 } = await rpDenoiseAudio(audio);
return NextResponse.json({ audio: audioB64 });
} catch (e) {
const msg = e instanceof RunpodError ? e.message : e instanceof Error ? e.message : "Denoise failed.";
return NextResponse.json({ error: msg }, { status: 502 });
}
}
+419
View File
@@ -0,0 +1,419 @@
import { type NextRequest, NextResponse } from "next/server";
import { RunpodError } from "@/lib/server/runpod/client";
import {
rpImg2Img,
rpInpaint,
rpRemoveBackground,
rpUpscale,
} from "@/lib/server/runpod/endpoints";
import { guard } from "../_guard";
export const runtime = "nodejs";
export const maxDuration = 60; // SD / cold-start models can take a while
export const dynamic = "force-dynamic";
/**
* Generative image-edit proxy. The BACKEND is pluggable — pick one with env
* `AI_EDIT_PROVIDER` (default `auto`):
*
* - `runpod` → RunPod serverless GPU endpoints (one per model). Maps each
* op → endpoint: restore/upscale → Real-ESRGAN (#7), colorize
* → img2img (#10), replace-sky / magic-eraser / generative-fill
* → SD 3.5 masked inpaint (#9), prompt → SD 3.5 img2img (#10).
* Env: RUNPOD_API_KEY + per-model RUNPOD_*_URL. Key stays
* server-side.
* - `local` → your own Stable Diffusion server (Automatic1111 / Forge /
* SD.Next img2img API). Env: LOCAL_SD_URL.
* - `huggingface` → Hugging Face Inference API. Env: HF_API_TOKEN, HF_IMAGE_MODEL.
* - `gemini` → Google Gemini image model (needs a billed key for image output).
* Env: GEMINI_API_KEY, GEMINI_IMAGE_MODEL.
* - `auto` → first configured of: runpod → local → huggingface → gemini.
*
* NOTE: `remove-background` runs in-browser by default (@imgly, no key), so it
* usually never reaches here. Object detection uses its own route (./classify).
*
* POST { imageBase64, mimeType?, op, maskBase64?, params? } -> { imageBase64, mimeType }
* Auth: session-gated. Rate limit: 12/min (the most expensive route).
*/
const OP_PROMPTS: Record<string, string> = {
restore:
"Restore and enhance this photograph: improve sharpness and clarity, correct exposure and white balance, reduce noise and compression artifacts, recover detail. Keep it natural and photorealistic.",
colorize: "Colorize this image with natural, realistic, well-balanced colors.",
"replace-sky":
"Replace the sky with a dramatic, beautiful golden-hour sky with soft clouds. Keep the foreground subject unchanged and the result photorealistic.",
};
const MAX_BASE64 = 4_000_000; // ~3 MB decoded — stays under serverless body limits
type Provider = "runpod" | "local" | "huggingface" | "gemini" | "none";
function resolveProvider(): Provider {
const explicit = (process.env.AI_EDIT_PROVIDER || "auto").toLowerCase();
if (
explicit === "runpod" ||
explicit === "local" ||
explicit === "huggingface" ||
explicit === "gemini"
)
return explicit;
if (explicit === "none") return "none";
// auto: prefer RunPod GPU endpoints, then a private local server, then HF, then Gemini.
// Detect RunPod when the key + ANY image endpoint URL is set (an upscale/colorize-only
// deployment is valid — not just the SD ones).
if (
process.env.RUNPOD_API_KEY &&
(process.env.RUNPOD_SD_IMG2IMG_URL ||
process.env.RUNPOD_SD_INPAINT_URL ||
process.env.RUNPOD_UPSCALE_URL ||
process.env.RUNPOD_COLORIZE_URL ||
process.env.RUNPOD_BG_REMOVE_URL)
)
return "runpod";
if (process.env.LOCAL_SD_URL) return "local";
if (process.env.HF_API_TOKEN) return "huggingface";
if (process.env.GEMINI_API_KEY) return "gemini";
return "none";
}
/** Ops that only the RunPod (mask/fixed-function) backend can serve. */
const RUNPOD_ONLY_OPS = new Set(["upscale", "magic-eraser", "generative-fill"]);
interface EditResult {
imageBase64: string;
mimeType: string;
}
export async function POST(req: NextRequest) {
const gate = await guard(req, "edit");
if (!gate.ok) return gate.response;
const provider = resolveProvider();
if (provider === "none") {
return NextResponse.json(
{
error:
"AI image editing is not configured. Set AI_EDIT_PROVIDER=runpod + RUNPOD_API_KEY + the per-model RUNPOD_*_URL vars (RunPod GPU), or LOCAL_SD_URL (own Stable Diffusion), HF_API_TOKEN (Hugging Face), or GEMINI_API_KEY. Background removal and all analysis still work with no key.",
},
{ status: 503 },
);
}
let body: unknown;
try {
body = await req.json();
} catch {
return NextResponse.json({ error: "Invalid request body." }, { status: 400 });
}
const { imageBase64, mimeType, op, maskBase64, params } = (body ?? {}) as {
imageBase64?: unknown;
mimeType?: unknown;
op?: { type?: string; prompt?: string; factor?: number };
maskBase64?: unknown;
params?: unknown;
};
if (typeof imageBase64 !== "string" || imageBase64.length === 0) {
return NextResponse.json({ error: "Invalid image." }, { status: 400 });
}
const hasMask = typeof maskBase64 === "string" && maskBase64.length > 0;
// Image + mask share one request body — budget them together against the cap.
if (imageBase64.length + (hasMask ? (maskBase64 as string).length : 0) > MAX_BASE64) {
return NextResponse.json(
{ error: "Image (plus mask) is too large — try a smaller image." },
{ status: 400 },
);
}
const safeMime =
typeof mimeType === "string" && /^image\/(jpeg|png|webp)$/.test(mimeType)
? mimeType
: "image/jpeg";
const opType = op?.type ?? "";
if (provider !== "runpod" && RUNPOD_ONLY_OPS.has(opType)) {
return NextResponse.json(
{ error: "This edit needs the RunPod backend (set AI_EDIT_PROVIDER=runpod)." },
{ status: 400 },
);
}
// Build the instruction from an allow-listed op (never trust arbitrary server prompts).
let instruction = "";
if (opType === "prompt" || opType === "generative-fill") {
const p = typeof op?.prompt === "string" ? op.prompt.trim() : "";
if (!p) return NextResponse.json({ error: "Empty prompt." }, { status: 400 });
instruction = p.slice(0, 500);
} else if (opType === "replace-sky") {
instruction =
typeof op?.prompt === "string" && op.prompt.trim()
? `Replace the sky with: ${op.prompt.trim().slice(0, 300)}. Keep the foreground unchanged and photorealistic.`
: OP_PROMPTS["replace-sky"]!;
} else if (opType === "magic-eraser") {
instruction =
"Fill the selected region with a clean, seamless, plausible background. Photorealistic.";
} else if (OP_PROMPTS[opType]) {
instruction = OP_PROMPTS[opType]!;
} else if (opType !== "upscale" && opType !== "remove-background") {
return NextResponse.json({ error: "Unsupported operation." }, { status: 400 });
}
try {
let result: EditResult;
if (provider === "runpod")
result = await editRunPod(
op ?? {},
imageBase64,
instruction,
hasMask ? (maskBase64 as string) : undefined,
params,
);
else if (provider === "local") result = await editLocal(instruction, imageBase64);
else if (provider === "huggingface") result = await editHuggingFace(instruction, imageBase64);
else result = await editGemini(instruction, imageBase64, safeMime);
return NextResponse.json(result);
} catch (err) {
const message = err instanceof Error ? err.message : "AI request failed.";
const status = err instanceof AiError || err instanceof RunpodError ? err.status : 502;
return NextResponse.json({ error: message }, { status });
}
}
// ---------------------------------------------------------------------------
// Backend: RunPod serverless GPU endpoints (one model per endpoint).
// Each op maps to its endpoint; the API key + URLs stay server-side.
// ---------------------------------------------------------------------------
interface SdParams {
negativePrompt?: string;
strength?: number;
steps?: number;
seed?: number;
guidanceScale?: number;
}
function sanitizeParams(raw: unknown): SdParams {
const p = (raw ?? {}) as Record<string, unknown>;
const out: SdParams = {};
if (typeof p.negativePrompt === "string" && p.negativePrompt.trim())
out.negativePrompt = p.negativePrompt.trim().slice(0, 300);
const strength = Number(p.strength);
if (Number.isFinite(strength)) out.strength = Math.max(0, Math.min(1, strength));
const steps = Number(p.steps);
if (Number.isFinite(steps)) out.steps = Math.max(1, Math.min(60, Math.round(steps)));
const guidance = Number(p.guidanceScale);
if (Number.isFinite(guidance)) out.guidanceScale = Math.max(1, Math.min(20, guidance));
const seed = Number(p.seed);
if (Number.isFinite(seed)) out.seed = Math.max(0, Math.min(2_147_483_647, Math.round(seed)));
return out;
}
async function editRunPod(
op: { type?: string; prompt?: string; factor?: number },
imageBase64: string,
instruction: string,
maskBase64: string | undefined,
rawParams: unknown,
): Promise<EditResult> {
const params = sanitizeParams(rawParams);
switch (op.type) {
case "remove-background":
// U²-Net via rembg (#6) — a real endpoint replacing the flaky in-browser remover.
return rpRemoveBackground(imageBase64);
case "restore":
// Real-ESRGAN (#7) with the GFPGAN face pass = "Restore & Enhance".
return rpUpscale(imageBase64, 4, true);
case "upscale":
return rpUpscale(imageBase64, op.factor === 4 ? 4 : 2, false);
case "colorize":
// The dedicated DDColor endpoint kept hard-crashing (modelscope). Route
// colorize through the img2img model as an instruction instead.
return rpImg2Img({ imageB64: imageBase64, prompt: instruction, ...params });
case "prompt":
return rpImg2Img({ imageB64: imageBase64, prompt: instruction, ...params }); // SD 3.5 img2img (#10)
case "replace-sky":
// True sky replacement is masked inpaint (#9). Without a mask (no in-app sky
// segmentation yet) degrade to a low-strength img2img (#10) so the foreground
// is mostly preserved.
if (maskBase64)
return rpInpaint({
imageB64: imageBase64,
maskB64: maskBase64,
prompt: instruction,
...params,
});
return rpImg2Img({
imageB64: imageBase64,
prompt: instruction,
...params,
strength: params.strength ?? 0.4,
});
case "magic-eraser":
case "generative-fill":
// SD 3.5 masked inpaint (#9) — white in the mask = the region to regenerate.
if (!maskBase64) throw new AiError("This edit needs a mask/selection.", 400);
return rpInpaint({
imageB64: imageBase64,
maskB64: maskBase64,
prompt: instruction,
...params,
});
default:
throw new AiError("Unsupported operation.", 400);
}
}
class AiError extends Error {
status: number;
constructor(message: string, status = 502) {
super(message);
this.status = status;
}
}
// ---------------------------------------------------------------------------
// Backend: local Stable Diffusion (Automatic1111 / Forge / SD.Next img2img API)
// ---------------------------------------------------------------------------
async function editLocal(instruction: string, imageBase64: string): Promise<EditResult> {
const base = process.env.LOCAL_SD_URL;
if (!base || !/^https?:\/\//i.test(base)) {
throw new AiError("LOCAL_SD_URL is not a valid http(s) URL.", 500);
}
const url = `${base.replace(/\/$/, "")}/sdapi/v1/img2img`;
let res: Response;
try {
res = await fetch(url, {
method: "POST",
headers: { "content-type": "application/json" },
body: JSON.stringify({
init_images: [imageBase64],
prompt: instruction,
denoising_strength: Number(process.env.LOCAL_SD_DENOISE ?? 0.55),
steps: Number(process.env.LOCAL_SD_STEPS ?? 25),
cfg_scale: 7,
sampler_name: process.env.LOCAL_SD_SAMPLER || "Euler a",
}),
cache: "no-store",
});
} catch {
throw new AiError("Could not reach your local Stable Diffusion server (LOCAL_SD_URL).", 502);
}
if (!res.ok) {
throw new AiError(`Local SD server error (${res.status}).`, 502);
}
const data = (await res.json().catch(() => null)) as { images?: string[] } | null;
const out = data?.images?.[0];
if (!out) throw new AiError("Local SD server did not return an image.", 502);
// A1111 returns raw base64 PNG (no data: prefix).
return { imageBase64: out.includes(",") ? out.split(",")[1]! : out, mimeType: "image/png" };
}
// ---------------------------------------------------------------------------
// Backend: Hugging Face Inference API — instruction image editing.
// ---------------------------------------------------------------------------
async function editHuggingFace(instruction: string, imageBase64: string): Promise<EditResult> {
const token = process.env.HF_API_TOKEN;
if (!token) throw new AiError("HF_API_TOKEN is not set.", 500);
const model = process.env.HF_IMAGE_MODEL || "timbrooks/instruct-pix2pix";
let res: Response;
try {
res = await fetch(`https://api-inference.huggingface.co/models/${model}`, {
method: "POST",
headers: {
authorization: `Bearer ${token}`,
"content-type": "application/json",
// Wait for the model to warm up instead of a fast 503.
"x-wait-for-model": "true",
},
body: JSON.stringify({
inputs: imageBase64,
parameters: { prompt: instruction, guidance_scale: 7, image_guidance_scale: 1.5 },
}),
cache: "no-store",
});
} catch {
throw new AiError("Could not reach the Hugging Face Inference API.", 502);
}
if (!res.ok) {
// Truncated on purpose — never surface a full upstream body.
const detail = (await res.text().catch(() => "")).slice(0, 160);
if (res.status === 503) throw new AiError("The model is loading — try again in ~20s.", 503);
throw new AiError(`Hugging Face error (${res.status}). ${detail}`, 502);
}
// Success returns raw image bytes.
const outMime = res.headers.get("content-type") || "image/png";
if (outMime.startsWith("application/json")) {
const j = (await res.json().catch(() => null)) as { error?: string } | null;
throw new AiError(
j?.error ? `Hugging Face: ${j.error}` : "Hugging Face returned no image.",
502,
);
}
const buf = await res.arrayBuffer();
return { imageBase64: Buffer.from(buf).toString("base64"), mimeType: outMime };
}
// ---------------------------------------------------------------------------
// Backend: Google Gemini image model (needs a billed key for image output).
// ---------------------------------------------------------------------------
async function editGemini(
instruction: string,
imageBase64: string,
safeMime: string,
): Promise<EditResult> {
const apiKey = process.env.GEMINI_API_KEY;
if (!apiKey) throw new AiError("GEMINI_API_KEY is not set.", 500);
const model = process.env.GEMINI_IMAGE_MODEL || "gemini-2.5-flash-image";
const prompt = `Edit this image as follows: ${instruction}. Preserve realism unless explicitly asked otherwise.`;
let res: Response;
try {
res = await fetch(
`https://generativelanguage.googleapis.com/v1beta/models/${model}:generateContent`,
{
method: "POST",
headers: { "content-type": "application/json", "x-goog-api-key": apiKey },
body: JSON.stringify({
contents: [
{
role: "user",
parts: [
{ inlineData: { mimeType: safeMime, data: imageBase64 } },
{ text: prompt },
],
},
],
generationConfig: { responseModalities: ["IMAGE"] },
}),
cache: "no-store",
},
);
} catch {
throw new AiError("Could not reach the AI service.", 502);
}
if (!res.ok) {
// Truncated on purpose — never surface a full upstream body.
const detail = (await res.text().catch(() => "")).slice(0, 160);
throw new AiError(`AI service error (${res.status}). ${detail}`, 502);
}
const data = (await res.json().catch(() => null)) as GeminiResponse | null;
const parts = data?.candidates?.[0]?.content?.parts ?? [];
const imgPart = parts.find((p) => p.inlineData?.data || p.inline_data?.data);
const out = imgPart?.inlineData?.data ?? imgPart?.inline_data?.data;
if (!out)
throw new AiError(
"The model did not return an image (the free Gemini tier has no image output — use LOCAL_SD_URL or HF_API_TOKEN instead).",
502,
);
const outMime = imgPart?.inlineData?.mimeType ?? imgPart?.inline_data?.mime_type ?? "image/png";
return { imageBase64: out, mimeType: outMime };
}
interface GeminiPart {
text?: string;
inlineData?: { mimeType?: string; data?: string };
inline_data?: { mime_type?: string; data?: string };
}
interface GeminiResponse {
candidates?: Array<{ content?: { parts?: GeminiPart[] } }>;
}
+48
View File
@@ -0,0 +1,48 @@
import { type NextRequest, NextResponse } from "next/server";
import { RunpodError } from "@/lib/server/runpod/client";
import { rpTilt } from "@/lib/server/runpod/endpoints";
import { guard } from "../_guard";
export const runtime = "nodejs";
export const maxDuration = 60; // cold-start tilt worker can take a while
export const dynamic = "force-dynamic";
/**
* Camera-tilt estimation proxy. Accepts a base64 image and returns
* {rollDegrees, pitchDegrees, fovDegrees} from the RunPod tilt endpoint
* (RUNPOD_TILT_URL) so the editor can auto-straighten.
*
* POST { image } -> { rollDegrees, pitchDegrees, fovDegrees }
* Auth: session-gated. Rate limit: 30/min.
*/
const MAX_BASE64 = 4_000_000; // ~3 MB decoded
export async function POST(req: NextRequest) {
const gate = await guard(req, "tilt");
if (!gate.ok) return gate.response;
let body: { image?: unknown };
try {
body = await req.json();
} catch {
return NextResponse.json({ error: "Invalid JSON body." }, { status: 400 });
}
const image = typeof body.image === "string" ? body.image : "";
if (!image) return NextResponse.json({ error: "Missing image." }, { status: 400 });
if (image.length > MAX_BASE64) {
return NextResponse.json({ error: "Image too large." }, { status: 413 });
}
try {
const tilt = await rpTilt(image);
return NextResponse.json(tilt);
} catch (e) {
const msg =
e instanceof RunpodError ? e.message : e instanceof Error ? e.message : "Tilt estimate failed.";
return NextResponse.json({ error: msg }, { status: 502 });
}
}
@@ -0,0 +1,49 @@
import { type NextRequest, NextResponse } from "next/server";
import { RunpodError } from "@/lib/server/runpod/client";
import { rpTranscribe } from "@/lib/server/runpod/endpoints";
import { guard } from "../_guard";
export const runtime = "nodejs";
export const maxDuration = 60; // cold-start STT worker can take a while
export const dynamic = "force-dynamic";
/**
* Speech-to-text proxy for voice annotations. Accepts base64 WAV (16 kHz mono
* PCM16, produced in-browser) and returns the transcript. Calls the RunPod
* voice-to-text endpoint (RUNPOD_STT_URL) — the key stays server-side.
*
* POST { audio, language? } -> { transcript, segments? }
* Auth: session-gated. Rate limit: 20/min.
*/
const MAX_BASE64 = 8_000_000; // ~6 MB decoded WAV — stays under serverless body limits
export async function POST(req: NextRequest) {
const gate = await guard(req, "transcribe");
if (!gate.ok) return gate.response;
let body: { audio?: unknown; language?: unknown };
try {
body = await req.json();
} catch {
return NextResponse.json({ error: "Invalid JSON body." }, { status: 400 });
}
const audio = typeof body.audio === "string" ? body.audio : "";
const language = typeof body.language === "string" ? body.language : undefined;
if (!audio) return NextResponse.json({ error: "Missing audio." }, { status: 400 });
if (audio.length > MAX_BASE64) {
return NextResponse.json({ error: "Audio too long — keep it under ~30s." }, { status: 413 });
}
try {
const { transcript, segments } = await rpTranscribe(audio, { language, punctuation: true });
return NextResponse.json({ transcript, segments });
} catch (e) {
const msg =
e instanceof RunpodError ? e.message : e instanceof Error ? e.message : "Transcription failed.";
return NextResponse.json({ error: msg }, { status: 502 });
}
}
+77 -1
View File
@@ -1140,4 +1140,80 @@
.dash-root .ai-bubble { max-width: 86%; }
.dash-root .ai-view { height: calc(100vh - 150px); }
}
/* =========================================================================
Smart Gallery — the embedded @photo-gallery/sdk surface.
The SDK is themed entirely through the token map in lib/gallery-api.ts
(--apg-* -> this file's own vars), so the rules below only handle the
host chrome: sizing, the demo banner, and the load/error placeholders.
========================================================================= */
/* The gallery is the one view that wants the whole viewport: it has its own sidebar, toolbar and
scrollers, so any height we leave on the table is wasted chrome. The old big PageHead cost ~90px;
a slim header (~40px) + compact banner + tight gaps hand almost all of that back to the shell.
96px = the dashboard's top padding + the slim header row; `.gal-shell` (flex:1; min-height:0)
consumes whatever is left after the header and the optional demo banner. */
.dash-root .gal { display: flex; flex-direction: column; gap: 10px; height: calc(100vh - 96px); min-height: 700px; }
/* Slim inline header — replaces the tall PageHead. One row, ~40px, so the shell keeps the height. */
.dash-root .gal-head { display: flex; align-items: center; gap: 10px; min-height: 36px; flex: 0 0 auto; }
.dash-root .gal-head-ic { width: 28px; height: 28px; border-radius: 9px; display: grid; place-items: center; color: #fff; background: var(--grad-brand); box-shadow: var(--glow-orange); flex: 0 0 auto; }
.dash-root .gal-head-title { font-size: 16px; font-weight: 700; line-height: 1.1; margin: 0; }
.dash-root .gal-head-sub { color: var(--muted); font-size: 12.5px; font-weight: 500; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; min-width: 0; }
@media (max-width: 720px) { .dash-root .gal-head-sub { display: none; } }
/* `.view` sets `z-index: 1`, which makes it a stacking context and traps the SDK's
full-screen overlays (lightbox z1000, editors z1050, camera z1080, modals z1100)
underneath the topbar's `z-index: 20`. Opting this one view out of the stacking
context lets those overlays cover the whole dashboard, as they must. The view still
paints above the ambient `.dash-content::before` glow because it follows it in the DOM. */
.dash-root .view.gal { z-index: auto; }
/* Compact single-line demo banner (~34px). Truncates rather than wrapping so it never steals a
second row of height from the shell. */
.dash-root .gal-banner { display: flex; align-items: center; gap: 8px; min-height: 34px; padding: 6px 12px; border-radius: 11px; border: 1px solid var(--border); background: color-mix(in srgb, var(--orange) 9%, var(--panel-2)); color: var(--text-2); font-size: 12px; font-weight: 500; flex: 0 0 auto; white-space: nowrap; overflow: hidden; }
.dash-root .gal-banner span { overflow: hidden; text-overflow: ellipsis; }
.dash-root .gal-banner svg { color: var(--orange); flex: 0 0 auto; }
/* The gallery's own viewport. `overflow: hidden` keeps the SDK's internal scrollers
in charge; its full-screen overlays (lightbox/editor/camera) are position:fixed
and deliberately escape this box to cover the whole dashboard. */
.dash-root .gal-shell { flex: 1; min-height: 0; position: relative; border-radius: 18px; border: 1px solid var(--border); background: var(--panel-2); overflow: hidden; box-shadow: var(--card-hi), 0 1px 2px rgba(0, 0, 0, 0.18); }
.dash-root[data-theme="dark"] .gal-shell { border: 0.5px solid #452b1a; border-radius: 20px; }
/* The SDK's embedded root fills this box. (--apg-overlay-top is set from the
component's `style` prop — the SDK writes an inline default that a stylesheet
rule could not override.) */
.dash-root .gal-shell .apg { height: 100%; }
/* ---- Fullscreen (the SDK puts `.apg--fullscreen` on its root: position:fixed; inset:0) ----
A position:fixed box is only clipped by an ancestor that is its CONTAINING BLOCK, which
`overflow`/`border-radius`/`box-shadow` alone never create — only transform / filter /
perspective / backdrop-filter / will-change / contain do. Nothing on the path
(.dash-content > .view.gal > .gal-shell) uses any of those: `.view`'s `ds-fade` animates
opacity only, and `.view.gal` already drops the `z-index: 1` stacking context. So the
fullscreen root does escape today — these rules make that survive an edit above. */
/* `.dash-root .gal-shell .apg` (0,3,0) would otherwise out-specify the SDK's own sizing; with
inset:0 driving the box, height must get out of the way. */
.dash-root .gal-shell .apg.apg--fullscreen {
height: auto;
/* Above the topbar (z-index: 20) and the sidebar, below the SDK's own overlays (1000+). */
z-index: 900;
}
/* Belt and braces: if a future rule ever DOES make `.gal-shell` a containing block, an
`overflow: hidden` on it would crop the fullscreen root to the embedded box. Drop the clip
(and the rounded corner it exists to enforce) for exactly as long as fullscreen is on. */
.dash-root .gal-shell:has(.apg--fullscreen) { overflow: visible; }
.dash-root .gal-placeholder { height: 100%; min-height: 320px; display: flex; flex-direction: column; align-items: center; justify-content: center; gap: 12px; text-align: center; padding: 28px; }
.dash-root .gal-placeholder-ic { width: 66px; height: 66px; border-radius: 20px; display: grid; place-items: center; color: #fff; background: var(--grad-brand); box-shadow: var(--glow-orange); }
.dash-root .gal-placeholder p { color: var(--muted); font-size: 13px; max-width: 420px; }
.dash-root .gal-placeholder h3 { font-size: 16px; font-weight: 700; }
.dash-root .gal-placeholder-error .gal-placeholder-ic { background: color-mix(in srgb, var(--red) 88%, #000); box-shadow: 0 10px 28px -12px color-mix(in srgb, var(--red) 60%, transparent); }
@media (max-width: 920px) {
/* Narrower chrome: a little less top offset, and a smaller floor so short viewports still work. */
.dash-root .gal { height: calc(100vh - 84px); min-height: 560px; }
}
+2
View File
@@ -17,6 +17,7 @@ import { AiAssistant } from "./ai-assistant";
import { TeamManagement } from "./team-management";
import { Messenger } from "./messenger";
import { Inbox } from "./inbox";
import { SmartGallery } from "./smart-gallery";
import "../../app/dashboard/dashboard.css";
export function Dashboard() {
@@ -49,6 +50,7 @@ export function Dashboard() {
: active === "ai" ? <AiAssistant />
: active === "messenger" ? <Messenger />
: active === "inbox" ? <Inbox />
: active === "gallery" ? <SmartGallery theme={theme} />
: active === "team" ? <TeamManagement />
: <ComingSoon title={title} icon={item?.icon ?? "dashboard"} onGo={setActive} />}
</ToastProvider>
+10 -1
View File
@@ -34,6 +34,7 @@ export const NAV_GROUPS: NavGroup[] = [
items: [
{ key: "messenger", label: "Messenger", icon: "send", subtitle: "Chat with your team and clients" },
{ key: "inbox", label: "Inbox", icon: "bell", subtitle: "Mentions, messages, alerts and mail — all in one" },
{ key: "gallery", label: "Smart Gallery", icon: "gallery", subtitle: "Every photo and video for your jobs — searchable, editable and shareable" },
],
},
{
@@ -79,7 +80,15 @@ export const NAV_ITEMS: NavItem[] = NAV_GROUPS.flatMap((g) => g.items);
// brand-new user with no membership); every other item requires membership, and the
// items mapped here additionally require the given permission. Unmapped items are
// shown to any member. This is UX only — be-crm still enforces every action.
const ALWAYS_VISIBLE = new Set(["dashboard", "profile", "messenger", "inbox"]);
//
// `gallery` DECISION: it stays in ALWAYS_VISIBLE (nav row always shown to members) rather than
// being gated here by `media.view`. The real access gate lives INSIDE the view (SmartGallery reads
// `useGalleryFeatures().canView`), which uses the permissive "member with zero media.* perms → all
// enabled" fallback. Gating the nav row here would use the stricter sidebar rule (member without the
// perm → hidden) and so would hide the gallery from freshly-seeded members before an admin has
// configured any Media perms — regressing the demo and the common member. So: always-visible row,
// real gating in-view. (be-crm enforces data access regardless of what the nav shows.)
const ALWAYS_VISIBLE = new Set(["dashboard", "profile", "messenger", "inbox", "gallery"]);
const NAV_PERMISSION: Record<string, string | undefined> = {
team: "team.manage",
people: "team.manage",
@@ -0,0 +1,69 @@
"use client";
// ============================================================
// The actual @photo-gallery/sdk mount. Split out of smart-gallery.tsx so it can be
// loaded with next/dynamic({ ssr: false }) — the SDK is browser-only (matchMedia,
// IndexedDB, Leaflet, canvas) and pulls in heavy optional ML models on demand, so it
// must stay out of the dashboard's initial bundle and out of the server render.
// ============================================================
import { useMemo } from "react";
import { PhotoGallery, type ViewId } from "@photo-gallery/sdk";
import { createCrmAIProvider } from "@/lib/gallery-ai";
import {
GALLERY_THEME_TOKENS,
useGalleryFeatures,
useGalleryLockProvider,
useGalleryStorage,
useGalleryUser,
} from "@/lib/gallery-api";
import "@photo-gallery/sdk/styles.css";
import "leaflet/dist/leaflet.css";
export interface SmartGalleryMountProps {
/** The dashboard's current appearance — the gallery must never diverge from the host. */
theme: "dark" | "light";
}
export default function SmartGalleryMount({ theme }: SmartGalleryMountProps) {
const { adapter } = useGalleryStorage();
const currentUser = useGalleryUser();
// Server-backed Recently Deleted lock when the Shell is wired; `undefined` in demo mode, which
// leaves the SDK on its own device-local lock (see useGalleryLockProvider).
const lockProvider = useGalleryLockProvider();
// Feature toggles resolved from the caller's CRM permissions (Media group). Superadmins/owners and
// the demo see everything; see useGalleryFeatures for the safe permissive fallback.
const { features } = useGalleryFeatures();
// Sidebar rows + Collections sections the CRM never wants to surface. The SDK hides both the row and
// the matching Collections section for each id. Screenshots + Documents aren't part of a roofing CRM.
const hiddenViews: ViewId[] = ["screenshots", "sys:documents"];
// One provider per mount. Every model is dynamically imported inside it, so constructing
// it is cheap; the weight only arrives when a photo is actually analyzed.
const ai = useMemo(() => createCrmAIProvider(), []);
return (
<PhotoGallery
embedded
adapter={adapter}
ai={ai}
// The CRM owns light/dark, so the in-gallery Appearance switcher is suppressed and the
// theme is driven straight off the dashboard's own toggle.
theme={theme}
chrome={{ titlebar: false, themeSwitcher: false }}
themeTokens={GALLERY_THEME_TOKENS}
borderRadius={12}
currentUser={currentUser}
lockProvider={lockProvider}
title="Smart Gallery"
// The SDK's floating Info panel defaults to 64px from the top — the height of its
// OWN toolbar. Inside the dashboard it has to clear the 84px CRM topbar instead.
// `style` is applied after the token vars, so this wins over the SDK's inline default.
style={{ ["--apg-overlay-top" as string]: "96px" }}
hiddenViews={hiddenViews}
features={features}
/>
);
}
+122
View File
@@ -0,0 +1,122 @@
"use client";
// ============================================================
// Smart Gallery — the tenant's media library, powered by @photo-gallery/sdk embedded
// inside the dashboard shell. The SDK owns the gallery experience (grid, lightbox,
// photo + video editors, map, people, versions, comments, AI); this file owns the
// LynkedUp chrome around it: page head, demo-mode banner, sizing, and failure
// containment so a gallery fault can never take the dashboard down.
//
// Storage + identity come from `@/lib/gallery-api` (be-crm data door when the Shell is
// configured, device-local otherwise). Theming comes from GALLERY_THEME_TOKENS, which
// maps the SDK's tokens onto this dashboard's own CSS variables.
// ============================================================
import { Component, type ErrorInfo, type ReactNode } from "react";
import dynamic from "next/dynamic";
import { Icon } from "./ui";
import { useGalleryFeatures, useGalleryStorage } from "@/lib/gallery-api";
const SmartGalleryMount = dynamic(() => import("./smart-gallery-mount"), {
ssr: false,
loading: () => <GalleryPlaceholder label="Loading your library…" />,
});
export function SmartGallery({ theme }: { theme: "dark" | "light" }) {
const { live } = useGalleryStorage();
// Whole-view gate. `media.view` with the same permissive fallback the feature toggles use, so a
// superadmin/owner and the demo always pass, and a member is only blocked if they hold some Media
// perms but not `media.view`. The sidebar keeps the row visible (see sidebar.tsx §4) — the real
// gate is here.
const { canView } = useGalleryFeatures();
return (
<div className="view gal">
{/* Slim header the big PageHead ate vertical space the gallery needs. The CRM topbar already
shows the "Smart Gallery" title; this compact row (~40px) just adds context and the icon. */}
<div className="gal-head">
<span className="gal-head-ic">
<Icon name="gallery" size={16} />
</span>
<h1 className="gal-head-title">Smart Gallery</h1>
<span className="gal-head-sub">Every photo and video for your jobs searchable, editable and shareable.</span>
</div>
{!live && (
<div className="gal-banner">
<Icon name="info" size={14} />
<span>Demo mode stored on this device only. It goes live once the Shell + be-crm are connected.</span>
</div>
)}
{canView ? (
<div className="gal-shell">
<GalleryBoundary>
<SmartGalleryMount theme={theme} />
</GalleryBoundary>
</div>
) : (
<div className="gal-shell">
<div className="gal-placeholder">
<span className="gal-placeholder-ic">
<Icon name="lock" size={28} />
</span>
<h3>You don&apos;t have access to the gallery</h3>
<p>Ask a workspace admin to grant you the &ldquo;View Smart Gallery&rdquo; permission.</p>
</div>
</div>
)}
</div>
);
}
/* ------------------------------------------------------------------ */
function GalleryPlaceholder({ label }: { label: string }) {
return (
<div className="gal-placeholder">
<span className="gal-placeholder-ic">
<Icon name="gallery" size={30} />
</span>
<p>{label}</p>
</div>
);
}
interface BoundaryState {
error: Error | null;
}
/**
* The gallery is a large third-party surface with canvas, workers and optional ML models. A render
* fault inside it must degrade to a message rather than blanking the whole dashboard, so it gets its
* own error boundary. (Error boundaries still require a class component in React 19.)
*/
class GalleryBoundary extends Component<{ children: ReactNode }, BoundaryState> {
state: BoundaryState = { error: null };
static getDerivedStateFromError(error: Error): BoundaryState {
return { error };
}
componentDidCatch(error: Error, info: ErrorInfo): void {
console.error("[smart-gallery] render failed", error, info.componentStack);
}
render(): ReactNode {
const { error } = this.state;
if (!error) return this.props.children;
return (
<div className="gal-placeholder gal-placeholder-error">
<span className="gal-placeholder-ic">
<Icon name="alert" size={30} />
</span>
<h3>The gallery could not be displayed</h3>
<p>{error.message || "An unexpected error occurred."}</p>
<button className="ds-btn v-outline s-sm" onClick={() => this.setState({ error: null })}>
<Icon name="refresh" size={14} /> Try again
</button>
</div>
);
}
}
+6 -1
View File
@@ -24,7 +24,8 @@ import {
LayoutDashboard, Building2, FolderKanban, UserPlus, BadgeCheck, Filter,
Truck, CloudLightning, Map as MapIcon, PenTool, Calculator, CalendarDays,
Trophy, ListChecks, Users, Settings, Sparkles, MoreHorizontal,
UsersRound, type LucideIcon,
UsersRound, Image as ImageIcon, Images, File, FolderOpen, Download,
Video, Play, LayoutGrid, ZoomIn, type LucideIcon,
} from "lucide-react";
/* ---------------------------------------------------------- */
@@ -50,6 +51,10 @@ const ICONS: Record<string, LucideIcon> = {
estimates: Calculator, schedule: CalendarDays, leaderboard: Trophy,
subtasks: ListChecks, people: Users, settings: Settings, ai: Sparkles,
team: UsersRound, dots: MoreHorizontal,
// media / gallery (also used by messenger.tsx, which already asks for image/file)
image: ImageIcon, gallery: Images, file: File, folder: FolderOpen,
download: Download, video: Video, play: Play, grid: LayoutGrid,
filter: Filter, zoom: ZoomIn, sparkle: Sparkles, "map-pin": MapPin,
};
export function Icon({ name, size = 18, className, strokeWidth = 2 }: { name: string; size?: number; className?: string; strokeWidth?: number }) {
+734
View File
@@ -0,0 +1,734 @@
"use client";
/**
* The CRM's AIProvider for the Smart Gallery.
*
* PROVENANCE: a port of the SDK demo's `createDemoAIProvider()` plus its
* clip/face/ocr/tensorflow/runpod-yolo providers and `imageEncode.ts` helpers
* (advance-photo-gallery-web-sdk/apps/web/src/lib/ai/*), collapsed into one
* module and retargeted from `/api/ai/*` to `/api/gallery/ai/*`.
*
* Capability split:
* - object detection: TensorFlow.js COCO-SSD, fully in-browser (no key), or the
* RunPod YOLO classifier via /api/gallery/ai/classify when
* NEXT_PUBLIC_APG_RUNPOD_DETECT=true (COCO-SSD is the automatic fallback)
* - face detection + recognition: face-api.js in-browser clustered into People
* - OCR: tesseract.js in-browser searchable text + the Documents album
* - semantic search: CLIP via transformers.js in-browser
* - background removal: @imgly in-browser WASM, or the RunPod U²-Net endpoint
* when NEXT_PUBLIC_APG_RUNPOD_BG=true (in-browser is the fallback)
* - other generative edits / transcription / denoise / tilt: proxied through the
* server routes so the RunPod key never reaches the browser
*
* EVERY heavy model is behind `await import(...)` so none of it lands in the
* initial bundle, and every capability degrades to []/''/null with a
* console.warn rather than throwing a failed model must never break the
* gallery UI.
*
* The in-browser models fetch weights from public CDNs (jsdelivr, huggingface,
* storage.googleapis.com, staticimgly.com). See docs/SMART_GALLERY.md for the
* list that would need CSP allow-listing.
*/
import type { AIProvider, GenerativeEditOp, MediaItem } from "@photo-gallery/sdk";
// Derived from the provider interface so we import only the three public types.
type DetectedObject = Awaited<ReturnType<NonNullable<AIProvider["detectObjects"]>>>[number];
type DetectedFace = Awaited<ReturnType<NonNullable<AIProvider["detectFaces"]>>>[number];
type ImageSource = ImageBitmap | HTMLImageElement;
const API = "/api/gallery/ai";
// ---------------------------------------------------------------------------
// imageEncode helpers (ported from apps/web/src/lib/ai/imageEncode.ts)
// ---------------------------------------------------------------------------
export interface EncodedImage {
/** base64 JPEG (no data: prefix). */
data: string;
mimeType: string;
/** Actual pixel dims of the encoded image (after downscale). */
width: number;
height: number;
}
/** Draw an image to a downscaled canvas and return base64 JPEG + its dims. */
export function imageToBase64(image: ImageSource, maxDim: number): EncodedImage {
const w = (image as HTMLImageElement).naturalWidth || (image as ImageBitmap).width;
const h = (image as HTMLImageElement).naturalHeight || (image as ImageBitmap).height;
const scale = Math.min(1, maxDim / Math.max(w, h));
const cw = Math.max(1, Math.round(w * scale));
const ch = Math.max(1, Math.round(h * scale));
const canvas = document.createElement("canvas");
canvas.width = cw;
canvas.height = ch;
const ctx = canvas.getContext("2d");
if (!ctx) throw new Error("Canvas not supported.");
ctx.drawImage(image as CanvasImageSource, 0, 0, cw, ch);
const dataUrl = canvas.toDataURL("image/jpeg", 0.9);
return { data: dataUrl.split(",")[1] ?? "", mimeType: "image/jpeg", width: cw, height: ch };
}
/**
* Rasterize a mask (ImageData, white = region to regenerate) to a PNG base64
* scaled to targetW×targetH so it matches the encoded image exactly SD 3.5
* requires image and mask to be identical pixel sizes.
*/
export function maskToBase64(mask: ImageData, targetW: number, targetH: number): string {
const tmp = document.createElement("canvas");
tmp.width = mask.width;
tmp.height = mask.height;
const tctx = tmp.getContext("2d");
if (!tctx) throw new Error("Canvas not supported.");
tctx.putImageData(mask, 0, 0);
const out = document.createElement("canvas");
out.width = targetW;
out.height = targetH;
const octx = out.getContext("2d");
if (!octx) throw new Error("Canvas not supported.");
// Nearest-neighbour, not bilinear — keep the mask strictly binary so SD gets
// crisp white(regenerate)/black(keep) edges instead of an anti-aliased grey halo.
octx.imageSmoothingEnabled = false;
octx.drawImage(tmp, 0, 0, targetW, targetH);
return out.toDataURL("image/png").split(",")[1] ?? "";
}
export function base64ToBlob(base64: string, mime: string): Blob {
const bin = atob(base64);
const bytes = new Uint8Array(bin.length);
for (let i = 0; i < bin.length; i++) bytes[i] = bin.charCodeAt(i);
return new Blob([bytes], { type: mime });
}
/**
* Pad an image with a neutral border for outpaint and return {imageBase64, maskBase64}
* as base64 PNG the border is WHITE in the mask (regenerate), the original image
* area BLACK (keep). Capped at 1280px on the long side.
*/
export function padForOutpaint(
image: ImageSource,
factor: number,
): { imageBase64: string; maskBase64: string } {
const w = (image as HTMLImageElement).naturalWidth || (image as ImageBitmap).width;
const h = (image as HTMLImageElement).naturalHeight || (image as ImageBitmap).height;
const f = Math.max(1.1, Math.min(2, factor));
const maxDim = 1280;
let pw = Math.round(w * f);
let ph = Math.round(h * f);
const scale = Math.min(1, maxDim / Math.max(pw, ph));
pw = Math.max(16, Math.round(pw * scale));
ph = Math.max(16, Math.round(ph * scale));
const iw = Math.max(1, Math.round(w * scale));
const ih = Math.max(1, Math.round(h * scale));
const ox = Math.floor((pw - iw) / 2);
const oy = Math.floor((ph - ih) / 2);
const imgCanvas = document.createElement("canvas");
imgCanvas.width = pw;
imgCanvas.height = ph;
const ictx = imgCanvas.getContext("2d");
if (!ictx) throw new Error("Canvas not supported.");
// Fill the new border with a blurred, stretched copy of the photo so the model
// has real color/context to continue from — flat gray gives it nothing.
ictx.filter = "blur(28px)";
ictx.drawImage(image as CanvasImageSource, 0, 0, pw, ph);
ictx.filter = "none";
ictx.drawImage(image as CanvasImageSource, ox, oy, iw, ih);
const maskCanvas = document.createElement("canvas");
maskCanvas.width = pw;
maskCanvas.height = ph;
const mctx = maskCanvas.getContext("2d");
if (!mctx) throw new Error("Canvas not supported.");
mctx.fillStyle = "#ffffff";
mctx.fillRect(0, 0, pw, ph);
mctx.fillStyle = "#000000";
mctx.fillRect(ox, oy, iw, ih);
return {
imageBase64: imgCanvas.toDataURL("image/png").split(",")[1] ?? "",
maskBase64: maskCanvas.toDataURL("image/png").split(",")[1] ?? "",
};
}
/** Draw an image to a canvas (downscaled) and return a JPEG Blob. */
function canvasBlob(image: ImageSource, maxDim: number): Promise<Blob> {
const w = (image as HTMLImageElement).naturalWidth || (image as ImageBitmap).width;
const h = (image as HTMLImageElement).naturalHeight || (image as ImageBitmap).height;
const scale = Math.min(1, maxDim / Math.max(w, h));
const cw = Math.max(1, Math.round(w * scale));
const ch = Math.max(1, Math.round(h * scale));
const canvas = document.createElement("canvas");
canvas.width = cw;
canvas.height = ch;
const ctx = canvas.getContext("2d");
if (!ctx) return Promise.reject(new Error("Canvas not supported."));
ctx.drawImage(image as CanvasImageSource, 0, 0, cw, ch);
return new Promise((resolve, reject) =>
canvas.toBlob((b) => (b ? resolve(b) : reject(new Error("toBlob failed"))), "image/jpeg", 0.92),
);
}
function clamp01(n: number): number {
return n < 0 ? 0 : n > 1 ? 1 : n;
}
// ---------------------------------------------------------------------------
// Object detection — TensorFlow.js COCO-SSD, in-browser
// ---------------------------------------------------------------------------
interface CocoPrediction {
bbox: [number, number, number, number];
class: string;
score: number;
}
interface CocoModel {
detect(
img: HTMLImageElement | HTMLCanvasElement | HTMLVideoElement,
maxNumBoxes?: number,
minScore?: number,
): Promise<CocoPrediction[]>;
}
let cocoPromise: Promise<CocoModel | null> | null = null;
/** Load tfjs + COCO-SSD exactly once; resolves to null if anything fails. */
function ensureCoco(): Promise<CocoModel | null> {
cocoPromise ??= (async () => {
try {
const tf = await import("@tensorflow/tfjs");
try {
await tf.setBackend("webgl");
} catch {
// Fall back to the default backend if WebGL is unavailable.
}
await tf.ready();
const cocoSsd = await import("@tensorflow-models/coco-ssd");
return (await cocoSsd.load({ base: "lite_mobilenet_v2" })) as unknown as CocoModel;
} catch (err) {
console.warn("[gallery-ai] COCO-SSD load failed; object detection disabled.", err);
return null;
}
})();
return cocoPromise;
}
async function detectObjectsInBrowser(
item: MediaItem,
image: ImageSource,
): Promise<DetectedObject[]> {
const model = await ensureCoco();
if (!model) return [];
try {
const el = image as HTMLImageElement;
const w = el.naturalWidth || el.width || item.width || 1;
const h = el.naturalHeight || el.height || item.height || 1;
const predictions = await model.detect(el, 20, 0.4);
return predictions.map((p) => ({
label: p.class,
confidence: p.score,
box: { x: p.bbox[0] / w, y: p.bbox[1] / h, width: p.bbox[2] / w, height: p.bbox[3] / h },
})) as DetectedObject[];
} catch (err) {
console.warn("[gallery-ai] object detection failed.", err);
return [];
}
}
/**
* Server-side YOLO detection via /api/gallery/ai/classify, with the in-browser
* COCO-SSD as the automatic fallback so detection never hard-fails.
*/
async function detectObjectsViaRunpod(
item: MediaItem,
image: ImageSource,
): Promise<DetectedObject[]> {
try {
const { data, mimeType, width, height } = imageToBase64(image, 1280);
const res = await fetch(`${API}/classify`, {
method: "POST",
headers: { "content-type": "application/json" },
body: JSON.stringify({ imageBase64: data, mimeType, width, height }),
});
if (!res.ok) throw new Error(`classify failed (${res.status})`);
const { objects } = (await res.json()) as { objects?: DetectedObject[] };
if (Array.isArray(objects)) return objects;
throw new Error("classify returned no objects");
} catch (err) {
console.warn("[gallery-ai] RunPod detection failed; falling back to COCO-SSD.", err);
return detectObjectsInBrowser(item, image);
}
}
// ---------------------------------------------------------------------------
// Faces — @vladmandic/face-api, in-browser (128-D descriptors → People)
// ---------------------------------------------------------------------------
const FACE_MODEL_URL = "https://cdn.jsdelivr.net/npm/@vladmandic/face-api@1.7.15/model";
type FaceApi = typeof import("@vladmandic/face-api");
let facePromise: Promise<FaceApi | null> | null = null;
function ensureFaceModels(): Promise<FaceApi | null> {
facePromise ??= (async () => {
try {
const faceapi = await import("@vladmandic/face-api");
// The bundled tf re-export is typed narrowly; backend control lives on the
// runtime object. Prefer WebGL (no eval; CSP-friendly), fall back gracefully.
const tf = faceapi.tf as unknown as {
setBackend: (b: string) => Promise<boolean>;
ready: () => Promise<void>;
};
try {
await tf.setBackend("webgl");
} catch {
/* keep default backend */
}
await tf.ready();
await Promise.all([
faceapi.nets.tinyFaceDetector.loadFromUri(FACE_MODEL_URL),
faceapi.nets.faceLandmark68Net.loadFromUri(FACE_MODEL_URL),
faceapi.nets.faceRecognitionNet.loadFromUri(FACE_MODEL_URL),
]);
return faceapi;
} catch (err) {
// Degrade gracefully — People simply stays empty if models can't load.
console.warn("[gallery-ai] face model load failed; face clustering disabled.", err);
return null;
}
})();
return facePromise;
}
let faceWarned = false;
async function detectFaces(item: MediaItem, image: ImageSource): Promise<DetectedFace[]> {
const faceapi = await ensureFaceModels();
if (!faceapi) return [];
const w = (image as HTMLImageElement).naturalWidth || (image as ImageBitmap).width || 1;
const h = (image as HTMLImageElement).naturalHeight || (image as ImageBitmap).height || 1;
type TNetInput = Parameters<FaceApi["detectAllFaces"]>[0];
let results;
try {
results = await faceapi
.detectAllFaces(
image as unknown as TNetInput,
new faceapi.TinyFaceDetectorOptions({ inputSize: 416, scoreThreshold: 0.5 }),
)
.withFaceLandmarks()
.withFaceDescriptors();
} catch (err) {
if (!faceWarned) {
faceWarned = true;
console.warn("[gallery-ai] face detection failed on", item.name, err);
}
return [];
}
return results.map((r) => {
const b = r.detection.box;
return {
confidence: r.detection.score,
box: {
x: clamp01(b.x / w),
y: clamp01(b.y / h),
width: clamp01(b.width / w),
height: clamp01(b.height / h),
},
embedding: Array.from(r.descriptor),
};
}) as DetectedFace[];
}
// ---------------------------------------------------------------------------
// OCR — tesseract.js, in-browser
// ---------------------------------------------------------------------------
// Must equal the EXACT tesseract.js version in package.json (pinned, no caret)
// so the worker CDN URL can never drift from the installed main-thread code.
const TESSERACT_VERSION = "5.1.1";
const WORKER_PATH = `https://cdn.jsdelivr.net/npm/tesseract.js@${TESSERACT_VERSION}/dist/worker.min.js`;
const CORE_PATH = "https://cdn.jsdelivr.net/npm/tesseract.js-core@5";
// jsDelivr's GitHub mirror of naptha/tessdata (same files as projectnaptha.com),
// so every asset comes from ONE host that a CSP can allow-list.
const LANG_PATH = "https://cdn.jsdelivr.net/gh/naptha/tessdata@gh-pages/4.0.0";
interface OcrWord {
text?: string;
confidence?: number;
}
interface OcrData {
text?: string;
confidence?: number;
words?: OcrWord[];
blocks?: Array<{ paragraphs?: Array<{ lines?: Array<{ words?: OcrWord[] }> }> }> | null;
}
type TesseractWorker = import("tesseract.js").Worker;
let ocrWorkerPromise: Promise<TesseractWorker | null> | null = null;
function ensureOcrWorker(): Promise<TesseractWorker | null> {
ocrWorkerPromise ??= (async () => {
try {
const { createWorker } = await import("tesseract.js");
// v5: createWorker(langs, oem, options) already loads + initializes the
// language internally — do NOT call the removed v4 worker.load().
return await createWorker("eng", 1, {
workerPath: WORKER_PATH,
corePath: CORE_PATH,
langPath: LANG_PATH,
});
} catch (err) {
console.warn("[gallery-ai] tesseract worker init failed; OCR disabled.", err);
return null;
}
})();
return ocrWorkerPromise;
}
const WORD_CONFIDENCE = 70; // a word tesseract is actually sure about
const MIN_WORDS = 4; // need several confident words to call it a document
const MIN_CHARS = 10;
function collectWords(data: OcrData): OcrWord[] {
if (Array.isArray(data.words) && data.words.length) return data.words;
const out: OcrWord[] = [];
for (const b of data.blocks ?? [])
for (const p of b.paragraphs ?? [])
for (const l of p.lines ?? []) for (const w of l.words ?? []) out.push(w);
return out;
}
/**
* Return real text or '' (not a document). tesseract hallucinates low-confidence
* gibberish for photos with no text, so we keep only high-confidence, word-shaped
* tokens and require several of them.
*/
function meaningfulText(data: OcrData): string {
const words = collectWords(data);
if (words.length > 0) {
const good = words.filter(
(w) => (w.confidence ?? 0) >= WORD_CONFIDENCE && /[A-Za-z0-9]{2,}/.test(w.text ?? ""),
);
const text = good
.map((w) => (w.text ?? "").trim())
.filter(Boolean)
.join(" ")
.trim();
return good.length >= MIN_WORDS && text.length >= MIN_CHARS ? text : "";
}
// Fallback: overall confidence + count of word-shaped tokens.
const raw = (data.text ?? "").trim();
const conf = typeof data.confidence === "number" ? data.confidence : 0;
const realWords = raw.match(/[A-Za-z]{3,}/g) ?? [];
return conf >= 72 && realWords.length >= 6 ? raw : "";
}
async function ocr(_item: MediaItem, image: ImageSource): Promise<string> {
const worker = await ensureOcrWorker();
if (!worker) return "";
try {
// Request the block hierarchy so per-word confidence is available.
const { data } = (await worker.recognize(
image as unknown as HTMLImageElement,
{},
{ text: true, blocks: true },
)) as { data: OcrData };
return meaningfulText(data);
} catch (err) {
console.warn("[gallery-ai] OCR failed.", err);
return "";
}
}
// ---------------------------------------------------------------------------
// Semantic search — CLIP via transformers.js (ONNX-WASM), in-browser
// ---------------------------------------------------------------------------
const CLIP_MODEL_ID = "Xenova/clip-vit-base-patch16";
type Transformers = typeof import("@huggingface/transformers");
let transformersMod: Transformers | null = null;
/* eslint-disable @typescript-eslint/no-explicit-any -- transformers.js pipelines are untyped */
let clipVisionPromise: Promise<{ processor: any; model: any } | null> | null = null;
let clipTextPromise: Promise<{ tokenizer: any; model: any } | null> | null = null;
async function loadTransformers(): Promise<Transformers> {
if (!transformersMod) {
transformersMod = await import("@huggingface/transformers");
// Remote-only (models from the HF CDN); rely on browser cache between sessions.
transformersMod.env.allowLocalModels = false;
}
return transformersMod;
}
function ensureClipVision() {
clipVisionPromise ??= (async () => {
try {
const tf = await loadTransformers();
const [processor, model] = await Promise.all([
tf.AutoProcessor.from_pretrained(CLIP_MODEL_ID),
tf.CLIPVisionModelWithProjection.from_pretrained(CLIP_MODEL_ID),
]);
return { processor, model };
} catch (err) {
console.warn("[gallery-ai] CLIP vision load failed; semantic search disabled.", err);
return null;
}
})();
return clipVisionPromise;
}
function ensureClipText() {
clipTextPromise ??= (async () => {
try {
const tf = await loadTransformers();
const [tokenizer, model] = await Promise.all([
tf.AutoTokenizer.from_pretrained(CLIP_MODEL_ID),
tf.CLIPTextModelWithProjection.from_pretrained(CLIP_MODEL_ID),
]);
return { tokenizer, model };
} catch (err) {
console.warn("[gallery-ai] CLIP text load failed; semantic search disabled.", err);
return null;
}
})();
return clipTextPromise;
}
/** Draw an image onto a canvas (downscaled) for the CLIP image processor. */
function toCanvas(image: ImageSource, maxDim = 384): HTMLCanvasElement {
const w = (image as HTMLImageElement).naturalWidth || (image as ImageBitmap).width;
const h = (image as HTMLImageElement).naturalHeight || (image as ImageBitmap).height;
const scale = Math.min(1, maxDim / Math.max(w, h));
const canvas = document.createElement("canvas");
canvas.width = Math.max(1, Math.round(w * scale));
canvas.height = Math.max(1, Math.round(h * scale));
const ctx = canvas.getContext("2d");
if (!ctx) throw new Error("Canvas not supported.");
ctx.drawImage(image as CanvasImageSource, 0, 0, canvas.width, canvas.height);
return canvas;
}
function tensorToArray(t: any): number[] {
const data: Float32Array = t?.data ?? t;
return Array.from(data as ArrayLike<number>);
}
async function embedImage(_item: MediaItem, image: ImageSource): Promise<number[]> {
const v = await ensureClipVision();
if (!v) return [];
try {
const tf = await loadTransformers();
const canvas = toCanvas(image);
const ctx = canvas.getContext("2d");
if (!ctx) return [];
const imageData = ctx.getImageData(0, 0, canvas.width, canvas.height);
const raw = new tf.RawImage(imageData.data, canvas.width, canvas.height, 4).rgb();
const inputs = await v.processor(raw);
const out = await v.model(inputs);
return tensorToArray(out.image_embeds);
} catch (err) {
console.warn("[gallery-ai] embedImage failed.", err);
return [];
}
}
async function embedText(query: string): Promise<number[]> {
const t = await ensureClipText();
if (!t) return [];
try {
const inputs = t.tokenizer([query], { padding: true, truncation: true });
const out = await t.model(inputs);
return tensorToArray(out.text_embeds);
} catch (err) {
console.warn("[gallery-ai] embedText failed.", err);
return [];
}
}
/* eslint-enable @typescript-eslint/no-explicit-any */
// ---------------------------------------------------------------------------
// The provider
// ---------------------------------------------------------------------------
/** `true` only for the literal string "true", matching the SDK demo's semantics. */
function flag(v: string | undefined): boolean {
return v === "true";
}
export function createCrmAIProvider(): AIProvider {
// NEXT_PUBLIC_* are inlined at build time, so these must be read as full
// static member expressions — do NOT refactor to dynamic indexing.
const useRunpodDetect = flag(process.env.NEXT_PUBLIC_APG_RUNPOD_DETECT);
const useRunpodBg = flag(process.env.NEXT_PUBLIC_APG_RUNPOD_BG);
const useRunpodTilt = flag(process.env.NEXT_PUBLIC_APG_RUNPOD_TILT);
return {
name: "crm-ai (coco-ssd/yolo + face-api + tesseract + clip + runpod-edit)",
detectObjects: useRunpodDetect ? detectObjectsViaRunpod : detectObjectsInBrowser,
detectFaces,
ocr,
embedImage,
embedText,
async generativeEdit(item: MediaItem, image: ImageSource, op: GenerativeEditOp) {
// Remove Background runs fully in-browser (no key) via @imgly — works even
// with no backend. Other ops go through the /api/gallery/ai/edit route.
if (op.type === "remove-background") {
// Prefer the RunPod U²-Net endpoint when enabled; fall back to in-browser
// @imgly if it's off or the request fails, so this always produces a result.
if (useRunpodBg) {
try {
const { data, mimeType } = imageToBase64(image, 1600);
const res = await fetch(`${API}/edit`, {
method: "POST",
headers: { "content-type": "application/json" },
body: JSON.stringify({
imageBase64: data,
mimeType,
op: { type: "remove-background" },
params: {},
}),
});
if (res.ok) {
const { imageBase64: out, mimeType: outMime } = (await res.json()) as {
imageBase64: string;
mimeType?: string;
};
return base64ToBlob(out, outMime || "image/png");
}
} catch {
/* fall through to the in-browser remover */
}
}
const inputBlob = await canvasBlob(image, 1600);
const { removeBackground } = await import("@imgly/background-removal");
return removeBackground(inputBlob, { output: { format: "image/png" } });
}
// Outpaint / expand-canvas: pad the image with a neutral border, mark that
// border WHITE in the mask, and run it through the same inpaint path as
// generative-fill — no extra backend route needed.
if (op.type === "outpaint") {
const { imageBase64: padded, maskBase64: border } = padForOutpaint(
image,
typeof op.factor === "number" ? op.factor : 1.5,
);
const outParams: Record<string, unknown> = {
strength: typeof op.strength === "number" ? op.strength : 0.85,
};
const outRes = await fetch(`${API}/edit`, {
method: "POST",
headers: { "content-type": "application/json" },
body: JSON.stringify({
imageBase64: padded,
mimeType: "image/png",
op: {
type: "generative-fill",
prompt:
op.prompt ||
"Extend and continue the scene naturally, matching lighting, colors and perspective.",
},
maskBase64: border,
params: outParams,
}),
});
if (!outRes.ok) {
const err = (await outRes.json().catch(() => ({}))) as { error?: string };
throw new Error(err.error || `AI request failed (${outRes.status}).`);
}
const outJson = (await outRes.json()) as { imageBase64: string; mimeType?: string };
return base64ToBlob(outJson.imageBase64, outJson.mimeType || "image/png");
}
const { data, mimeType, width, height } = imageToBase64(image, 1280);
// Masked ops carry an ImageData mask — rasterize it to a PNG matched to the
// (downscaled) image dims, and strip it from the op since ImageData is not
// JSON-serializable.
const maskBase64 = "mask" in op ? maskToBase64(op.mask, width, height) : undefined;
const wireOp: Record<string, unknown> = { type: op.type };
if ("prompt" in op && typeof op.prompt === "string") wireOp.prompt = op.prompt;
if ("factor" in op && typeof op.factor === "number") wireOp.factor = op.factor;
// Forward the "edit strength" slider (0..1) so the backend can scale the edit.
const params: Record<string, unknown> = {};
if ("strength" in op && typeof op.strength === "number") params.strength = op.strength;
const res = await fetch(`${API}/edit`, {
method: "POST",
headers: { "content-type": "application/json" },
body: JSON.stringify({ imageBase64: data, mimeType, op: wireOp, maskBase64, params }),
});
if (!res.ok) {
const err = (await res.json().catch(() => ({}))) as { error?: string };
throw new Error(err.error || `AI request failed (${res.status}).`);
}
const { imageBase64, mimeType: outMime } = (await res.json()) as {
imageBase64: string;
mimeType?: string;
};
return base64ToBlob(imageBase64, outMime || "image/png");
},
// Voice annotation: record → (optional denoise) → transcribe. Both proxy
// through server routes so the RunPod key stays server-side.
async transcribeAudio(audioBase64: string) {
const res = await fetch(`${API}/transcribe`, {
method: "POST",
headers: { "content-type": "application/json" },
body: JSON.stringify({ audio: audioBase64 }),
});
if (!res.ok) {
const err = (await res.json().catch(() => ({}))) as { error?: string };
throw new Error(err.error || `Transcription failed (${res.status}).`);
}
const { transcript } = (await res.json()) as { transcript?: string };
return (transcript ?? "").trim();
},
async denoiseAudio(audioBase64: string) {
const res = await fetch(`${API}/denoise`, {
method: "POST",
headers: { "content-type": "application/json" },
body: JSON.stringify({ audio: audioBase64 }),
});
if (!res.ok) {
const err = (await res.json().catch(() => ({}))) as { error?: string };
throw new Error(err.error || `Denoise failed (${res.status}).`);
}
const { audio } = (await res.json()) as { audio?: string };
return audio ?? audioBase64;
},
// Camera-tilt estimation is opt-in (needs the RunPod tilt endpoint deployed);
// gate it so the editor's Auto-straighten button only appears when configured.
estimateTilt: useRunpodTilt
? async (_item: MediaItem, image: ImageSource) => {
const { data } = imageToBase64(image, 1024);
const res = await fetch(`${API}/tilt`, {
method: "POST",
headers: { "content-type": "application/json" },
body: JSON.stringify({ image: data }),
});
if (!res.ok) {
const err = (await res.json().catch(() => ({}))) as { error?: string };
throw new Error(err.error || `Tilt estimate failed (${res.status}).`);
}
return (await res.json()) as {
rollDegrees: number;
pitchDegrees: number;
fovDegrees: number;
};
}
: undefined,
};
}
+418
View File
@@ -0,0 +1,418 @@
"use client";
// Smart Gallery data layer. Serves EITHER a device-local mock (when the Shell isn't configured — the
// demo keeps working offline) OR the live be-crm data door (crm.gallery.*), behind one StorageAdapter
// so the embedded @photo-gallery/sdk is mode-agnostic. Tenant scoping, comment authorship and byte
// authorization are all enforced server-side; this is just glue.
//
// Live contract (be-crm):
// query crm.gallery.state.load {} -> PersistedState
// cmd crm.gallery.state.apply StateChanges -> { upserted, removed }
// cmd crm.gallery.media.presignUpload { mediaId, mime, sizeBytes, filename? } -> { ref, uploadUrl, method, headers? }
// cmd crm.gallery.media.presignDownload { refs: string[] } -> { urls, expiresInSeconds }
// query crm.gallery.stats {} -> { items, albums, people, bytes }
// query crm.gallery.lock.status {} -> { hasPassword }
// cmd crm.gallery.lock.set { password: string | null } -> { hasPassword }
// cmd crm.gallery.lock.verify { password: string } -> { ok }
//
// BYTES NEVER PASS THROUGH be-crm OR THE BFF. `putMedia` mints a short-lived signed PUT URL and the
// browser transfers straight to object storage — the same rule media-api.ts follows for attachments.
import { useMemo } from "react";
import { useAppShell, useAuth } from "@abe-kap/appshell-sdk/react";
import {
createLocalStorageAdapter,
type GalleryFeatures,
type GalleryUser,
type MediaItem,
type PersistedState,
type StateChanges,
type StorageAdapter,
type StoredBlob,
type ThemeTokens,
} from "@photo-gallery/sdk";
import { user as demoUser } from "@/components/dashboard/account-data";
import { useMyAccess } from "@/lib/access";
import { isShellConfigured } from "./appshell";
const SHELL = isShellConfigured();
/** Device-local store key for demo mode. Namespaced so it can't collide with the SDK's own demo. */
const LOCAL_STORE_KEY = "lup:smart-gallery:v1";
/** be-crm caps a single presignDownload at 500 refs — chunk anything larger. */
const PRESIGN_CHUNK = 500;
/** Largest single upload the gallery will attempt (matches be-crm's gallery cap). */
export const MAX_GALLERY_BYTES = 200 * 1024 * 1024;
/* ======================================================================== */
/* Wire types (be-crm shapes) */
/* ======================================================================== */
interface StateLoadDTO {
media: MediaItem[];
albums: PersistedState["albums"];
people: PersistedState["people"];
labelAliases?: Record<string, string>;
deletedLabels?: string[];
}
interface PresignUploadDTO {
ref: string;
uploadUrl: string;
method: "PUT";
headers?: Record<string, string>;
}
interface PresignDownloadDTO {
urls: Record<string, string>;
expiresInSeconds: number;
}
/** The subset of the AppShell SDK this module needs — keeps the adapter unit-testable. */
interface DataDoor {
query<T>(action: string, variables?: Record<string, unknown>): Promise<T>;
command<T>(action: string, variables?: Record<string, unknown>): Promise<T>;
}
/* ======================================================================== */
/* The live adapter — be-crm data door */
/* ======================================================================== */
function chunk<T>(items: T[], size: number): T[][] {
const out: T[][] = [];
for (let i = 0; i < items.length; i += size) out.push(items.slice(i, i + size));
return out;
}
/**
* A StorageAdapter backed by the be-crm data door.
*
* Metadata (media/albums/people) rides the door as JSON; bytes go direct to object storage via
* short-lived signed URLs. `storageRef` is the durable handle we persist `src` is only ever a
* signed URL with a TTL, so it is re-resolved from the refs on every `load()`.
*/
export function createDataDoorAdapter(sdk: DataDoor): StorageAdapter {
/** Resolve durable refs → fresh signed GET URLs, in chunks the door will accept. */
async function resolveRefs(refs: string[]): Promise<Record<string, string>> {
const unique = [...new Set(refs.filter(Boolean))];
if (!unique.length) return {};
const urls: Record<string, string> = {};
for (const group of chunk(unique, PRESIGN_CHUNK)) {
const res = await sdk.command<PresignDownloadDTO>("crm.gallery.media.presignDownload", { refs: group });
Object.assign(urls, res.urls ?? {});
}
return urls;
}
return {
name: "crm-data-door",
async load(): Promise<PersistedState | null> {
const state = await sdk.query<StateLoadDTO>("crm.gallery.state.load", {});
const media = state.media ?? [];
// Signed URLs expire, so `src` is always rebuilt from `storageRef` at load time. Items with no
// ref (e.g. a seeded remote URL) keep whatever `src` they were stored with.
const urls = await resolveRefs(media.map((m) => m.storageRef ?? "").filter(Boolean));
return {
media: media.map((m) => (m.storageRef && urls[m.storageRef] ? { ...m, src: urls[m.storageRef]! } : m)),
albums: state.albums ?? [],
people: state.people ?? [],
labelAliases: state.labelAliases ?? {},
deletedLabels: state.deletedLabels ?? [],
version: 1,
};
},
// Incremental sync is the real path (see applyChanges). `save` only runs if the store ever falls
// back to whole-state persistence; express it as one big change set so behaviour is identical.
async save(state: PersistedState): Promise<void> {
await sdk.command("crm.gallery.state.apply", {
upsertMedia: state.media,
upsertAlbums: state.albums,
upsertPeople: state.people,
labelAliases: state.labelAliases ?? {},
deletedLabels: state.deletedLabels ?? [],
});
},
/**
* Incremental persistence only the entities that actually changed. This is what makes a shared
* tenant library safe for concurrent editors: two people touching different photos never clobber
* each other, because neither sends the other's rows.
*/
async applyChanges(changes: StateChanges): Promise<void> {
// StateChanges is a closed interface; the door takes an open variables bag.
await sdk.command("crm.gallery.state.apply", { ...changes });
},
/** Presign → direct PUT → resolve a display URL. Bytes never touch be-crm or the BFF. */
async putMedia(id: string, blob: Blob, meta: { name: string; mime: string }): Promise<StoredBlob> {
if (blob.size > MAX_GALLERY_BYTES) {
throw new Error(`File is too large (max ${Math.floor(MAX_GALLERY_BYTES / (1024 * 1024))} MB).`);
}
const mime = meta.mime || blob.type || "application/octet-stream";
const presigned = await sdk.command<PresignUploadDTO>("crm.gallery.media.presignUpload", {
mediaId: id,
mime,
sizeBytes: blob.size,
filename: meta.name,
});
const res = await fetch(presigned.uploadUrl, {
method: presigned.method ?? "PUT",
headers: { "content-type": mime, ...(presigned.headers ?? {}) },
body: blob,
});
if (!res.ok) throw new Error(`Upload failed (${res.status}).`);
const urls = await resolveRefs([presigned.ref]);
return { ref: presigned.ref, url: urls[presigned.ref] ?? presigned.ref };
},
};
}
/* ======================================================================== */
/* Public hooks */
/* ======================================================================== */
export interface GalleryStorage {
/** True when persisting to be-crm; false when running on the device-local demo store. */
live: boolean;
adapter: StorageAdapter;
}
/**
* The storage backend for the embedded gallery. Live against the be-crm data door once the Shell is
* configured, otherwise a device-local store so the demo works with no backend at all.
*
* SHELL is a build-time constant, so this branch is stable across renders (Rules-of-Hooks safe).
*/
export function useGalleryStorage(): GalleryStorage {
const { sdk } = useAppShell();
// The adapter identity must be stable — PhotoGallery captures it in a ref on first render.
return useMemo<GalleryStorage>(
() =>
SHELL
? { live: true, adapter: createDataDoorAdapter(sdk as unknown as DataDoor) }
: { live: false, adapter: createLocalStorageAdapter(LOCAL_STORE_KEY) },
[sdk],
);
}
/**
* The signed-in identity handed to the gallery so comments are attributed to a real person rather
* than a free-text name the author can type themselves.
*
* Falls back to the same static demo user the sidebar and topbar use when the Shell isn't wired, so
* the comment module behaves identically in the demo the CRM never shows an anonymous author.
* be-crm re-stamps `authorId` from the PAT on every write regardless, so this value is a display
* convenience, never the source of authority.
*/
export function useGalleryUser(): GalleryUser {
const { user, context } = useAuth();
return useMemo<GalleryUser>(() => {
if (!user) return { id: demoUser.id, name: demoUser.name };
const avatarUrl = user.avatarUrl ?? context?.principal?.avatarUrl;
return {
id: user.id,
name: user.displayName || user.email || demoUser.name,
...(user.email ? { email: user.email } : {}),
...(avatarUrl ? { avatarUrl } : {}),
};
}, [user, context]);
}
/* ======================================================================== */
/* Recently Deleted lock */
/* ======================================================================== */
/**
* The SDK's `lockProvider` contract, declared here rather than imported so this module keeps
* compiling against an SDK build that predates the prop. It is structurally identical to
* `PhotoGalleryProps['lockProvider']`.
*/
export interface GalleryLockProvider {
status(): Promise<{ hasPassword: boolean }>;
/** `null` clears the password. */
set(password: string | null): Promise<void>;
verify(password: string): Promise<boolean>;
}
/**
* A server-backed, per-user lock for the Recently Deleted view.
*
* Without this the SDK falls back to a device-local `localStorage` hash: the lock exists only on
* the browser that set it, so the same account on a second device sees an unlocked trash. Backed
* by `crm.gallery.lock.*`, the password becomes an account property be-crm stores only a
* scrypt hash with a per-record random salt, keyed by (tenant, principal), and rate-limits verify.
*
* Returns `undefined` in demo mode ON PURPOSE: with no backend there is nowhere to put the hash,
* and the SDK's own device-local behaviour is the right fallback for a demo.
*
* SHELL is a build-time constant, so this branch is stable across renders (Rules-of-Hooks safe).
*/
export function useGalleryLockProvider(): GalleryLockProvider | undefined {
const { sdk } = useAppShell();
return useMemo<GalleryLockProvider | undefined>(() => {
if (!SHELL) return undefined;
const door = sdk as unknown as DataDoor;
return {
status: () => door.query<{ hasPassword: boolean }>("crm.gallery.lock.status", {}),
// The SDK's contract returns void; the door's `{ hasPassword }` is redundant after a set.
set: async (password) => {
await door.command("crm.gallery.lock.set", { password });
},
// A wrong password is a normal `{ ok: false }`, not an error. A 403 (the verify lockout)
// still throws, which is what the SDK's prompt should surface.
verify: async (password) => (await door.command<{ ok: boolean }>("crm.gallery.lock.verify", { password })).ok,
};
}, [sdk]);
}
/* ======================================================================== */
/* Permission-gated features */
/* ======================================================================== */
/**
* The gallery capabilities the CRM gates behind team permissions. Each SDK feature toggle
* (and the whole-view `media.view` gate) maps to one permission id from be-crm's Media group.
* The SDK still enforces nothing here hiding a control is UX; be-crm enforces every write via
* GALLERY_VISIBILITY + tenant scoping regardless of what the UI shows.
*/
const MEDIA_PERMISSIONS = [
"media.view", "media.upload", "media.capture", "media.edit",
"media.delete", "media.export", "media.share", "media.map", "media.ai",
] as const;
/** SDK feature toggle → the permission id that unlocks it. */
const FEATURE_PERMISSION: Record<keyof GalleryFeatures, string> = {
editor: "media.edit",
camera: "media.capture",
import: "media.upload",
export: "media.export",
sharing: "media.share",
map: "media.map",
ai: "media.ai",
};
export interface ResolvedGalleryFeatures {
/** Feature toggles to hand the SDK's `features` prop, resolved from the caller's permissions. */
features: GalleryFeatures;
/** Whether the whole Smart Gallery view should render at all (`media.view`). */
canView: boolean;
}
/**
* Resolve the gallery's feature flags + view access from the signed-in user's CRM permissions.
*
* Fallback rule (SAFE default a control is enabled unless the user is positively known to lack it):
* 1. While access is still loading, stay permissive so features don't flash on then vanish.
* 2. Superadmins and owners get everything (they hold every permission anyway; this is explicit).
* 3. A member who has been assigned AT LEAST ONE `media.*` permission is gated precisely: each
* feature is enabled only if its mapped permission is in their list.
* 4. A member with ZERO `media.*` permissions assigned is treated as fully enabled. Roles are not
* configured with Media perms until an admin opts in (§5 just made them assignable), so gating a
* freshly-seeded member down to nothing would cripple the gallery before anyone could grant them
* anything. The "has at least one media.* perm" signal is what flips a role from this permissive
* default into precise per-feature gating.
*
* Because be-crm's mock access (demo, no Shell) returns superadmin + all perms, the demo shows
* everything via rule 2.
*/
export function useGalleryFeatures(): ResolvedGalleryFeatures {
const access = useMyAccess();
return useMemo<ResolvedGalleryFeatures>(() => {
const has = (p: string) => access.permissions.includes(p);
const privileged = access.isSuperadmin || access.roleSlugs.includes("owner");
const hasAnyMedia = MEDIA_PERMISSIONS.some(has);
// Permissive whenever we can't (yet) prove the user lacks a permission: loading, privileged, or a
// member who has no Media perms assigned at all. Otherwise gate precisely on the mapped id.
const allow = (perm: string) => access.loading || privileged || !hasAnyMedia || has(perm);
const features: GalleryFeatures = {
editor: allow(FEATURE_PERMISSION.editor),
camera: allow(FEATURE_PERMISSION.camera),
ai: allow(FEATURE_PERMISSION.ai),
map: allow(FEATURE_PERMISSION.map),
import: allow(FEATURE_PERMISSION.import),
export: allow(FEATURE_PERMISSION.export),
sharing: allow(FEATURE_PERMISSION.sharing),
};
return { features, canView: allow("media.view") };
}, [access]);
}
/* ======================================================================== */
/* Theme bridge */
/* ======================================================================== */
/**
* Maps the gallery's design tokens onto the CRM's own CSS variables.
*
* Every value is a `var(--crm-token)` reference rather than a literal hex, so the gallery inherits the
* dashboard's palette through the SAME `[data-theme]` cascade the rest of the app uses light/dark
* switch together, and a future palette change reaches the gallery with no code edit here.
*/
export const GALLERY_THEME_TOKENS: ThemeTokens = {
// Surfaces
bgLight: "var(--bg)",
bgDark: "var(--bg)",
elevatedLight: "var(--panel-3)",
elevatedDark: "var(--panel-3)",
sidebarBgLight: "var(--sidebar)",
sidebarBgDark: "var(--sidebar)",
toolbarBgLight: "var(--panel)",
toolbarBgDark: "var(--panel)",
cardLight: "var(--panel)",
cardDark: "var(--panel)",
cardHoverLight: "var(--panel-3)",
cardHoverDark: "var(--panel-3)",
menuBgLight: "var(--panel-3)",
menuBgDark: "var(--panel-3)",
// Text
textLight: "var(--text)",
textDark: "var(--text)",
textSecondaryLight: "var(--muted)",
textSecondaryDark: "var(--muted)",
textTertiaryLight: "var(--faint)",
textTertiaryDark: "var(--faint)",
// Lines + washes
separatorLight: "var(--border)",
separatorDark: "var(--border)",
separatorStrongLight: "var(--border-2)",
separatorStrongDark: "var(--border-2)",
hoverLight: "var(--track)",
hoverDark: "var(--track)",
activeLight: "var(--border-2)",
activeDark: "var(--border-2)",
sidebarSelectedLight: "var(--track)",
sidebarSelectedDark: "var(--track)",
glassBorderLight: "var(--border-2)",
glassBorderDark: "var(--border-2)",
// Brand
accent: "var(--orange)",
accentStrongLight: "var(--orange-2)",
accentStrongDark: "var(--orange-2)",
accentContrast: "#ffffff",
dangerLight: "var(--red)",
dangerDark: "var(--red)",
tileFav: "var(--red)",
segmentedActive: "var(--panel-3)",
// Overlays + chrome
overlayBg: "rgba(2,2,6,0.94)",
editorBg: "var(--panel-2)",
shadowSm: "0 1px 2px rgba(0,0,0,0.18)",
shadowMdLight: "var(--shadow)",
shadowMdDark: "var(--shadow)",
shadowLgLight: "0 30px 70px -20px rgba(15,23,42,0.25)",
shadowLgDark: "0 30px 70px -20px rgba(0,0,0,0.7)",
fontFamily: "var(--font)",
radiusMenu: 14,
sidebarRadius: 14,
};
+86
View File
@@ -0,0 +1,86 @@
/**
* Tiny in-process fixed-window rate limiter for the Smart Gallery AI routes.
*
* ---------------------------------------------------------------------------
* SCOPE AND LIMITATIONS READ BEFORE RELYING ON THIS
* ---------------------------------------------------------------------------
* State lives in a plain `Map` in THIS process's memory. That means:
*
* - PER-INSTANCE, NOT GLOBAL. With N app instances behind a load balancer a
* caller gets up to N x the configured budget. On serverless platforms each
* cold start begins with an empty map, so the effective limit is weaker
* still.
* - NOT DURABLE. A restart or redeploy clears every counter.
* - FIXED WINDOW, NOT SLIDING. A caller can burst `max` at the very end of one
* window and `max` again at the start of the next up to 2x `max` across a
* window boundary. Acceptable here; the goal is to bound runaway cost, not
* to meter precisely.
*
* It exists because the routes it guards spend real money on GPU inference and
* shipping them with NO limit at all is worse than shipping an imperfect one.
*
* REPLACE WITH REDIS (or the platform's rate limiter) BEFORE RUNNING MORE THAN
* ONE INSTANCE. The `limit()` signature is deliberately narrow so a Redis-backed
* implementation can drop straight in the only change needed is making it
* async at the call sites.
*
* This is a throttle, not an authorization check. See src/lib/server/session.ts.
*/
interface Window {
/** Requests counted so far in the current window. */
count: number;
/** Epoch ms at which the current window ends. */
resetAt: number;
}
const windows = new Map<string, Window>();
/** Drop expired entries so the map cannot grow without bound. */
const SWEEP_INTERVAL_MS = 60_000;
let lastSweep = 0;
function sweep(now: number): void {
if (now - lastSweep < SWEEP_INTERVAL_MS) return;
lastSweep = now;
for (const [key, w] of windows) {
if (w.resetAt <= now) windows.delete(key);
}
}
export interface LimitResult {
ok: boolean;
/** Seconds until the window resets. Send as `Retry-After` when `ok` is false. */
retryAfter: number;
}
/**
* Count one request against `key` and report whether it is allowed.
*
* @param key Caller identity use `rateLimitKey()` from ./session.
* @param max Requests allowed per window.
* @param windowMs Window length in ms.
*/
export function limit(key: string, max: number, windowMs: number): LimitResult {
const now = Date.now();
sweep(now);
const existing = windows.get(key);
if (!existing || existing.resetAt <= now) {
windows.set(key, { count: 1, resetAt: now + windowMs });
return { ok: true, retryAfter: 0 };
}
if (existing.count >= max) {
return { ok: false, retryAfter: Math.max(1, Math.ceil((existing.resetAt - now) / 1000)) };
}
existing.count += 1;
return { ok: true, retryAfter: 0 };
}
/** Test/maintenance helper — clears all counters. */
export function resetAllLimits(): void {
windows.clear();
lastSweep = 0;
}
+86
View File
@@ -0,0 +1,86 @@
/**
* base64 / data-URI helpers shared by the RunPod endpoint wrappers.
* Different endpoints name their image field differently and some prefix a
* `data:` URI these helpers normalize both.
*
* PROVENANCE: a faithful port of
* `advance-photo-gallery-web-sdk/apps/web/src/lib/runpod/base64.ts`.
* The `pickOutputImage` normalization is intentionally identical so both apps
* tolerate the same set of endpoint response shapes.
*
* SERVER-ONLY.
*/
/** "data:image/png;base64,XXXX" -> "XXXX" (leaves a bare base64 string untouched). */
export function stripDataUri(s: string): string {
if (!s.startsWith("data:")) return s;
const i = s.indexOf(",");
return i === -1 ? s : s.slice(i + 1);
}
/** Wrap a bare base64 string in a data: URI. */
export function toDataUri(b64: string, mime: string): string {
return `data:${mime};base64,${stripDataUri(b64)}`;
}
/**
* Pull the first image-like base64 string out of a RunPod endpoint's `output`,
* regardless of which field name it used. Handles the common shapes:
* "iVBOR..." (raw string)
* { image: "..." } / { image_png } / { image_base64 }
* { images: ["..."] } (array)
* { output: { image: "..." } } (nested)
*/
export function pickOutputImage(output: unknown): string {
const s = findImageString(output, false);
if (!s) throw new Error("RunPod output did not contain an image.");
return stripDataUri(s);
}
/** Keys whose name implies the value IS the image — any non-empty string is accepted. */
const IMAGE_KEYS = ["image_png", "image", "image_base64", "images"] as const;
/** Generic wrapper keys — a string here must actually look like image data. */
const CONTAINER_KEYS = ["output", "result", "data"] as const;
/** A base64 image payload is long; a status/id string ("success", "job-abc") is short. */
function looksLikeImageData(s: string): boolean {
if (s.startsWith("data:image/")) return true;
return s.length >= 256 && /^[A-Za-z0-9+/=\s]+$/.test(s.slice(0, 256));
}
/**
* `strict` is true when we descended through a generic wrapper key (output/result/
* data), where a bare string could be a status/id rather than an image so it must
* pass `looksLikeImageData`. Under an explicit image key (or at top level) any
* non-empty string is taken as the image.
*/
function findImageString(value: unknown, strict: boolean, depth = 0): string | undefined {
if (depth > 5) return undefined;
if (typeof value === "string") {
if (!value) return undefined;
return !strict || looksLikeImageData(value) ? value : undefined;
}
if (Array.isArray(value)) {
for (const el of value) {
const s = findImageString(el, strict, depth + 1);
if (s) return s;
}
return undefined;
}
if (value && typeof value === "object") {
const o = value as Record<string, unknown>;
for (const key of IMAGE_KEYS) {
if (key in o) {
const s = findImageString(o[key], false, depth + 1);
if (s) return s;
}
}
for (const key of CONTAINER_KEYS) {
if (key in o) {
const s = findImageString(o[key], true, depth + 1);
if (s) return s;
}
}
}
return undefined;
}
+134
View File
@@ -0,0 +1,134 @@
/**
* Low-level RunPod transport. Every model runs as its own RunPod endpoint; this
* handles the common request envelope, bearer auth, and both response modes:
*
* - `/runsync` (preferred): the job runs synchronously and the body already
* contains `output` (or IS the output for a custom handler).
* - `/run`: returns `{ id, status }`; we poll `/status/{id}` until the job
* reaches a terminal state or the time budget (kept under the 60s serverless
* function cap) is exhausted.
*
* RUNPOD_API_KEY is read here so callers never handle the secret directly, and
* it is NEVER echoed into an error message. Upstream error bodies are truncated
* to 160 chars before being surfaced.
*
* PROVENANCE: a faithful port of
* `advance-photo-gallery-web-sdk/apps/web/src/lib/runpod/client.ts` (same 55s
* budget, same polling, same RunpodError.status mapping).
*
* SERVER-ONLY.
*/
export class RunpodError extends Error {
status: number;
constructor(message: string, status = 502) {
super(message);
this.name = "RunpodError";
this.status = status;
}
}
export interface RunpodCallOpts {
/** Human label used in error messages, e.g. "sd-inpaint". */
name: string;
/** Full endpoint URL from the per-model env var (…/runsync or …/run). */
url: string;
input: Record<string, unknown>;
/** Total budget for the whole call incl. polling. Default 55s (< the 60s cap). */
timeoutMs?: number;
pollIntervalMs?: number;
}
interface RunpodEnvelope {
id?: string;
status?: string;
output?: unknown;
error?: unknown;
}
export async function runpodCall<TOut = unknown>(opts: RunpodCallOpts): Promise<TOut> {
const { name, url, input } = opts;
const timeoutMs = opts.timeoutMs ?? 55_000;
const pollIntervalMs = opts.pollIntervalMs ?? 1500;
const key = process.env.RUNPOD_API_KEY;
if (!key) throw new RunpodError("RUNPOD_API_KEY is not set.", 500);
const deadline = Date.now() + timeoutMs;
const authHeaders = { authorization: `Bearer ${key}` };
let res: Response;
try {
res = await fetch(url, {
method: "POST",
headers: { "content-type": "application/json", ...authHeaders },
body: JSON.stringify({ input }),
signal: AbortSignal.timeout(timeoutMs),
cache: "no-store",
});
} catch {
throw new RunpodError(`Could not reach RunPod (${name}).`, 502);
}
if (!res.ok) {
// Truncated on purpose — never surface a full upstream body to the client.
const detail = (await res.text().catch(() => "")).slice(0, 160);
throw new RunpodError(`RunPod ${name} error (${res.status}). ${detail}`.trim(), 502);
}
const data = (await res.json().catch(() => null)) as RunpodEnvelope | null;
if (!data || typeof data !== "object") {
throw new RunpodError(`RunPod ${name} returned an invalid response.`, 502);
}
// Terminal failure reported in a job envelope.
if (data.status === "FAILED" || data.status === "CANCELLED") {
throw new RunpodError(`RunPod ${name} job ${data.status.toLowerCase()}.`, 502);
}
// No job id → this is not an async envelope; the body itself is the output.
// Covers custom /runsync handlers that return their result directly, even when
// it carries a `status` field (e.g. "success").
if (data.id === undefined) {
return (data.output !== undefined ? data.output : data) as TOut;
}
// Job envelope that already carries a completed/inline output.
if (data.output !== undefined && (data.status === undefined || data.status === "COMPLETED")) {
return data.output as TOut;
}
// Async: poll /status/{id} until COMPLETED / FAILED / the time budget runs out.
// Each wait + fetch is clamped to the remaining budget so the whole call stays
// under `timeoutMs` (kept below the 60s function cap).
const statusUrl = url.replace(/\/run(sync)?(\/?)$/, `/status/${data.id}`);
for (;;) {
const remaining = deadline - Date.now();
if (remaining < 500) break; // not enough budget for another round
await sleep(Math.min(pollIntervalMs, remaining));
const left = deadline - Date.now();
if (left <= 0) break;
let sres: Response;
try {
sres = await fetch(statusUrl, {
headers: authHeaders,
signal: AbortSignal.timeout(Math.min(10_000, Math.max(1000, left))),
cache: "no-store",
});
} catch {
continue; // transient — keep polling until the deadline
}
if (!sres.ok) continue;
const sdata = (await sres.json().catch(() => null)) as RunpodEnvelope | null;
if (!sdata) continue;
if (sdata.status === "COMPLETED") return sdata.output as TOut;
if (sdata.status === "FAILED" || sdata.status === "CANCELLED") {
throw new RunpodError(`RunPod ${name} job ${sdata.status.toLowerCase()}.`, 502);
}
}
throw new RunpodError(
`RunPod ${name} timed out (raise the endpoint's speed or use /runsync).`,
504,
);
}
function sleep(ms: number): Promise<void> {
return new Promise((resolve) => setTimeout(resolve, ms));
}
+301
View File
@@ -0,0 +1,301 @@
/**
* Typed wrappers for each RunPod endpoint behind the Smart Gallery.
* Every function reads its own `RUNPOD_*_URL` env var, sends the exact `input`
* contract the model spec documents, and normalizes the response.
*
* PROVENANCE: a faithful port of
* `advance-photo-gallery-web-sdk/apps/web/src/lib/runpod/endpoints.ts`
* identical `normalizeDetections` / `normalizeBox` logic and identical env var
* names, so an endpoint deployed for the SDK demo works here unchanged.
*
* SERVER-ONLY. Missing/invalid URLs throw a RunpodError(500) so an unconfigured
* op surfaces as a clear message rather than a crash.
*/
import { stripDataUri, pickOutputImage } from "./base64";
import { RunpodError, runpodCall } from "./client";
import type {
Img2ImgReq,
InpaintReq,
RunpodDetection,
RunpodImageResult,
TiltResult,
TranscriptResult,
} from "./types";
function envNum(v: string | undefined, fallback: number): number {
// Treat a blank/whitespace env var as unset — Number('') is 0 (finite), which
// would otherwise send e.g. strength:0 for `RUNPOD_SD_STRENGTH=`.
if (v == null || v.trim() === "") return fallback;
const n = Number(v);
return Number.isFinite(n) ? n : fallback;
}
function endpointUrl(envVar: string): string {
const url = process.env[envVar];
if (!url || !/^https?:\/\//i.test(url)) {
throw new RunpodError(`${envVar} is not set (or is not an http(s) URL).`, 500);
}
return url;
}
/** Run an image-in/image-out endpoint and normalize the result to base64 PNG. */
async function imageOp(
name: string,
envVar: string,
input: Record<string, unknown>,
): Promise<RunpodImageResult> {
const output = await runpodCall({ name, url: endpointUrl(envVar), input });
return { imageBase64: pickOutputImage(output), mimeType: "image/png" };
}
// ---------------------------------------------------------------------------
// Image endpoints
// ---------------------------------------------------------------------------
/** #6 Background removal (U²-Net via rembg). model: u2net | u2netp | u2net_human_seg */
export function rpRemoveBackground(imageB64: string, model?: string): Promise<RunpodImageResult> {
return imageOp("background-removal", "RUNPOD_BG_REMOVE_URL", {
task: "remove-bg",
image: imageB64,
...(model ? { model_name: model } : {}),
});
}
/** #7 Real-ESRGAN enhance/upscale (RealESRGAN_x4plus). The caller's scale (from the
* op / restore pass) is authoritative it is not overridden by any env default. */
export function rpUpscale(
imageB64: string,
scale: 2 | 4,
faceEnhance = false,
): Promise<RunpodImageResult> {
return imageOp("upscale", "RUNPOD_UPSCALE_URL", {
task: "upscale",
image: imageB64,
scale,
face_enhance: faceEnhance,
});
}
/** #8 DDColor B&W → colorize. */
export function rpColorize(imageB64: string, inputSize?: number): Promise<RunpodImageResult> {
return imageOp("colorize", "RUNPOD_COLORIZE_URL", {
image: imageB64,
...(inputSize ? { input_size: inputSize } : {}),
});
}
/** #9 SD 3.5 masked inpainting (sky fix / eraser / fill). Also #11 outpaint (pre-padded). */
export function rpInpaint(p: InpaintReq): Promise<RunpodImageResult> {
const input: Record<string, unknown> = {
task: "inpaint",
image: p.imageB64,
mask: p.maskB64,
prompt: p.prompt,
strength: p.strength ?? envNum(process.env.RUNPOD_SD_STRENGTH, 0.8),
guidance_scale: p.guidanceScale ?? envNum(process.env.RUNPOD_SD_GUIDANCE, 7),
num_inference_steps: p.steps ?? envNum(process.env.RUNPOD_SD_STEPS, 35),
};
const negative = p.negativePrompt ?? process.env.RUNPOD_SD_NEGATIVE_PROMPT;
if (negative) input.negative_prompt = negative;
if (p.seed != null) input.seed = p.seed;
return imageOp("sd-inpaint", "RUNPOD_SD_INPAINT_URL", input);
}
/** #10 SD 3.5 general prompt edit (img2img, no mask). */
export function rpImg2Img(p: Img2ImgReq): Promise<RunpodImageResult> {
const input: Record<string, unknown> = {
task: "img2img",
image: p.imageB64,
prompt: p.prompt,
strength: p.strength ?? envNum(process.env.RUNPOD_SD_STRENGTH, 0.6),
guidance_scale: p.guidanceScale ?? envNum(process.env.RUNPOD_SD_GUIDANCE, 7),
num_inference_steps: p.steps ?? envNum(process.env.RUNPOD_SD_STEPS, 35),
};
const negative = p.negativePrompt ?? process.env.RUNPOD_SD_NEGATIVE_PROMPT;
if (negative) input.negative_prompt = negative;
if (p.seed != null) input.seed = p.seed;
return imageOp("sd-img2img", "RUNPOD_SD_IMG2IMG_URL", input);
}
// ---------------------------------------------------------------------------
// #1 YOLO detection → SDK DetectedObject shape (box as fractions 0..1)
// ---------------------------------------------------------------------------
export async function rpDetect(
imageB64: string,
width: number,
height: number,
): Promise<RunpodDetection[]> {
const output = await runpodCall<unknown>({
name: "yolo-detect",
url: endpointUrl("RUNPOD_YOLO_URL"),
input: { image: imageB64, task: "detect" },
});
return normalizeDetections(output, width, height);
}
function extractDetectionArray(output: unknown): unknown[] {
if (Array.isArray(output)) return output;
if (output && typeof output === "object") {
const o = output as Record<string, unknown>;
for (const key of ["detections", "predictions", "objects", "results", "boxes"]) {
if (Array.isArray(o[key])) return o[key] as unknown[];
}
}
return [];
}
/** First non-empty STRING among the args (numbers ignored — a numeric `class` is an index, not a name). */
function firstLabel(...vals: unknown[]): string | undefined {
for (const v of vals) {
if (typeof v === "string" && v.trim()) return v.trim();
}
return undefined;
}
function normalizeDetections(output: unknown, width: number, height: number): RunpodDetection[] {
const out: RunpodDetection[] = [];
for (const raw of extractDetectionArray(output)) {
if (!raw || typeof raw !== "object") continue;
const o = raw as Record<string, unknown>;
// Prefer a human-readable name (ultralytics tojson puts the string in `name`
// and a numeric index in `class`); fall back to class_<id>. Using firstLabel
// (not `??`) also means an explicit empty-string label doesn't get kept + dropped.
const classId = o.class_id ?? (typeof o.class === "number" ? o.class : undefined);
const label = (
firstLabel(o.label, o.name, o.class_name, typeof o.class === "string" ? o.class : undefined) ??
(classId != null ? `class_${classId}` : "object")
).toLowerCase();
const confidence = Number(o.confidence ?? o.score ?? o.conf ?? 0) || 0;
const box = normalizeBox(o, width, height);
if (box && label) out.push({ label, confidence, box });
}
return out;
}
function num(v: unknown): number | null {
const n = Number(v);
return Number.isFinite(n) ? n : null;
}
function asNum4(v: unknown): [number, number, number, number] | null {
if (!Array.isArray(v) || v.length < 4) return null;
const a = num(v[0]);
const b = num(v[1]);
const c = num(v[2]);
const d = num(v[3]);
return a === null || b === null || c === null || d === null ? null : [a, b, c, d];
}
/**
* Normalize a detection box to {x, y, width, height} as fractions 0..1 of the
* image, from whatever shape the endpoint emits:
* - `xyxy: [x1,y1,x2,y2]` (ultralytics) and generic `box: [...]` corner form
* - `xywh: [...]` and COCO `bbox: [x,y,w,h]` x/y/width/height form
* - object `{x1,y1,x2,y2}` / `{left,top,right,bottom}` (ultralytics tojson) corners
* - object `{x,y,width,height}` x/y/width/height
* Pixel values (any component > 1) are divided by the image dims; already-
* normalized fractions pass through.
*/
function normalizeBox(
o: Record<string, unknown>,
width: number,
height: number,
): RunpodDetection["box"] | null {
const W = width || 1;
const H = height || 1;
const clamp01 = (n: number) => Math.max(0, Math.min(1, n));
const frac = (x: number, y: number, w: number, h: number): RunpodDetection["box"] => {
if (Math.max(Math.abs(x), Math.abs(y), Math.abs(w), Math.abs(h)) > 1) {
x /= W;
y /= H;
w /= W;
h /= H;
}
return { x: clamp01(x), y: clamp01(y), width: clamp01(w), height: clamp01(h) };
};
const fromXyxy = (x1: number, y1: number, x2: number, y2: number) =>
frac(x1, y1, x2 - x1, y2 - y1);
// 1. Array boxes, interpreted by which key holds them.
const xyxyArr = asNum4(o.xyxy);
if (xyxyArr) return fromXyxy(xyxyArr[0], xyxyArr[1], xyxyArr[2], xyxyArr[3]);
const xywhArr = asNum4(o.xywh) ?? asNum4(o.bbox); // COCO `bbox` is [x,y,w,h]
if (xywhArr) return frac(xywhArr[0], xywhArr[1], xywhArr[2], xywhArr[3]);
const boxArr = asNum4(o.box); // generic array box → assume corner form
if (boxArr) return fromXyxy(boxArr[0], boxArr[1], boxArr[2], boxArr[3]);
// 2. Object boxes (either nested under `box` or directly on the detection).
const src =
o.box && typeof o.box === "object" && !Array.isArray(o.box)
? (o.box as Record<string, unknown>)
: o;
const x1 = num(src.x1 ?? src.left);
const y1 = num(src.y1 ?? src.top);
const x2 = num(src.x2 ?? src.right);
const y2 = num(src.y2 ?? src.bottom);
if (x1 !== null && y1 !== null && x2 !== null && y2 !== null) return fromXyxy(x1, y1, x2, y2);
const x = num(src.x);
const y = num(src.y);
const w = num(src.width);
const h = num(src.height);
if (x !== null && y !== null && w !== null && h !== null) return frac(x, y, w, h);
return null;
}
// ---------------------------------------------------------------------------
// Audio / calibration endpoints
// ---------------------------------------------------------------------------
/** #2 Camera tilt (DeepSingleImageCalibration). */
export async function rpTilt(imageB64: string): Promise<TiltResult> {
const o = await runpodCall<Record<string, unknown>>({
name: "tilt",
url: endpointUrl("RUNPOD_TILT_URL"),
input: { image: imageB64 },
});
return {
rollDegrees: Number(o.roll_degrees ?? 0) || 0,
pitchDegrees: Number(o.pitch_degrees ?? 0) || 0,
fovDegrees: Number(o.fov_degrees ?? 0) || 0,
};
}
/** #3 Voice-to-text (Parakeet). Audio must be WAV 16kHz mono PCM16. */
export async function rpTranscribe(
audioB64: string,
opts?: { language?: string; timestamps?: boolean; punctuation?: boolean },
): Promise<TranscriptResult> {
const o = await runpodCall<Record<string, unknown>>({
name: "transcribe",
url: endpointUrl("RUNPOD_STT_URL"),
input: { audio: audioB64, task: "transcribe", ...(opts ?? {}) },
});
const rawSegments = Array.isArray(o.segments) ? o.segments : [];
const segments = rawSegments.map((s) => {
const seg = (s ?? {}) as Record<string, unknown>;
return {
text: String(seg.text ?? ""),
startSec: Number(seg.start_sec ?? 0) || 0,
endSec: Number(seg.end_sec ?? 0) || 0,
};
});
return {
transcript: String(o.transcript ?? ""),
segments: segments.length ? segments : undefined,
};
}
/** #12 Audio noise removal (RNNoise). Audio must be WAV 48kHz mono 16-bit PCM. */
export async function rpDenoiseAudio(audioB64: string): Promise<{ audioB64: string }> {
const o = await runpodCall<Record<string, unknown>>({
name: "audio-denoise",
url: endpointUrl("RUNPOD_AUDIO_DENOISE_URL"),
input: { audio: audioB64, task: "denoise" },
});
const a = o.audio ?? o.output ?? "";
return { audioB64: stripDataUri(String(a)) };
}
+68
View File
@@ -0,0 +1,68 @@
/**
* Request/response types for the RunPod endpoints behind the Smart Gallery.
*
* PROVENANCE: a faithful port of
* `advance-photo-gallery-web-sdk/apps/web/src/lib/runpod/types.ts`.
* Keep the two in sync the RunPod endpoints are shared between the standalone
* SDK demo and this CRM.
*
* SERVER-ONLY. Imported by src/app/api/gallery/ai/* route handlers, never by a
* client component (which must not see a RunPod URL or key).
*/
/** Normalized image result returned by every image endpoint (raw base64, no data: prefix). */
export interface RunpodImageResult {
imageBase64: string;
mimeType: string;
}
/** #9 SD 3.5 masked inpainting (also #11 outpaint, pre-padded). white in mask = regenerate. */
export interface InpaintReq {
imageB64: string;
maskB64: string;
prompt: string;
negativePrompt?: string;
strength?: number;
guidanceScale?: number;
steps?: number;
seed?: number;
}
/** #10 SD 3.5 general prompt edit (img2img, no mask). */
export interface Img2ImgReq {
imageB64: string;
prompt: string;
negativePrompt?: string;
strength?: number;
guidanceScale?: number;
steps?: number;
seed?: number;
}
/**
* #1 YOLO construction-material classifier, normalized to the SDK's
* `DetectedObject` shape (box as fractions 0..1 of the image).
*/
export interface RunpodDetection {
label: string;
confidence: number;
box: { x: number; y: number; width: number; height: number };
}
/** #2 DeepSingleImageCalibration — camera tilt. */
export interface TiltResult {
rollDegrees: number;
pitchDegrees: number;
fovDegrees: number;
}
/** #3 Parakeet voice-to-text. */
export interface TranscriptSegment {
text: string;
startSec: number;
endSec: number;
}
export interface TranscriptResult {
transcript: string;
segments?: TranscriptSegment[];
}
+139
View File
@@ -0,0 +1,139 @@
/**
* Session gate for the Smart Gallery AI routes.
*
* ---------------------------------------------------------------------------
* TRUST MODEL
* ---------------------------------------------------------------------------
* These routes proxy a paid, rate-limited GPU backend (RunPod) using a secret
* held only on this server. An unauthenticated route is therefore not merely an
* information-disclosure problem it is a billable-resource problem: anyone who
* can reach the URL can spend the operator's GPU budget, and can use the CRM as
* an open relay for arbitrary image/audio processing.
*
* The upstream SDK demo's routes (apps/web/src/app/api/ai/*) are COMPLETELY
* unauthenticated. This module is the fix; every ported route must call it
* before doing any work.
*
* WHO IS TRUSTED
* We do not verify a JWT here and we do not hold any signing key. The single
* source of truth for "is this caller signed in" is the Shell BFF
* (`${BFF_ORIGIN}/api/session/context`), which owns the HttpOnly session cookie.
* We forward the caller's raw `cookie` header to it and treat a 200 as proof of
* a session. Consequences of that choice, stated explicitly:
*
* - The BFF is trusted absolutely. If it is compromised or misconfigured to
* answer 200 for anonymous callers, these routes are open. BFF_ORIGIN must
* therefore only ever point at an origin the operator controls.
* - We forward the cookie header verbatim and nothing else. No Authorization
* header, no bearer token, and never the RunPod key.
* - NOTHING IS CACHED. A cached "yes" would keep a revoked/expired session
* alive for the cache lifetime, so every AI request costs one BFF round
* trip. That is deliberate: correctness over latency for a spend gate.
* - A network failure reaching the BFF returns 503 (fail CLOSED), never 200.
* If we cannot prove a session, we do not spend GPU budget.
*
* DEMO MODE
* When the Shell is not configured (`NEXT_PUBLIC_SUPABASE_URL` unset) the CRM
* runs on its mock portal and there is no session to check, so we allow the
* request and log ONCE at startup-of-first-use. This is the same gate
* `isShellConfigured()` uses for mock-vs-real auth elsewhere in the app.
* IMPORTANT: never deploy to a public origin with the Shell unconfigured AND a
* real RUNPOD_API_KEY present that combination is an open, billable endpoint.
*
* WHAT THIS IS NOT
* This is authentication only, not authorization. It answers "is there a valid
* session", not "may this principal use the gallery". Per-resource policy for
* gallery data lives in be-crm behind the `crm.gallery` resource; if these AI
* routes ever need the same, check it there rather than re-deriving it here.
*/
export type GallerySession =
| { ok: true; principalId?: string }
| { ok: false; status: number; error: string };
/** Mirrors src/lib/appshell.ts — kept local so this stays server-only. */
function isShellConfigured(): boolean {
return Boolean(process.env.NEXT_PUBLIC_SUPABASE_URL);
}
let demoModeWarned = false;
/**
* Resolve the caller's session. Returns `{ ok: true }` (optionally with the
* principal id, used to key rate limits) or a ready-to-return failure with the
* status and message the route should emit.
*/
export async function requireGallerySession(req: Request): Promise<GallerySession> {
if (!isShellConfigured()) {
if (!demoModeWarned) {
demoModeWarned = true;
console.warn(
"[gallery-ai] Shell is not configured (NEXT_PUBLIC_SUPABASE_URL unset) — " +
"AI routes are UNAUTHENTICATED in demo mode. Do not expose this deployment publicly " +
"while RUNPOD_API_KEY is set.",
);
}
return { ok: true };
}
const cookie = req.headers.get("cookie");
if (!cookie) return { ok: false, status: 401, error: "Not signed in" };
const origin = (process.env.BFF_ORIGIN ?? "http://localhost:4000").replace(/\/$/, "");
let res: Response;
try {
res = await fetch(`${origin}/api/session/context`, {
headers: { cookie, accept: "application/json" },
// Never cache an auth decision — see the trust-model note above.
cache: "no-store",
signal: AbortSignal.timeout(5000),
});
} catch {
// Fail closed: we could not prove a session, so we do not spend GPU budget.
return { ok: false, status: 503, error: "Session service unavailable" };
}
if (res.status === 401) return { ok: false, status: 401, error: "Not signed in" };
if (!res.ok) return { ok: false, status: 503, error: "Session service unavailable" };
// The principal id is best-effort: it only sharpens the rate-limit key, so a
// shape we don't recognize degrades to IP-keyed limiting rather than failing.
let principalId: string | undefined;
try {
const data = (await res.json()) as Record<string, unknown> | null;
principalId = pickPrincipalId(data);
} catch {
/* ignore — see above */
}
return principalId ? { ok: true, principalId } : { ok: true };
}
function pickPrincipalId(data: Record<string, unknown> | null): string | undefined {
if (!data) return undefined;
const direct = data.userId ?? data.principalId ?? data.sub ?? data.id;
if (typeof direct === "string" && direct) return direct;
const user = data.user;
if (user && typeof user === "object") {
const u = user as Record<string, unknown>;
const nested = u.id ?? u.userId ?? u.sub;
if (typeof nested === "string" && nested) return nested;
}
return undefined;
}
/**
* Rate-limit key for a request: the authenticated principal when known,
* otherwise the first hop of `x-forwarded-for`.
*
* NOTE the first hop is client-controlled unless a trusted proxy overwrites the
* header. It is good enough to throttle honest clients and casual abuse; it is
* NOT a security boundary. The session gate above is the security boundary.
*/
export function rateLimitKey(req: Request, principalId?: string): string {
if (principalId) return `u:${principalId}`;
const xff = req.headers.get("x-forwarded-for") ?? "";
const first = xff.split(",")[0]?.trim();
return `ip:${first || "unknown"}`;
}