Files
Epicure/apps/web/app/api/v1/ai/import-photo/route.ts
T
Arnaud f0340859fa feat: split photo-import into vision recognition + text generation
The photo-import flow used one vision-capable model to both read the
photo and structure the full recipe (quantities, steps, timing) in a
single call. Split it into two: a vision model recognizes what's in
the picture, then a text model reconstructs the recipe from that
description — same pattern the pantry photo-scan already uses for
recognition, and lets structuring use whichever model is actually
configured for text generation. Still counts as one AI-quota unit.

v0.37.0
2026-07-17 16:12:39 +02:00

112 lines
4.0 KiB
TypeScript

import { NextRequest, NextResponse } from "next/server";
import { z } from "zod";
import { db, recipes, recipeIngredients, recipeSteps } from "@epicure/db";
import { requireSessionOrApiKey } from "@/lib/api-auth";
import { applyRateLimit } from "@/lib/rate-limit";
import { withAiQuota, resolveAiConfigOrError } from "@/lib/ai/ai-error";
import { importFromPhoto } from "@/lib/ai/features/import-photo";
import { getModelConfigForUseCase } from "@/lib/ai/resolve-user-key";
const Schema = z.object({
imageBase64: z.string().max(14_000_000),
mimeType: z.enum(["image/jpeg", "image/png", "image/webp"]),
});
export async function POST(req: NextRequest) {
const { session, response } = await requireSessionOrApiKey(req);
if (response) return response;
const body = await req.json() as unknown;
const parsed = Schema.safeParse(body);
if (!parsed.success) {
return NextResponse.json({ error: "Validation error", issues: parsed.error.issues }, { status: 400 });
}
const limited = await applyRateLimit(`rl:ai:${session!.user.id}`, 5, 60);
if (limited) return limited;
const userId = session!.user.id;
const locale = (session!.user as { locale?: string }).locale ?? "en";
const visionConfigResult = await resolveAiConfigOrError(() => getModelConfigForUseCase(userId, "vision"));
if (!visionConfigResult.ok) return visionConfigResult.response;
const visionConfig = visionConfigResult.data;
// Fall back to vision-capable defaults if no explicit model configured
if (!visionConfig.model) {
if (visionConfig.provider === "openai") visionConfig.model = "gpt-4o";
else if (visionConfig.provider === "anthropic") visionConfig.model = "claude-sonnet-4-6";
}
const textConfigResult = await resolveAiConfigOrError(() => getModelConfigForUseCase(userId, "text"));
if (!textConfigResult.ok) return textConfigResult.response;
const textConfig = textConfigResult.data;
const result = await withAiQuota(userId, session!.user.tier as "free" | "pro", () =>
importFromPhoto(parsed.data.imageBase64, parsed.data.mimeType, visionConfig, textConfig, locale),
{ skipQuota: visionConfig.isByok && textConfig.isByok }
);
if (!result.ok) return result.response;
const recipe = result.data;
if (!recipe.found) {
return NextResponse.json({ error: "No recipe recognized in photo" }, { status: 422 });
}
const newRecipeId = crypto.randomUUID();
const now = new Date();
await db.transaction(async (tx) => {
await tx.insert(recipes).values({
id: newRecipeId,
authorId: userId,
title: recipe.title,
description: recipe.description,
baseServings: recipe.baseServings ?? 4,
recipeType: recipe.recipeType,
visibility: "private",
difficulty: recipe.difficulty,
prepMins: recipe.prepMins,
cookMins: recipe.cookMins,
dietaryTags: recipe.dietaryTags ?? {},
aiGenerated: true,
// The extraction prompt always writes the recipe in the caller's own
// locale (see importFromPhoto's langInstruction) regardless of what
// language the photo/label was in — so this *is* the recipe's
// language, not a guess, and lets the Translate button correctly
// stay hidden until the user's app language changes.
language: locale,
createdAt: now,
updatedAt: now,
});
if (recipe.ingredients.length > 0) {
await tx.insert(recipeIngredients).values(
recipe.ingredients.map((ing, i) => ({
id: crypto.randomUUID(),
recipeId: newRecipeId,
rawName: ing.rawName,
quantity: ing.quantity !== undefined ? String(ing.quantity) : undefined,
unit: ing.unit,
note: ing.note,
order: i,
}))
);
}
if (recipe.steps.length > 0) {
await tx.insert(recipeSteps).values(
recipe.steps.map((step, i) => ({
id: crypto.randomUUID(),
recipeId: newRecipeId,
instruction: step.instruction,
timerSeconds: step.timerSeconds,
order: i,
}))
);
}
});
return NextResponse.json({ id: newRecipeId });
}