@fudrouter/fsrouter 0.6.64 → 0.6.66

package/open-sse/handlers/imageProviders/index.js

@@ -0,0 +1,44 @@
+// Image provider adapter registry
+import createOpenAIAdapter from "./openai.js";
+import gemini from "./gemini.js";
+import codex from "./codex.js";
+import sdwebui from "./sdwebui.js";
+import comfyui from "./comfyui.js";
+import huggingface from "./huggingface.js";
+import nanobanana from "./nanobanana.js";
+import falAi from "./falAi.js";
+import stabilityAi from "./stabilityAi.js";
+import blackForestLabs from "./blackForestLabs.js";
+import runwayml from "./runwayml.js";
+import cloudflareAi from "./cloudflareAi.js";
+import leonardo from "./leonardo.js";
+import weavy from "./weavy.js";
+
+const ADAPTERS = {
+  openai: createOpenAIAdapter("openai"),
+  minimax: createOpenAIAdapter("minimax"),
+  openrouter: createOpenAIAdapter("openrouter"),
+  recraft: createOpenAIAdapter("recraft"),
+  xai: createOpenAIAdapter("xai"),
+  gemini,
+  codex,
+  sdwebui,
+  comfyui,
+  huggingface,
+  nanobanana,
+  "fal-ai": falAi,
+  "stability-ai": stabilityAi,
+  "black-forest-labs": blackForestLabs,
+  runwayml,
+  "cloudflare-ai": cloudflareAi,
+  leonardo,
+  weavy,
+};
+
+export function getImageAdapter(provider) {
+  return ADAPTERS[provider] || null;
+}
+
+export function isImageProvider(provider) {
+  return provider in ADAPTERS;
+}

package/open-sse/handlers/imageProviders/leonardo.js

@@ -0,0 +1,435 @@
+/**
+ * Leonardo AI image and video generation adapter.
+ *
+ * Providers: leonardo
+ * Auth: JWT Bearer token + Sentry trace headers (randomized per request)
+ * Format: GraphQL mutation (Generate) → async polling via generationId
+ * Polling: Yes — polls /generations/{id} until status == "COMPLETE"
+ *
+ * Generation paths:
+ * - v1 (SDXL-based): phoenix, flux-dev, lucid-origin — uses modelId UUID + contrast/guidance
+ * - v2 (native):     seedream, flux-2-pro, gpt-image, sora, veo, kling, etc. — no modelId param
+ * - ideogram:        v4/v3 — uses aspectRatio instead of width/height
+ *
+ * IMPORTANT — params actually forwarded by this adapter:
+ * @param {string}  prompt              - (required) Image/video description (max 1400 chars)
+ * @param {string}  [model]             - Model alias (e.g. "leo-seedream", "leo-kling-3")
+ * @param {number}  [n=1]               - Number of images/clips, capped at 4
+ * @param {string}  [size="1024x1024"]  - Output dimensions, used to derive width/height/aspect ratio
+ * @param {string}  [negative_prompt]   - Elements to exclude (forwarded directly to Leonardo API)
+ *
+ * NOTE: For video models (kling, veo, sora, seedance, etc.), the Leonardo GraphQL
+ * API itself handles model routing — this adapter sends the same params as for images.
+ * duration, resolution, aspect_ratio, image_url, video_url, end_image_url, and
+ * motion_strength are NOT currently forwarded by this adapter. They are listed in
+ * providerModels.js for future adapter support and UI documentation purposes only.
+ */
+// Leonardo AI — GraphQL-based image generation adapter
+// Uses cookie → JWT auth and Leonardo's native GraphQL API.
+// Reference: kliperspro/backend/services/leonardo-service.js
+import { sleep, nowSec, POLL_INTERVAL_MS } from "./_base.js";
+import { randomBytes } from "crypto";
+
+// Read Leonardo admin config (promptEnhance, capabilities etc.) from KV store.
+// Falls back to empty object if DB not available (e.g., first boot).
+async function getLeonardoAdminConfig() {
+  try {
+    const { getAdapter } = await import('../../dist/lib/db/driver.js');
+    const db = await getAdapter();
+    const row = db.get(`SELECT value FROM kv WHERE scope = 'leonardo' AND key = 'admin_config'`);
+    if (!row?.value) return null;
+    return typeof row.value === "string" ? JSON.parse(row.value) : row.value;
+  } catch {
+    return null;
+  }
+}
+
+let _adminConfigCache = null;
+let _adminConfigAt = 0;
+const ADMIN_CONFIG_TTL = 30_000; // 30 seconds
+
+async function getModelConfig(modelId) {
+  const now = Date.now();
+  if (!_adminConfigCache || now - _adminConfigAt > ADMIN_CONFIG_TTL) {
+    _adminConfigCache = await getLeonardoAdminConfig();
+    _adminConfigAt = now;
+  }
+  if (!_adminConfigCache?.models) return null;
+  const slug = (modelId || "").replace(/^leo-/, "").toLowerCase();
+  return _adminConfigCache.models.find(m =>
+    (m.id || "").replace(/^leo-/, "").toLowerCase() === slug ||
+    (m.apiModelName || "").toLowerCase() === slug
+  ) || null;
+}
+
+const GRAPHQL_URL = "https://api.leonardo.ai/v1/graphql";
+const SENTRY_REL = "6a0bd1b5b7ef23a4f22608a2ed90c5e753cbc669";
+
+const POLL_TIMEOUT_MS = 120_000; // 2 minutes
+const POLL_INTERVAL_LEONARDO = 3000; // 3 seconds
+
+// ============================================
+// MODEL UUID REGISTRY — mirrors leoapi-main model_id.txt
+// ============================================
+const MODEL_REGISTRY = {
+  // Nano Banana
+  "nano-banana-2":    "7418e71f-4133-4e1b-9895-bee19f48f2ce",
+  "nano-banana-pro":  "7c02ef35-3a6b-4df6-b78d-873e5032c3b4",
+  "nano-banana":      "4a008a65-8d97-44f5-97a0-66c431612614",
+  // Seedream
+  "seedream-4.5":     "f1c295ea-1575-445f-89ae-9b4013a6a37c",
+  "seedream-4":       "94515e81-e589-4a5b-aeae-10ced50142c2",
+  "seedream":         "f1c295ea-1575-445f-89ae-9b4013a6a37c",
+  // GPT Image
+  "gpt-image-2":      "99ecc726-3404-412c-9dc1-24d4cdef2299",
+  "gpt-image-1.5":    "99ecc726-3404-412c-9dc1-24d4cdef2299",
+  "gpt-image-1":      "f75b1998-e5cb-4fdf-9eef-98e8186c2c2f",
+  "gpt-image":        "99ecc726-3404-412c-9dc1-24d4cdef2299",
+  // Flux
+  "flux-2-pro":       "5478273a-68e1-4efe-a0c4-3fe84e4c16a8",
+  "flux-2-dev":       "c19631a4-21b4-4dbd-b015-b446b7a4e0e0",
+  "flux-dev":         "b2614463-296c-462a-9586-aafdb8f00e36",
+  "flux-schnell":     "1dd50843-d653-4516-a8e3-f0238ee453ff",
+  "flux-kontext-max": "02dff998-e678-416c-a8a7-ce93188f2e68",
+  "flux-kontext":     "28aeddf8-bd19-4803-80fc-79602d1a9989",
+  // Lucid
+  "lucid-origin":     "7b592283-e8a7-4c5a-9ba6-d18c31f258b9",
+  "lucid-realism":    "05ce0082-2d80-4a2d-8653-4d1c85e2418e",
+  // Ideogram (uuid only for v3; v4 = null → native v2)
+  "ideogram-3":       "f9672904-3313-4867-b883-407ef6a0edec",
+  // Phoenix
+  phoenix:            "de7d3faf-762f-48e0-b3b7-9d0ac3a3fcf3",
+  "phoenix-1.0":      "de7d3faf-762f-48e0-b3b7-9d0ac3a3fcf3",
+  // XL models
+  "kino-xl":          "aa77f04e-3eec-4034-9c07-d0f619684628",
+  "anime-xl":         "e71a1c2f-4f80-4800-934f-2c68979d8cc8",
+  "lightning-xl":     "b24e16ff-06e3-43eb-8d33-4416c2d75876",
+  "vision-xl":        "5c232a9e-9061-4777-980a-ddc8e65647c6",
+  "diffusion-xl":     "1e60896f-3c26-4296-8ecc-53e2afecc132",
+  // LTX video UUIDs (needed for video adapter if passed as image model by mistake)
+  "ltxv-2.3-pro":     "938ebb92-045c-48b5-a02b-9489b4933f60",
+  "ltxv-2.3-fast":    "0eba7ca6-57b0-4b9f-9ff3-fd78af067109",
+  // Kling video UUIDs
+  "kling":            "6c904469-5291-4043-b610-f53b50dfd6ff",
+  "kling-omni":       "0d5109cf-d256-4720-86d3-d8e5ff5a3ce2",
+  // Note: gemini-image-2, recraft-v4, recraft-v4-pro, ideogram-v4, sora-2,
+  // veo-*, hailuo-*, seedance-*, happy-horse all have uuid=null in DEFAULT_CONFIG
+  // → they are treated as native v2 models (no modelId in generation params)
+};
+
+// ============================================
+// DIMENSION TABLES
+// ============================================
+const STANDARD_SIZES  = { "16:9": [1920, 1080], "9:16": [1080, 1920], "1:1": [1024, 1024], "4:3": [1440, 1080], "3:4": [1080, 1440], "3:2": [1620, 1080], "2:3": [1080, 1620] };
+const SEEDREAM_SIZES  = { "16:9": [2752, 1536], "9:16": [1536, 2752], "1:1": [1536, 1536], "4:3": [2048, 1536], "3:4": [1536, 2048], "3:2": [2048, 1360], "2:3": [1360, 2048] };
+const PHOENIX_SIZES   = { "16:9": [2752, 1536], "9:16": [1536, 2752], "1:1": [1536, 1536], "4:3": [2048, 1536], "3:2": [2048, 1360] };
+const GPT_IMAGE_SIZES = { "16:9": [1376,  768], "9:16": [ 768, 1376], "1:1": [1024, 1024], "4:3": [1184,  888], "3:4": [ 888, 1184], "3:2": [1248,  832], "2:3": [ 832, 1248] };
+
+// ============================================
+// HELPERS
+// ============================================
+
+function makeHex(n) { return randomBytes(n).toString("hex"); }
+
+/** Build Sentry trace headers (required by Leonardo API gateway) */
+function sentryHeaders(token) {
+  const tid = makeHex(8) + makeHex(8);
+  return {
+    authorization: `Bearer ${token}`,
+    "sentry-trace": `${tid}-${makeHex(8).slice(0, 16)}-0`,
+    baggage: `sentry-environment=vercel-production,sentry-release=${SENTRY_REL},sentry-public_key=a851bd902378477eae99cf74c62e142a,sentry-trace_id=${tid},sentry-org_id=4504767521292288,sentry-sampled=false`,
+  };
+}
+
+/** Execute a GraphQL request against Leonardo's API */
+async function gql(token, payload) {
+  const res = await fetch(GRAPHQL_URL, {
+    method: "POST",
+    headers: {
+      accept: "*/*",
+      "accept-language": "en-US,en;q=0.9",
+      "content-type": "application/json",
+      origin: "https://app.leonardo.ai",
+      referer: "https://app.leonardo.ai/",
+      "user-agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 Chrome/120.0.0.0 Safari/537.36",
+      "x-leo-schema-version": "latest",
+      "sec-fetch-dest": "empty",
+      "sec-fetch-mode": "cors",
+      "sec-fetch-site": "same-site",
+      ...sentryHeaders(token),
+    },
+    body: JSON.stringify(payload),
+  });
+  if (!res.ok) throw new Error(`Leonardo GraphQL HTTP ${res.status}`);
+  return res.json();
+}
+
+/** Resolve model UUID from name/id pattern */
+function resolveModelUuid(modelId) {
+  if (!modelId) return null;
+  const input = modelId.trim();
+
+  // Pass-through raw UUIDs
+  if (/^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i.test(input)) {
+    return input;
+  }
+
+  // Strip 'leo-' prefix
+  const name = input.replace(/^leo-/, "").toLowerCase();
+  if (MODEL_REGISTRY[name]) return MODEL_REGISTRY[name];
+  const normalized = name.replace(/[_.]/g, "-");
+  if (MODEL_REGISTRY[normalized]) return MODEL_REGISTRY[normalized];
+  for (const [key, uuid] of Object.entries(MODEL_REGISTRY)) {
+    if (key.includes(name) || name.includes(key)) return uuid;
+  }
+  return null;
+}
+
+/** Convert OpenAI size string (e.g. "1024x1024") → Leonardo aspect ratio → [width, height] */
+function sizeToLeonardoDimensions(size, modelId = "") {
+  const sizeMap = {
+    "1024x1024": "1:1",
+    "1024x1792": "9:16",
+    "1792x1024": "16:9",
+    "1024x1536": "2:3",
+    "1536x1024": "3:2",
+  };
+  const ar = sizeMap[size] || "1:1";
+
+  const lower = (modelId || "").toLowerCase();
+  const isSeedream = lower.includes("seedream");
+  const isPhoenix  = lower.includes("phoenix") || lower.includes("nano-banana");
+  const isGpt      = lower.includes("gpt-image");
+
+  if (isSeedream  && SEEDREAM_SIZES[ar])  return { ar, dims: SEEDREAM_SIZES[ar] };
+  if (isPhoenix   && PHOENIX_SIZES[ar])   return { ar, dims: PHOENIX_SIZES[ar] };
+  if (isGpt       && GPT_IMAGE_SIZES[ar]) return { ar, dims: GPT_IMAGE_SIZES[ar] };
+  return { ar, dims: STANDARD_SIZES[ar] || [1024, 1024] };
+}
+
+/** Derive the API model name sent in the mutation (strip leading 'leo-') */
+function resolveApiModelName(modelId) {
+  const clean = (modelId || "").replace(/^leo-/, "").toLowerCase();
+  const maps = {
+    "phoenix": "phoenix-v1.0",
+    "seedream": "seedream-4.5",
+    "gpt-image": "gpt-image-1.5",
+    "ideogram-v4": "ideogram-v4.0",
+    "recraft-v4-pro": "recraft-v4",
+    "happy-horse": "happy-horse",
+    "kling": "kling-2.6",
+    "kling-3": "kling-3.0",
+    "kling-3-turbo": "kling-3.0-turbo",
+    "kling-o1": "kling-video-o-1",
+    "kling-o3": "kling-video-o-3",
+    "ltx-pro": "ltxv-2.3-pro",
+    "ltx-fast": "ltxv-2.3-fast",
+    "ltx-2-pro": "ltxv-2.3-pro",
+    "ltx-2-fast": "ltxv-2.3-fast",
+  };
+  return maps[clean] || clean;
+}
+
+/**
+ * Returns the generation API version for a model:
+ *  - "v1" → legacy SDXL path: num_images + modelId UUID + contrast (Phoenix, Flux Dev, Lucid Origin)
+ *  - "v2" → native path: quantity, no modelId (Seedream, GPT Image, Flux 2 Pro, etc.)
+ */
+function getModelApiVersion(modelId) {
+  const name = (modelId || "").toLowerCase();
+  const isV1 = (
+    name === "leo-flux-dev" || name === "flux-dev" ||
+    name.includes("lucid-origin")
+  );
+  return isV1 ? "v1" : "v2";
+}
+
+// ============================================
+// POLLING
+// ============================================
+
+async function pollGenerationStatus(token, genId) {
+  const query = {
+    operationName: "GetAIGenerationFeedStatuses",
+    variables: { where: { id: { _eq: genId } } },
+    query: `query GetAIGenerationFeedStatuses($where: generations_bool_exp = {}) {
+  generations(where: $where) { id status __typename }
+}`,
+  };
+  const rj = await gql(token, query);
+  const gen = (rj?.data?.generations || [])[0];
+  return gen?.status || "PENDING";
+}
+
+async function fetchGeneratedImageUrls(token, genId) {
+  const query = {
+    operationName: "GetAIGenerationFeed",
+    variables: { where: { id: { _eq: genId } }, limit: 4 },
+    query: `query GetAIGenerationFeed($where: generations_bool_exp = {}, $limit: Int) {
+  generations(where: $where, limit: $limit) {
+    generated_images(order_by: [{url: desc}]) { url id __typename }
+    __typename
+  }
+}`,
+  };
+  const rj = await gql(token, query);
+  const gen = (rj?.data?.generations || [])[0];
+  return (gen?.generated_images || []).map((img) => img.url).filter(Boolean);
+}
+
+async function waitForImages(token, genId, timeoutMs = POLL_TIMEOUT_MS) {
+  const deadline = Date.now() + timeoutMs;
+  while (Date.now() < deadline) {
+    await sleep(POLL_INTERVAL_LEONARDO);
+    const status = await pollGenerationStatus(token, genId);
+    if (status === "COMPLETE" || status === "COMPLETED") {
+      return await fetchGeneratedImageUrls(token, genId);
+    }
+    if (status === "FAILED" || status === "ERROR") {
+      throw new Error(`Leonardo generation failed (status=${status})`);
+    }
+  }
+  // Timeout — try one last fetch
+  try {
+    const urls = await fetchGeneratedImageUrls(token, genId);
+    if (urls.length) return urls;
+  } catch {}
+  throw new Error("Leonardo generation timed out");
+}
+
+// ============================================
+// ADAPTER
+// ============================================
+
+export default {
+  /** Credentials stored as cookie (apiKey) + resolved JWT (accessToken) */
+  buildUrl: (_model, _creds) => GRAPHQL_URL,
+
+  buildHeaders: (creds) => {
+    const token = creds?.accessToken || creds?.apiKey || "";
+    return {
+      accept: "*/*",
+      "accept-language": "en-US,en;q=0.9",
+      "content-type": "application/json",
+      origin: "https://app.leonardo.ai",
+      referer: "https://app.leonardo.ai/",
+      "user-agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 Chrome/120.0.0.0 Safari/537.36",
+      "x-leo-schema-version": "latest",
+      "sec-fetch-dest": "empty",
+      "sec-fetch-mode": "cors",
+      "sec-fetch-site": "same-site",
+      ...sentryHeaders(token),
+    };
+  },
+
+  buildBody: async (modelId, body) => {
+    const n = Math.min(body.n || 1, 4);
+    const { ar, dims } = sizeToLeonardoDimensions(body.size, modelId);
+    const [width, height] = dims;
+    const prompt = (body.prompt || "").slice(0, 1400);
+    const modelCfg = await getModelConfig(modelId);
+    const apiModelName = modelCfg?.apiModelName || resolveApiModelName(modelId);
+    const modelUuid = resolveModelUuid(modelId);
+    const apiVersion = getModelApiVersion(modelId);
+    const promptEnhance = body.prompt_enhance ?? (modelCfg?.promptEnhance || "OFF");
+
+    let params;
+
+    if (apiVersion === "v1") {
+      // ── v1 legacy path (Phoenix, Flux Dev, Lucid Origin) ──
+      // Requires: num_images, modelId UUID, contrast
+      params = {
+        width,
+        height,
+        prompt,
+        num_images: n,
+        modelId: modelUuid,
+        prompt_enhance: promptEnhance,
+        contrast: body.contrast ?? 3.5,  // required for Phoenix/Lucid
+      };
+      if (body.guidance_scale != null) params.guidance_scale = body.guidance_scale;
+      if (body.num_inference_steps != null) params.num_inference_steps = body.num_inference_steps;
+    } else {
+      // ── v2 native path (Seedream, GPT Image, Flux 2 Pro, etc.) ──
+      params = {
+        prompt,
+        quantity: n,
+        prompt_enhance: promptEnhance,
+      };
+      if (apiModelName.toLowerCase().includes("ideogram")) {
+        params.aspectRatio = ar;
+      } else {
+        params.width = width;
+        params.height = height;
+      }
+      if (modelCfg?.quality) params.quality = modelCfg.quality;
+    }
+
+    if (body.negative_prompt) params.negative_prompt = body.negative_prompt;
+
+    return {
+      operationName: "Generate",
+      variables: {
+        request: {
+          model: apiModelName,
+          public: body.is_public ?? false,
+          parameters: params,
+        },
+      },
+      query: `mutation Generate($request: CreateGenerationRequest!) {
+  generate(request: $request) {
+    apiCreditCost generationId __typename
+  }
+}`,
+    };
+  },
+
+  /**
+   * Overrides standard parse flow — after submitting the mutation we need to
+   * poll until the generation completes, then return OpenAI-format image URLs.
+   *
+   * The imageGenerationCore passes:
+   *   { headers, log, streamToClient, onRequestSuccess, url, requestBody, model, body }
+   * `headers` is the request headers object (plain dict) containing `authorization`.
+   */
+  async parseResponse(response, { headers, log }) {
+    const rj = await response.json();
+
+    // Propagate GraphQL errors
+    if (rj?.errors?.length) {
+      const msg = rj.errors.map((e) => e.message).join(" | ");
+      throw new Error(`Leonardo generate error: ${msg}`);
+    }
+
+    const genId = rj?.data?.generate?.generationId;
+    if (!genId) {
+      throw new Error(`Leonardo: no generationId in response: ${JSON.stringify(rj).slice(0, 300)}`);
+    }
+
+    log?.debug?.("LEONARDO", `Generation submitted genId=${genId}`);
+
+    // Extract the Bearer JWT from the request Authorization header
+    const authHeader = (headers?.authorization || headers?.Authorization || "");
+    const token = authHeader.replace(/^Bearer\s+/i, "").trim();
+
+    if (!token) {
+      throw new Error("Leonardo: no JWT token found in request headers");
+    }
+
+    const urls = await waitForImages(token, genId);
+    log?.debug?.("LEONARDO", `Generation complete — ${urls.length} image(s) fetched`);
+
+    // Return in OpenAI image-response format
+    return {
+      created: nowSec(),
+      data: urls.map((url) => ({ url })),
+    };
+  },
+
+  normalize: (parsed) => {
+    // Already normalized by parseResponse above
+    if (parsed?.created && Array.isArray(parsed?.data)) return parsed;
+    const urls = Array.isArray(parsed?.data) ? parsed.data : [];
+    return { created: nowSec(), data: urls };
+  },
+};

package/open-sse/handlers/imageProviders/nanobanana.js

@@ -0,0 +1,79 @@
+/**
+ * NanoBanana image generation adapter.
+ *
+ * Providers: nanobanana
+ * Auth: Bearer token (apiKey or accessToken)
+ * Format: JSON body → async polling via task status endpoint
+ * Polling: Yes — polls until resultImageUrl is available
+ *
+ * Supported request params:
+ * @param {string}  prompt               - (required) Image description
+ * @param {number}  [n=1]                - Number of images (numImages)
+ * @param {string}  [size="1024x1024"]   - Maps to aspect ratio: "1:1"|"9:16"|"16:9"|"4:3"|"3:2"|"2:3"
+ * @param {string}  [image]              - Reference image URL → enables IMAGETOIAMGE mode
+ * @param {string[]} [images]            - Multiple reference image URLs
+ *
+ * Request type:
+ * - No image → type: "TEXTTOIAMGE"
+ * - image/images provided → type: "IMAGETOIAMGE"
+ *
+ * Response normalize: extracts resultImageUrl (falls back to originImageUrl).
+ */
+// NanoBanana API — async submit + poll record-info
+import { sleep, nowSec, sizeToAspectRatio, POLL_INTERVAL_MS, POLL_TIMEOUT_MS } from "./_base.js";
+
+const SUBMIT_URL = "https://api.nanobananaapi.ai/api/v1/nanobanana/generate";
+const POLL_BASE = "https://api.nanobananaapi.ai/api/v1/nanobanana/record-info";
+
+export default {
+  async: true,
+  buildUrl: () => SUBMIT_URL,
+  buildHeaders: (creds) => {
+    const headers = { "Content-Type": "application/json" };
+    const key = creds?.apiKey || creds?.accessToken;
+    if (key) headers["Authorization"] = `Bearer ${key}`;
+    return headers;
+  },
+  buildBody: (_model, body) => {
+    const ratio = sizeToAspectRatio(body.size);
+    const isEdit = !!(body.image || (Array.isArray(body.images) && body.images.length));
+    const req = {
+      prompt: body.prompt,
+      type: isEdit ? "IMAGETOIAMGE" : "TEXTTOIAMGE",
+      numImages: body.n || 1,
+      image_size: ratio,
+      // API requires callBackUrl; we poll instead so a dummy URL is fine.
+      callBackUrl: "https://localhost/callback",
+    };
+    if (isEdit) {
+      const urls = Array.isArray(body.images) ? body.images.filter(Boolean) : [];
+      if (body.image) urls.push(body.image);
+      req.imageUrls = urls;
+    }
+    return req;
+  },
+  // Async: parse submit → poll until SUCCESS, return raw poll data
+  async parseResponse(response, { headers }) {
+    const submitData = await response.json();
+    if (submitData.code !== 200) throw new Error(submitData.msg || "NanoBanana submit failed");
+    const taskId = submitData.data?.taskId;
+    if (!taskId) throw new Error("NanoBanana: no taskId returned");
+    const pollUrl = `${POLL_BASE}?taskId=${encodeURIComponent(taskId)}`;
+    const deadline = Date.now() + POLL_TIMEOUT_MS;
+    while (Date.now() < deadline) {
+      await sleep(POLL_INTERVAL_MS);
+      const r = await fetch(pollUrl, { headers });
+      if (!r.ok) throw new Error(`NanoBanana status ${r.status}`);
+      const s = await r.json();
+      const flag = s.data?.successFlag;
+      if (flag === 1) return s.data;
+      if (flag === 2 || flag === 3) throw new Error(s.data?.errorMessage || "NanoBanana generation failed");
+    }
+    throw new Error("NanoBanana polling timeout");
+  },
+  normalize: (responseBody, prompt) => {
+    const url = responseBody.response?.resultImageUrl || responseBody.response?.originImageUrl;
+    if (url) return { created: nowSec(), data: [{ url, revised_prompt: prompt }] };
+    return { created: nowSec(), data: [] };
+  },
+};

package/open-sse/handlers/imageProviders/openai.js

@@ -0,0 +1,61 @@
+/**
+ * OpenAI-compatible image generation adapter.
+ *
+ * Providers: openai | minimax | openrouter | recraft | xai
+ * Auth: Bearer token via Authorization header (apiKey or accessToken)
+ * Format: JSON body → sync JSON response
+ * Polling: None (synchronous)
+ *
+ * Supported request params:
+ * @param {string}  prompt                    - (required) Image description
+ * @param {string}  [model]                   - Model ID (set by caller)
+ * @param {number}  [n=1]                     - Number of images (1–10 for DALL-E 2; 1 for DALL-E 3/GPT Image 1)
+ * @param {string}  [size="1024x1024"]        - Dimensions: "256x256"|"512x512"|"1024x1024"|"1792x1024"|"1024x1792"
+ * @param {string}  [quality="standard"]      - "standard"|"hd"|"low"|"medium"|"high"|"auto"
+ * @param {string}  [style]                   - DALL-E 3 only: "vivid"|"natural"
+ * @param {string}  [response_format="url"]   - "url"|"b64_json"
+ * @param {string}  [output_format]           - GPT Image 1 only: "png"|"jpeg"|"webp"
+ * @param {string}  [background]              - GPT Image 1 only: "auto"|"transparent"|"opaque"
+ *
+ * xAI note: only prompt, model, n, response_format are forwarded.
+ */
+// OpenAI-compatible adapter (used by openai, minimax, openrouter, recraft)
+
+const ENDPOINTS = {
+  openai: "https://api.openai.com/v1/images/generations",
+  minimax: "https://api.minimaxi.com/v1/images/generations",
+  openrouter: "https://openrouter.ai/api/v1/images/generations",
+  recraft: "https://external.api.recraft.ai/v1/images/generations",
+  xai: "https://api.x.ai/v1/images/generations",
+};
+
+export default function createOpenAIAdapter(providerId) {
+  return {
+    buildUrl: () => ENDPOINTS[providerId],
+    buildHeaders: (creds) => {
+      const headers = { "Content-Type": "application/json" };
+      const key = creds?.apiKey || creds?.accessToken;
+      if (key) headers["Authorization"] = `Bearer ${key}`;
+      if (providerId === "openrouter") {
+        headers["HTTP-Referer"] = "https://endpoint-proxy.local";
+        headers["X-Title"] = "Endpoint Proxy";
+      }
+      return headers;
+    },
+    buildBody: (model, body) => {
+      const { prompt, n = 1, size = "1024x1024", quality, style, response_format } = body;
+      // xAI only accepts prompt, model, n, response_format
+      if (providerId === "xai") {
+        const req = { model, prompt, n };
+        if (response_format) req.response_format = response_format;
+        return req;
+      }
+      const req = { model, prompt, n, size };
+      if (quality) req.quality = quality;
+      if (style) req.style = style;
+      if (response_format) req.response_format = response_format;
+      return req;
+    },
+    normalize: (responseBody) => responseBody,
+  };
+}