Feat : Skills

This commit is contained in:
decolua
2026-05-04 11:29:02 +07:00
parent f08fa5f78d
commit 9c6be62a54
52 changed files with 2666 additions and 1581 deletions

View File

@@ -0,0 +1,31 @@
// Shared helpers for image provider adapters
export const POLL_INTERVAL_MS = 1500;
export const POLL_TIMEOUT_MS = 120000;
export const sleep = (ms) => new Promise((r) => setTimeout(r, ms));
// Map OpenAI size to provider-specific aspect ratio
export function sizeToAspectRatio(size) {
if (!size || typeof size !== "string") return "1:1";
const map = {
"1024x1024": "1:1",
"1024x1792": "9:16",
"1792x1024": "16:9",
"1024x1536": "2:3",
"1536x1024": "3:2",
};
return map[size] || "1:1";
}
// Fetch URL → base64 (for providers returning image URLs)
export async function urlToBase64(url) {
const res = await fetch(url);
if (!res.ok) throw new Error(`Failed to fetch image: ${res.status}`);
const buf = await res.arrayBuffer();
return Buffer.from(buf).toString("base64");
}
export function nowSec() {
return Math.floor(Date.now() / 1000);
}

View File

@@ -0,0 +1,43 @@
// Black Forest Labs (FLUX) — async submit + polling_url
import { sleep, nowSec, POLL_INTERVAL_MS, POLL_TIMEOUT_MS } from "./_base.js";
const BASE_URL = "https://api.bfl.ai/v1";
export default {
async: true,
buildUrl: (model) => `${BASE_URL}/${model}`,
buildHeaders: (creds) => {
const key = creds?.apiKey || creds?.accessToken;
return { "Content-Type": "application/json", "x-key": key };
},
buildBody: (_model, body) => {
const req = { prompt: body.prompt };
if (body.size) {
const [w, h] = body.size.split("x").map(Number);
if (w) req.width = w;
if (h) req.height = h;
}
if (body.image) req.image_prompt = body.image;
return req;
},
async parseResponse(response, { headers }) {
const data = await response.json();
const pollingUrl = data.polling_url;
if (!pollingUrl) throw new Error("BFL: no polling_url returned");
const deadline = Date.now() + POLL_TIMEOUT_MS;
while (Date.now() < deadline) {
await sleep(POLL_INTERVAL_MS);
const r = await fetch(pollingUrl, { headers: { "x-key": headers["x-key"], "Accept": "application/json" } });
if (!r.ok) throw new Error(`BFL status ${r.status}`);
const s = await r.json();
if (s.status === "Ready") return s;
if (s.status === "Error" || s.status === "Failed") throw new Error(s.error || "BFL generation failed");
}
throw new Error("BFL polling timeout");
},
normalize: (responseBody) => {
const sample = responseBody.result?.sample;
if (sample) return { created: nowSec(), data: [{ url: sample }] };
return { created: nowSec(), data: [] };
},
};

View File

@@ -0,0 +1,198 @@
// Codex (ChatGPT Plus/Pro) image generation via Responses API + SSE
import { randomUUID } from "node:crypto";
import { nowSec } from "./_base.js";
const CODEX_RESPONSES_URL = "https://chatgpt.com/backend-api/codex/responses";
const CODEX_USER_AGENT = "codex-imagen/0.2.6";
const CODEX_VERSION = "0.122.0";
const CODEX_ORIGINATOR = "codex_cli_rs";
const CODEX_MODEL_SUFFIX = "-image";
const CODEX_REF_DETAIL = "high";
function decodeAccountId(idToken) {
try {
const parts = String(idToken || "").split(".");
if (parts.length !== 3) return null;
const b64 = parts[1].replace(/-/g, "+").replace(/_/g, "/");
const pad = (4 - (b64.length % 4)) % 4;
const payload = JSON.parse(Buffer.from(b64 + "=".repeat(pad), "base64").toString("utf8"));
return payload?.["https://api.openai.com/auth"]?.chatgpt_account_id || null;
} catch {
return null;
}
}
function stripImageSuffix(model) {
return model.endsWith(CODEX_MODEL_SUFFIX) ? model.slice(0, -CODEX_MODEL_SUFFIX.length) : model;
}
function toDataUrl(input) {
if (!input || typeof input !== "string") return null;
if (/^data:image\//i.test(input) || /^https?:\/\//i.test(input)) return input;
return `data:image/png;base64,${input}`;
}
function buildContent(prompt, refs, detail = CODEX_REF_DETAIL) {
const content = [];
refs.forEach((url, index) => {
content.push({ type: "input_text", text: `<image name=image${index + 1}>` });
content.push({ type: "input_image", image_url: url, detail });
content.push({ type: "input_text", text: "</image>" });
});
content.push({ type: "input_text", text: prompt });
return content;
}
// Parse Codex SSE stream → final base64 image. Optional callbacks for client streaming.
async function parseStream(response, log, callbacks = {}) {
const reader = response.body.getReader();
const decoder = new TextDecoder();
let buffer = "";
let imageB64 = null;
let lastEvent = null;
let bytesReceived = 0;
let lastProgressLogMs = 0;
while (true) {
const { done, value } = await reader.read();
if (done) break;
bytesReceived += value?.byteLength || 0;
buffer += decoder.decode(value, { stream: true });
let sepIdx;
while ((sepIdx = buffer.indexOf("\n\n")) !== -1) {
const block = buffer.slice(0, sepIdx);
buffer = buffer.slice(sepIdx + 2);
const lines = block.split("\n");
let eventName = null;
let dataStr = "";
for (const line of lines) {
if (line.startsWith("event:")) eventName = line.slice(6).trim();
else if (line.startsWith("data:")) dataStr += line.slice(5).trim();
}
if (!eventName) continue;
if (eventName !== lastEvent) {
log?.info?.("IMAGE", `codex progress: ${eventName}`);
lastEvent = eventName;
}
const now = Date.now();
if (callbacks.onProgress && now - lastProgressLogMs > 200) {
lastProgressLogMs = now;
callbacks.onProgress({ stage: eventName, bytesReceived });
}
if (eventName === "response.image_generation_call.partial_image" && dataStr) {
try {
const data = JSON.parse(dataStr);
if (callbacks.onPartialImage && data?.partial_image_b64) {
callbacks.onPartialImage({ b64_json: data.partial_image_b64, index: data.partial_image_index });
}
} catch {}
}
if (eventName === "response.output_item.done" && dataStr) {
try {
const data = JSON.parse(dataStr);
const item = data?.item;
if (item?.type === "image_generation_call" && item.result) {
imageB64 = item.result;
}
} catch {}
}
}
}
return imageB64;
}
// SSE Response that pipes codex progress + partial + done events to client
function buildSseResponse(providerResponse, log, onSuccess) {
const stream = new ReadableStream({
async start(controller) {
const enc = new TextEncoder();
const send = (event, data) => {
controller.enqueue(enc.encode(`event: ${event}\ndata: ${JSON.stringify(data)}\n\n`));
};
try {
const b64 = await parseStream(providerResponse, log, {
onProgress: (info) => send("progress", info),
onPartialImage: (info) => send("partial_image", info),
});
if (!b64) {
send("error", { message: "Codex did not return an image. Account may not be entitled (Plus/Pro required)." });
} else {
if (onSuccess) await onSuccess();
send("done", { created: nowSec(), data: [{ b64_json: b64 }] });
}
} catch (err) {
send("error", { message: err?.message || "Stream failed" });
} finally {
controller.close();
}
},
});
return new Response(stream, {
headers: {
"Content-Type": "text/event-stream",
"Cache-Control": "no-cache, no-transform",
"Connection": "keep-alive",
"X-Accel-Buffering": "no",
"Access-Control-Allow-Origin": "*",
},
});
}
export default {
stream: true,
buildUrl: () => CODEX_RESPONSES_URL,
buildHeaders: (creds) => {
const accountId = creds?.providerSpecificData?.chatgptAccountId || decodeAccountId(creds?.idToken);
return {
"accept": "text/event-stream, application/json",
"authorization": `Bearer ${creds?.accessToken || ""}`,
"chatgpt-account-id": accountId || "",
"content-type": "application/json",
"originator": CODEX_ORIGINATOR,
"session_id": randomUUID(),
"user-agent": CODEX_USER_AGENT,
"version": CODEX_VERSION,
"x-client-request-id": randomUUID(),
};
},
buildBody: (model, body) => {
const refs = [];
if (Array.isArray(body.images)) body.images.forEach((i) => { const u = toDataUrl(i); if (u) refs.push(u); });
const single = toDataUrl(body.image);
if (single) refs.push(single);
const detail = body.image_detail || CODEX_REF_DETAIL;
const imgTool = { type: "image_generation", output_format: (body.output_format || "png").toLowerCase() };
if (body.size && body.size !== "") imgTool.size = body.size;
if (body.quality && body.quality !== "") imgTool.quality = body.quality;
if (body.background && body.background !== "") imgTool.background = body.background;
return {
model: stripImageSuffix(model),
instructions: "",
input: [{ type: "message", role: "user", content: buildContent(body.prompt, refs, detail) }],
tools: [imgTool],
tool_choice: "auto",
parallel_tool_calls: false,
prompt_cache_key: randomUUID(),
stream: true,
store: false,
reasoning: null,
};
},
// Custom: codex parses SSE → either pipe to client or collect b64
async parseResponse(response, { log, streamToClient, onRequestSuccess }) {
if (streamToClient) {
return { sseResponse: buildSseResponse(response, log, onRequestSuccess) };
}
const b64 = await parseStream(response, log);
if (!b64) {
throw new Error("Codex did not return an image. Account may not be entitled (Plus/Pro required).");
}
return { created: nowSec(), data: [{ b64_json: b64 }] };
},
normalize: (responseBody) => responseBody,
};

View File

@@ -0,0 +1,8 @@
// ComfyUI — local, noAuth (placeholder; full graph workflow not implemented)
export default {
noAuth: true,
buildUrl: () => "http://localhost:8188",
buildHeaders: () => ({ "Content-Type": "application/json" }),
buildBody: (_model, body) => ({ prompt: body.prompt }),
normalize: (responseBody) => responseBody,
};

View File

@@ -0,0 +1,41 @@
// Fal.ai — async submit + queue polling
import { sleep, nowSec, sizeToAspectRatio, POLL_INTERVAL_MS, POLL_TIMEOUT_MS } from "./_base.js";
const BASE_URL = "https://queue.fal.run";
export default {
async: true,
buildUrl: (model) => `${BASE_URL}/${model}`,
buildHeaders: (creds) => {
const key = creds?.apiKey || creds?.accessToken;
return { "Content-Type": "application/json", "Authorization": `Key ${key}` };
},
buildBody: (_model, body) => {
const req = { prompt: body.prompt, num_images: body.n || 1 };
if (body.size) req.image_size = sizeToAspectRatio(body.size);
if (body.image) req.image_url = body.image;
return req;
},
async parseResponse(response, { headers }) {
const { status_url, response_url } = await response.json();
const deadline = Date.now() + POLL_TIMEOUT_MS;
while (Date.now() < deadline) {
await sleep(POLL_INTERVAL_MS);
const r = await fetch(status_url, { headers });
if (!r.ok) throw new Error(`Fal status ${r.status}`);
const s = await r.json();
if (s.status === "COMPLETED") {
const fr = await fetch(response_url, { headers });
return await fr.json();
}
if (s.status === "FAILED") throw new Error(s.error || "Fal generation failed");
}
throw new Error("Fal polling timeout");
},
normalize: (responseBody) => {
const images = Array.isArray(responseBody.images)
? responseBody.images
: (responseBody.image ? [responseBody.image] : []);
return { created: nowSec(), data: images.map((img) => ({ url: img.url || img })) };
},
};

View File

@@ -0,0 +1,25 @@
// Google Gemini adapter (Nano Banana models)
import { nowSec } from "./_base.js";
const BASE_URL = "https://generativelanguage.googleapis.com/v1beta/models";
export default {
buildUrl: (model, creds) => {
const apiKey = creds?.apiKey || creds?.accessToken;
const modelId = model.replace(/^models\//, "");
return `${BASE_URL}/${modelId}:generateContent?key=${encodeURIComponent(apiKey)}`;
},
buildHeaders: () => ({ "Content-Type": "application/json" }),
buildBody: (_model, body) => ({
contents: [{ parts: [{ text: body.prompt }] }],
generationConfig: { responseModalities: ["TEXT", "IMAGE"] },
}),
normalize: (responseBody, prompt) => {
const parts = responseBody.candidates?.[0]?.content?.parts || [];
const images = parts.filter((p) => p.inlineData?.data).map((p) => ({ b64_json: p.inlineData.data }));
return {
created: nowSec(),
data: images.length > 0 ? images : [{ b64_json: "", revised_prompt: prompt }],
};
},
};

View File

@@ -0,0 +1,22 @@
// HuggingFace Inference API — returns binary image
import { nowSec } from "./_base.js";
const BASE_URL = "https://api-inference.huggingface.co/models";
export default {
buildUrl: (model) => `${BASE_URL}/${model}`,
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) => ({ inputs: body.prompt }),
// HF returns raw image bytes — convert to b64_json
async parseResponse(response) {
const buf = await response.arrayBuffer();
const base64 = Buffer.from(buf).toString("base64");
return { created: nowSec(), data: [{ b64_json: base64 }] };
},
normalize: (responseBody) => responseBody,
};

View File

@@ -0,0 +1,37 @@
// 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";
const ADAPTERS = {
openai: createOpenAIAdapter("openai"),
minimax: createOpenAIAdapter("minimax"),
openrouter: createOpenAIAdapter("openrouter"),
recraft: createOpenAIAdapter("recraft"),
gemini,
codex,
sdwebui,
comfyui,
huggingface,
nanobanana,
"fal-ai": falAi,
"stability-ai": stabilityAi,
"black-forest-labs": blackForestLabs,
runwayml,
};
export function getImageAdapter(provider) {
return ADAPTERS[provider] || null;
}
export function isImageProvider(provider) {
return provider in ADAPTERS;
}

View File

@@ -0,0 +1,58 @@
// 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: [] };
},
};

View File

@@ -0,0 +1,33 @@
// 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",
};
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;
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,
};
}

View File

@@ -0,0 +1,47 @@
// Runway ML — async submit + /tasks/{id} polling
import { sleep, nowSec, sizeToAspectRatio, POLL_INTERVAL_MS, POLL_TIMEOUT_MS } from "./_base.js";
const BASE_URL = "https://api.dev.runwayml.com/v1";
export default {
async: true,
buildUrl: (model) => {
// Image models (gen4_image*) → text_to_image; video models → image_to_video
return `${BASE_URL}/${model.includes("image") ? "text_to_image" : "image_to_video"}`;
},
buildHeaders: (creds) => {
const key = creds?.apiKey || creds?.accessToken;
return {
"Content-Type": "application/json",
"Authorization": `Bearer ${key}`,
"X-Runway-Version": "2024-11-06",
};
},
buildBody: (model, body) => {
const isVideo = !model.includes("image");
const ratio = sizeToAspectRatio(body.size);
if (isVideo) {
return { promptText: body.prompt, model, ratio, duration: 5, ...(body.image ? { promptImage: body.image } : {}) };
}
return { promptText: body.prompt, model, ratio, ...(body.image ? { referenceImages: [{ uri: body.image }] } : {}) };
},
async parseResponse(response, { headers }) {
const { id } = await response.json();
if (!id) throw new Error("Runway: no task id returned");
const taskUrl = `${BASE_URL}/tasks/${id}`;
const deadline = Date.now() + POLL_TIMEOUT_MS;
while (Date.now() < deadline) {
await sleep(POLL_INTERVAL_MS);
const r = await fetch(taskUrl, { headers });
if (!r.ok) throw new Error(`Runway status ${r.status}`);
const s = await r.json();
if (s.status === "SUCCEEDED") return s;
if (s.status === "FAILED" || s.status === "CANCELLED") throw new Error(s.failure || "Runway task failed");
}
throw new Error("Runway polling timeout");
},
normalize: (responseBody) => {
const outputs = Array.isArray(responseBody.output) ? responseBody.output : [];
return { created: nowSec(), data: outputs.map((url) => ({ url })) };
},
};

View File

@@ -0,0 +1,17 @@
// SD WebUI (AUTOMATIC1111) — local, noAuth
import { nowSec } from "./_base.js";
export default {
noAuth: true,
buildUrl: () => "http://localhost:7860/sdapi/v1/txt2img",
buildHeaders: () => ({ "Content-Type": "application/json" }),
buildBody: (_model, body) => {
const { prompt, n = 1, size = "1024x1024" } = body;
const [width, height] = size.split("x").map(Number);
return { prompt, width: width || 512, height: height || 512, steps: 20, batch_size: n };
},
normalize: (responseBody) => {
const images = Array.isArray(responseBody.images) ? responseBody.images.map((img) => ({ b64_json: img })) : [];
return { created: nowSec(), data: images };
},
};

View File

@@ -0,0 +1,34 @@
// Stability AI v2 — sync, returns { image: "<b64>" }
import { nowSec, sizeToAspectRatio } from "./_base.js";
const BASE_URL = "https://api.stability.ai/v2beta/stable-image/generate";
// Map model id → endpoint segment
function modelToEndpoint(model) {
if (model.includes("ultra")) return "ultra";
if (model.includes("sd3")) return "sd3";
return "core";
}
export default {
buildUrl: (model) => `${BASE_URL}/${modelToEndpoint(model)}`,
buildHeaders: (creds) => {
const key = creds?.apiKey || creds?.accessToken;
return {
"Content-Type": "application/json",
"Authorization": `Bearer ${key}`,
"Accept": "application/json",
};
},
buildBody: (model, body) => {
const req = { prompt: body.prompt, output_format: (body.output_format || "png").toLowerCase() };
if (body.size) req.aspect_ratio = sizeToAspectRatio(body.size);
if (body.style) req.style_preset = body.style;
if (model.includes("sd3")) req.model = model;
return req;
},
normalize: (responseBody) => {
if (responseBody.image) return { created: nowSec(), data: [{ b64_json: responseBody.image }] };
return { created: nowSec(), data: [] };
},
};