From 579730640590ce9b3749391f34f01c499b7392d2 Mon Sep 17 00:00:00 2001 From: Glfcai-qiu Date: Wed, 24 Jun 2026 10:26:31 +0800 Subject: [PATCH 1/2] =?UTF-8?q?feat:=20=E6=96=B0=E5=A2=9E=20fal.ai=20?= =?UTF-8?q?=E5=8D=8F=E8=AE=AE=E8=AF=86=E5=88=AB=E3=80=81=E6=A8=A1=E5=9E=8B?= =?UTF-8?q?=E6=8B=89=E5=8F=96=E4=B8=8E=E5=9B=BE=E7=89=87=E7=94=9F=E6=88=90?= =?UTF-8?q?=E6=94=AF=E6=8C=81?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- main.py | 312 ++++++++++++++++++++++++++++++++++++-- static/api-settings.html | 15 +- static/js/api-settings.js | 59 ++++--- 3 files changed, 352 insertions(+), 34 deletions(-) diff --git a/main.py b/main.py index 74321e993..8ccf7c7c3 100644 --- a/main.py +++ b/main.py @@ -263,9 +263,17 @@ async def websocket_endpoint(websocket: WebSocket, client_id: str = None): } PROVIDER_ID_RE = re.compile(r"^[a-zA-Z0-9_-]{2,40}$") -SUPPORTED_PROVIDER_PROTOCOLS = {"openai", "apimart", "gemini", "volcengine", "runninghub", "jimeng"} +SUPPORTED_PROVIDER_PROTOCOLS = {"openai", "apimart", "gemini", "volcengine", "runninghub", "jimeng", "fal"} SUPPORTED_IMAGE_REQUEST_MODES = {"openai", "openai-json"} RUNNINGHUB_DEFAULT_BASE_URL = "https://www.runninghub.cn" +FAL_DEFAULT_BASE_URL = "https://queue.fal.run" +FAL_MODELS_API_URL = "https://api.fal.ai/v1/models" +FAL_DEFAULT_IMAGE_MODELS = [ + "fal-ai/flux/schnell", + "fal-ai/flux/dev", + "fal-ai/flux-pro", + "fal-ai/nano-banana", +] RUNNINGHUB_OPENAPI_BASE_URL = "https://www.runninghub.cn/openapi/v2" RUNNINGHUB_MODEL_REGISTRY_URL = "https://raw.githubusercontent.com/HM-RunningHub/ComfyUI_RH_OpenAPI/main/models_registry.json" RUNNINGHUB_LLM_BASE_URL = "https://llm.runninghub.cn/v1" @@ -771,6 +779,22 @@ def default_api_providers(): "ms_loras": [], "ms_defaults_version": 0, }, + { + "id": "fal", + "name": "fal.ai", + "base_url": FAL_DEFAULT_BASE_URL, + "protocol": "fal", + "image_request_mode": "openai", + "image_generation_endpoint": "", + "image_edit_endpoint": "", + "enabled": True, + "primary": False, + "image_models": FAL_DEFAULT_IMAGE_MODELS, + "chat_models": [], + "video_models": [], + "ms_loras": [], + "ms_defaults_version": 0, + }, ] def merge_default_api_providers(providers): @@ -850,6 +874,19 @@ def merge_default_api_providers(providers): protocols.update(normalize_model_protocols(lingjing_default.get("model_protocols"))) current["model_protocols"] = protocols # 即梦 CLI 不再是强制保留的默认平台：仅在用户已添加了即梦协议的平台时，规范化其默认模型/地址。 + fal_default = next((d for d in default_api_providers() if d["id"] == "fal"), None) + if fal_default: + current = next((item for item in merged if item.get("id") == "fal"), None) + if not current: + merged.append(fal_default) + else: + if not current.get("base_url"): + current["base_url"] = fal_default["base_url"] + current["protocol"] = "fal" + current["image_request_mode"] = normalize_image_request_mode(current.get("image_request_mode")) + current["image_models"] = model_list_from_values([*(current.get("image_models") or []), *(fal_default.get("image_models") or [])]) + current["chat_models"] = model_list_from_values(current.get("chat_models") or []) + current["video_models"] = [] for current in merged: if not is_jimeng_provider(current): continue @@ -1170,6 +1207,9 @@ def normalize_provider(item): if provider_id == "runninghub": protocol = "runninghub" base_url = base_url or RUNNINGHUB_DEFAULT_BASE_URL + if provider_id == "fal": + protocol = "fal" + base_url = base_url or FAL_DEFAULT_BASE_URL return { "id": provider_id, "name": name, @@ -3779,7 +3819,7 @@ def provider_protocol(provider): # 单模型可覆盖的协议（仅 OpenAI / Gemini，二者可共用同一站点的 Base URL + Key） PER_MODEL_PROTOCOL_OPTIONS = {"openai", "gemini"} # 协议固定、不支持单模型覆盖的内置平台 -FIXED_PROTOCOL_PROVIDER_IDS = {"modelscope", "volcengine", "jimeng", "runninghub"} +FIXED_PROTOCOL_PROVIDER_IDS = {"modelscope", "volcengine", "jimeng", "runninghub", "fal"} def normalize_model_protocols(value): """规整 {模型名: 协议} 覆盖表，仅保留 openai/gemini。""" @@ -3836,6 +3876,9 @@ def is_runninghub_provider(provider): def is_jimeng_provider(provider): return provider_protocol(provider) == "jimeng" or str((provider or {}).get("id") or "").strip().lower() == "jimeng" +def is_fal_provider(provider): + return provider_protocol(provider) == "fal" or str((provider or {}).get("id") or "").strip().lower() == "fal" + def is_yuli_provider(provider): # 玉玉API（yuli.host）的视频接口走自有格式（/v1/video/create + /v1/video/query）， # 与通用 OpenAI /v1/videos/generations 不同，需单独识别。 @@ -8625,6 +8668,132 @@ async def generate_runninghub_video(payload, provider): local_urls = [await save_remote_video_to_output(url, prefix="rh_video_") for url in urls] return {"videos": local_urls, "task_id": task_id, "raw": result} +def fal_api_key(provider=None): + provider_id = str((provider or {}).get("id") or "fal").strip().lower() or "fal" + api_key = str((provider or {}).get("api_key") or "").strip() or provider_env_key_value(provider_id) + if not api_key: + raise HTTPException(status_code=400, detail="未配置 fal.ai API Key，请在 API 设置中填写。") + return strip_auth_scheme(api_key, "Key") + +def fal_headers(provider=None, json_body=True): + headers = { + "Authorization": f"Key {fal_api_key(provider)}", + "Accept": "application/json", + } + if json_body: + headers["Content-Type"] = "application/json" + return headers + +def fal_queue_base_url(provider=None): + base_url = str((provider or {}).get("base_url") or FAL_DEFAULT_BASE_URL).strip().rstrip("/") + return base_url or FAL_DEFAULT_BASE_URL + +def fal_queue_url(provider, model): + model_id = selected_model(model, FAL_DEFAULT_IMAGE_MODELS[0]).strip().strip("/") + if not model_id: + raise HTTPException(status_code=400, detail="fal.ai 模型名称不能为空") + if model_id.startswith(("http://", "https://")): + return model_id + return f"{fal_queue_base_url(provider)}/{model_id}" + +def fal_image_size(size): + width, height = parse_size_pair(size) + if not width or not height: + return "square_hd" + ratio = width / max(1, height) + if abs(ratio - 1.0) < 0.08: + return "square_hd" + if abs(ratio - (4 / 3)) < 0.08: + return "landscape_4_3" + if abs(ratio - (3 / 4)) < 0.08: + return "portrait_4_3" + if abs(ratio - (16 / 9)) < 0.08: + return "landscape_16_9" + if abs(ratio - (9 / 16)) < 0.08: + return "portrait_16_9" + return {"width": width, "height": height} + +def fal_reference_url(ref): + return reference_to_data_url(ref, max_size=1536) + +def fal_submit_payload(prompt, size, model, reference_images=None): + body = { + "prompt": prompt, + "image_size": fal_image_size(size), + } + refs = [fal_reference_url(ref) for ref in (reference_images or [])[:ONLINE_IMAGE_REFERENCE_MAX]] + refs = [url for url in refs if url] + if refs: + model_lc = str(model or "").lower() + if len(refs) > 1 or "nano-banana" in model_lc: + body["image_urls"] = refs + else: + body["image_url"] = refs[0] + return body + +def fal_response_url(provider, model, request_id): + model_id = selected_model(model, FAL_DEFAULT_IMAGE_MODELS[0]).strip().strip("/") + base = fal_queue_base_url(provider) + return f"{base}/{model_id}/requests/{urllib.parse.quote(str(request_id), safe='')}" + +def fal_status_value(raw): + if not isinstance(raw, dict): + return "" + return str(raw.get("status") or raw.get("state") or "").strip().upper() + +def fal_failure_message(raw): + if not isinstance(raw, dict): + return str(raw or "") + error = raw.get("error") or raw.get("detail") or raw.get("message") or raw.get("logs") + if isinstance(error, str): + return error + if isinstance(error, (dict, list)): + return json.dumps(error, ensure_ascii=False)[:800] + return "" + +async def wait_for_fal_result(client, provider, model, submit_raw): + request_id = str((submit_raw or {}).get("request_id") or "").strip() + status_url = str((submit_raw or {}).get("status_url") or "").strip() + response_url = str((submit_raw or {}).get("response_url") or "").strip() + if not response_url and request_id: + response_url = fal_response_url(provider, model, request_id) + if not status_url and response_url: + status_url = response_url + if not response_url: + raise HTTPException(status_code=502, detail=f"fal.ai 未返回 response_url/request_id：{submit_raw}") + + deadline = time.monotonic() + IMAGE_TASK_TIMEOUT + last_payload = submit_raw + headers = fal_headers(provider, json_body=False) + while status_url and time.monotonic() < deadline: + await asyncio.sleep(IMAGE_POLL_INTERVAL) + status_response = await client.get(status_url, headers=headers) + status_response.raise_for_status() + last_payload = status_response.json() if status_response.text else {} + status = fal_status_value(last_payload) + if status in {"COMPLETED", "SUCCEEDED", "SUCCESS", "DONE"}: + break + if status in {"FAILED", "ERROR", "CANCELED", "CANCELLED"}: + raise HTTPException(status_code=502, detail=f"fal.ai 任务失败：{fal_failure_message(last_payload) or last_payload}") + else: + raise HTTPException(status_code=504, detail=f"fal.ai 生图任务超时：request_id={request_id or response_url}，最后响应：{last_payload}") + + result_response = await client.get(response_url, headers=headers) + result_response.raise_for_status() + return result_response.json() if result_response.text else {} + +async def generate_fal_provider_image(prompt, size, model, reference_images=None, provider=None): + provider = provider or get_api_provider_exact("fal") + body = fal_submit_payload(prompt, size, model, reference_images) + async with httpx.AsyncClient(timeout=httpx.Timeout(connect=20.0, read=1800.0, write=120.0, pool=20.0)) as client: + response = await client.post(fal_queue_url(provider, model), headers=fal_headers(provider), json=body) + response.raise_for_status() + submit_raw = response.json() if response.text else {} + raw = await wait_for_fal_result(client, provider, model, submit_raw) + if isinstance(raw, dict): + raw.setdefault("_fal_request", submit_raw) + return extract_image(raw), raw + async def generate_ai_image(prompt, size, quality, model, reference_images=None, provider_id="comfly"): provider = get_api_provider(provider_id) if provider["id"] == "modelscope": @@ -8637,6 +8806,8 @@ async def generate_ai_image(prompt, size, quality, model, reference_images=None, return await generate_gemini_provider_image(prompt, size, model, reference_images, provider) if is_volcengine_provider(provider): return await generate_volcengine_provider_image(prompt, size, model, reference_images, provider) + if is_fal_provider(provider): + return await generate_fal_provider_image(prompt, size, model, reference_images, provider) is_gpt2 = is_gpt_image_2_model(model) is_apimart = is_apimart_provider(provider) # 不对 GPT 尺寸做任何缩小/拦截：用户选什么尺寸就原样发给上游； @@ -10368,6 +10539,8 @@ async def save_providers(payload: List[ApiProviderPayload]): provider["protocol"] = "runninghub" if provider["id"] == "volcengine": provider["protocol"] = "volcengine" + if provider["id"] == "fal": + provider["protocol"] = "fal" if not providers: raise HTTPException(status_code=400, detail="至少保留一个 API 平台") # 强制最多一个 primary（取最后被标记的；都没标记则保持原样不强制） @@ -10416,9 +10589,13 @@ def protocol_from_payload(payload): return "runninghub" if provider_id == "jimeng": return "jimeng" + if provider_id == "fal": + return "fal" base_url = str(getattr(payload, "base_url", "") or "").strip().lower() if "runninghub.cn" in base_url or "runninghub.ai" in base_url: return "runninghub" + if "fal.ai" in base_url or "fal.run" in base_url: + return "fal" protocol = str(getattr(payload, "protocol", "") or "openai").strip().lower() return protocol if protocol in SUPPORTED_PROVIDER_PROTOCOLS else "openai" @@ -10438,6 +10615,8 @@ def api_key_from_payload(payload, protocol: str = ""): return value if protocol == "volcengine": return volcengine_provider_api_key("") + if protocol == "fal": + return provider_env_key_value("fal") return "" def upstream_models_url(base_url: str, protocol: str): @@ -10452,10 +10631,87 @@ def upstream_models_url(base_url: str, protocol: str): def upstream_model_headers(api_key: str, protocol: str): if protocol == "gemini": return {"x-goog-api-key": api_key, "Accept": "application/json"} + if protocol == "fal": + return {"Authorization": f"Key {strip_auth_scheme(api_key, 'Key')}", "Accept": "application/json"} if protocol == "runninghub": return {"Authorization": bearer_auth_value(api_key), "Accept": "application/json"} return {"Authorization": bearer_auth_value(api_key), "Accept": "application/json"} +def fal_model_id(item): + if isinstance(item, str): + return item.strip() + if not isinstance(item, dict): + return "" + return str( + item.get("id") + or item.get("endpoint_id") + or item.get("endpointId") + or item.get("model_id") + or item.get("modelId") + or item.get("slug") + or item.get("name") + or "" + ).strip() + +def fal_is_image_model(item): + if not isinstance(item, dict): + return True + text = " ".join( + str(item.get(key) or "") + for key in ("category", "type", "tags", "description", "id", "endpoint_id", "name") + ).lower() + if any(word in text for word in ("video", "audio", "speech", "voice", "music", "llm", "chat", "embedding", "3d")): + return False + return any(word in text for word in ("image", "flux", "stable-diffusion", "sdxl", "text-to-image", "image-to-image", "nano-banana")) + +async def fetch_fal_models_payload(api_key: str): + if not api_key: + raise HTTPException(status_code=400, detail="请先填写或保存 fal.ai API Key") + headers = upstream_model_headers(api_key, "fal") + categories = ["text-to-image", "image-to-image", "image-editing"] + ids = [] + raw_items = [] + async with httpx.AsyncClient(timeout=30) as client: + for category in categories: + cursor = "" + for _page in range(5): + params = {"category": category, "limit": 100} + if cursor: + params["cursor"] = cursor + response = await client.get(FAL_MODELS_API_URL, headers=headers, params=params) + if response.status_code >= 400 and category != categories[0]: + break + response.raise_for_status() + raw = response.json() if response.text else {} + items = [] + if isinstance(raw, dict): + for key in ("data", "models", "items", "results"): + if isinstance(raw.get(key), list): + items = raw[key] + break + cursor = str(raw.get("next_cursor") or raw.get("nextCursor") or raw.get("cursor") or "").strip() + elif isinstance(raw, list): + items = raw + cursor = "" + raw_items.extend(items) + for item in items: + model_id = fal_model_id(item) + if model_id and fal_is_image_model(item) and model_id not in ids: + ids.append(model_id) + if not cursor: + break + if not ids: + ids = list(FAL_DEFAULT_IMAGE_MODELS) + return { + "total": len(ids), + "protocol": "fal", + "image_models": ids, + "chat_models": [], + "video_models": [], + "all": ids, + "raw": {"models": raw_items[:200]}, + } + def volcengine_default_model_payload(status=200, message="", raw=None): return { "ok": True, @@ -10557,16 +10813,21 @@ async def probe_openai_models_endpoint(client, base_url: str, api_key: str): return False, {"status": response.status_code, "message": f"OpenAI /v1/models 不可用 (HTTP {response.status_code})", "raw": raw} return False, {"status": response.status_code, "message": f"OpenAI /v1/models 服务端错误 {response.status_code}", "raw": raw} -async def probe_volcengine_auto_detect(client, base_url: str, api_key: str): +def is_likely_volcengine_base_url(base_url: str): + host = urllib.parse.urlparse(str(base_url or "").strip()).netloc.lower() + return any(key in host for key in ("volces.com", "volcengine", "byteplusapi.com")) + +async def probe_volcengine_auto_detect(client, base_url: str, api_key: str, force: bool = False): + likely_volcengine = force or is_likely_volcengine_base_url(base_url) task_ok, task_probe = await probe_volcengine_task_endpoint(client, base_url, api_key) - if task_ok: + if task_ok and likely_volcengine and task_probe.get("status") != 404: return True, { "status": task_probe.get("status") or 200, "message": "检测到方舟/Ark 任务协议", "raw": {"task_probe": task_probe.get("raw")}, } compat_ok, compat_probe = await probe_openai_compat_bearer_endpoint(client, base_url, api_key) - if compat_ok: + if compat_ok and likely_volcengine: return True, { "status": compat_probe.get("status") or 200, "message": "检测到方舟/Ark Bearer 鉴权入口（OpenAI 兼容透传）", @@ -10659,6 +10920,21 @@ async def test_provider_connection(payload: TestConnectionPayload): "protocol": "runninghub", "raw": payload_models.get("raw"), } + if protocol == "fal": + api_key = api_key_from_payload(payload, protocol) + payload_models = await fetch_fal_models_payload(api_key) + return { + "ok": True, + "status": 200, + "message": "fal.ai API 可用，已拉取图像生成模型。", + "model_count": payload_models["total"], + "image_models": payload_models["image_models"], + "chat_models": [], + "video_models": [], + "all": payload_models["all"], + "protocol": "fal", + "raw": payload_models.get("raw"), + } base_url = (payload.base_url or "").strip().rstrip("/") if not base_url: raise HTTPException(status_code=400, detail="请先填写请求地址") @@ -10682,7 +10958,7 @@ async def test_provider_connection(payload: TestConnectionPayload): return {"ok": False, "status": resp.status_code, "message": f"上游 {endpoint_label} 返回网页 HTML，请检查请求地址是否为 API Base URL"} if resp.status_code >= 400: if protocol == "volcengine": - detected, probe = await probe_volcengine_auto_detect(client, base_url, api_key) + detected, probe = await probe_volcengine_auto_detect(client, base_url, api_key, force=True) if detected: message = f"{probe.get('message') or '方舟任务接口可达'}；但 /api/v3/models 不可用。请按实际方舟控制台模型名称手动填写视频模型。" return volcengine_default_model_payload(status=probe.get("status") or resp.status_code, message=message, raw={"models_error": resp.text[:300], **(probe.get("raw") or {})}) @@ -10696,7 +10972,7 @@ async def test_provider_connection(payload: TestConnectionPayload): grouped, ids = parse_upstream_models(data, protocol) grouped, ids = apply_agnes_model_defaults(base_url, grouped, ids) if protocol == "volcengine" and not ids: - detected, probe = await probe_volcengine_auto_detect(client, base_url, api_key) + detected, probe = await probe_volcengine_auto_detect(client, base_url, api_key, force=True) if detected: return volcengine_default_model_payload(status=resp.status_code, raw=data) return { @@ -10713,7 +10989,7 @@ async def test_provider_connection(payload: TestConnectionPayload): if protocol == "volcengine": try: async with httpx.AsyncClient(timeout=15) as client: - detected, probe = await probe_volcengine_auto_detect(client, base_url, api_key) + detected, probe = await probe_volcengine_auto_detect(client, base_url, api_key, force=True) if detected: message = f"{probe.get('message') or '方舟任务接口可达'}；但模型列表请求失败。请按实际方舟控制台模型名称手动填写视频模型。" return volcengine_default_model_payload(status=probe.get("status") or 0, message=message, raw={"models_error": str(e)[:300], **(probe.get("raw") or {})}) @@ -10732,6 +11008,20 @@ async def probe_async_endpoint(payload: TestConnectionPayload): api_key = api_key_from_payload(payload, protocol) if not api_key: raise HTTPException(status_code=400, detail="请先填写或保存 API Key") + if protocol == "fal": + payload_models = await fetch_fal_models_payload(api_key) + return { + "ok": True, + "protocol": "fal", + "status_code": 200, + "message": "fal.ai API 可用，已拉取图像生成模型。", + "raw": payload_models.get("raw"), + "model_count": payload_models["total"], + "image_models": payload_models["image_models"], + "chat_models": [], + "video_models": [], + "all": payload_models["all"], + } if protocol == "volcengine": try: async with httpx.AsyncClient(timeout=15) as client: @@ -10863,6 +11153,8 @@ async def fetch_models_from_upstream(base_url: str, api_key: str, protocol: str if protocol == "runninghub": provider = {"id": "runninghub", "name": "RunningHub", "base_url": base_url or RUNNINGHUB_DEFAULT_BASE_URL, "protocol": "runninghub", "api_key": api_key} return await runninghub_models_payload(provider) + if protocol == "fal": + return await fetch_fal_models_payload(api_key) base_url = (base_url or "").strip().rstrip("/") if not base_url: raise HTTPException(status_code=400, detail="请先填写请求地址") @@ -10885,7 +11177,7 @@ async def fetch_models_from_upstream(base_url: str, api_key: str, protocol: str raise HTTPException(status_code=400, detail=f"上游 {endpoint_label} 返回网页 HTML，请检查请求地址是否为 API Base URL") if resp.status_code >= 400: if protocol == "volcengine": - detected, probe = await probe_volcengine_auto_detect(client, base_url, api_key) + detected, probe = await probe_volcengine_auto_detect(client, base_url, api_key, force=True) if detected: payload = volcengine_default_model_payload( status=probe.get("status") or resp.status_code, @@ -10926,7 +11218,7 @@ async def fetch_models_from_upstream(base_url: str, api_key: str, protocol: str if protocol == "volcengine": try: async with httpx.AsyncClient(timeout=15) as client: - detected, probe = await probe_volcengine_auto_detect(client, base_url, api_key) + detected, probe = await probe_volcengine_auto_detect(client, base_url, api_key, force=True) if detected: payload = volcengine_default_model_payload( status=probe.get("status") or 0, diff --git a/static/api-settings.html b/static/api-settings.html index 3c9b6d145..e316c06f1 100644 --- a/static/api-settings.html +++ b/static/api-settings.html @@ -15,12 +15,12 @@ } catch(e) {} })(); - - - - - - + + + + + +
@@ -228,6 +228,7 @@ +
@@ -471,6 +472,6 @@
- + diff --git a/static/js/api-settings.js b/static/js/api-settings.js index d8fd99058..82a287d09 100644 --- a/static/js/api-settings.js +++ b/static/js/api-settings.js @@ -66,6 +66,8 @@ const MS_BUILTIN_IMAGE_MODELS = [ ]; const MS_DEFAULT_BASE_URL = 'https://api-inference.modelscope.cn/v1'; const RH_DEFAULT_BASE_URL = 'https://www.runninghub.cn'; +const FAL_DEFAULT_BASE_URL = 'https://queue.fal.run'; +const FAL_DEFAULT_IMAGE_MODELS = ['fal-ai/flux/schnell', 'fal-ai/flux/dev', 'fal-ai/flux-pro', 'fal-ai/nano-banana']; const LINGJING_DEFAULT_BASE_URL = 'https://apistudio.vip'; const LINGJING_REGISTER_URL = 'https://apistudio.vip/register?aff=g1CT'; const EXAMPLE_BASE_URL = 'https://api.example.com/v1'; @@ -246,7 +248,7 @@ function deriveIdFromName(name, existingId){ function updateIdPreview(){ const item = provider(); if(!item) return; - const isBuiltin = item.id === 'comfly' || item.id === 'modelscope' || item.id === 'runninghub' || item.id === 'volcengine' || item.id === 'lingjing' || item.id === 'jimeng'; + const isBuiltin = item.id === 'comfly' || item.id === 'modelscope' || item.id === 'runninghub' || item.id === 'volcengine' || item.id === 'lingjing' || item.id === 'jimeng' || item.id === 'fal'; const idPreview = document.getElementById('idPreview'); if(!idPreview) return; if(isBuiltin){ @@ -267,7 +269,7 @@ function visibleProviders(){ function isFixedProvider(itemOrId){ const id = typeof itemOrId === 'string' ? itemOrId : itemOrId?.id; // 即梦 CLI 不再是固定平台：可删除、可排序，未添加则不存在。 - return id === 'modelscope' || id === 'runninghub' || id === 'volcengine' || id === 'lingjing'; + return id === 'modelscope' || id === 'runninghub' || id === 'volcengine' || id === 'lingjing' || id === 'fal'; } function unique(values){ const seen = new Set(); @@ -647,6 +649,12 @@ function applyProviderOnboardingDefaults(id){ item.protocol = 'jimeng'; item.image_models = unique([...(item.image_models || []).filter(model => !JIMENG_LEGACY_IMAGE_MODELS.has(String(model || '').trim())), ...JIMENG_DEFAULT_IMAGE_MODELS]); item.video_models = unique([...(item.video_models || []).filter(model => !JIMENG_LEGACY_VIDEO_MODELS.has(String(model || '').trim())), ...JIMENG_DEFAULT_VIDEO_MODELS]); + } else if(id === 'fal'){ + item.base_url = item.base_url || FAL_DEFAULT_BASE_URL; + item.protocol = 'fal'; + item.image_models = unique([...(item.image_models || []), ...FAL_DEFAULT_IMAGE_MODELS]); + item.chat_models = unique(item.chat_models || []); + item.video_models = []; } selectedId = item.id; renderEditor(); @@ -660,18 +668,18 @@ function syncEditor(){ const item = provider(); if(!item) return; const oldId = item.id; - const isBuiltin = item.id === 'comfly' || item.id === 'modelscope' || item.id === 'runninghub' || item.id === 'volcengine' || item.id === 'lingjing' || item.id === 'jimeng'; + const isBuiltin = item.id === 'comfly' || item.id === 'modelscope' || item.id === 'runninghub' || item.id === 'volcengine' || item.id === 'lingjing' || item.id === 'jimeng' || item.id === 'fal'; // 内置和自定义平台的 ID 都保持稳定；新建时若没有 ID 才生成一次。 const nextId = isBuiltin ? item.id : deriveIdFromName(nameInput.value, item.id); item.id = nextId; if(oldId !== item.id) selectedId = item.id; item.name = nameInput.value.trim() || item.id; - const selectedProtocol = item.id === 'modelscope' ? 'openai' : item.id === 'runninghub' ? 'runninghub' : item.id === 'volcengine' ? 'volcengine' : item.id === 'jimeng' ? 'jimeng' : (protocolInput?.value || 'openai'); + const selectedProtocol = item.id === 'modelscope' ? 'openai' : item.id === 'runninghub' ? 'runninghub' : item.id === 'volcengine' ? 'volcengine' : item.id === 'jimeng' ? 'jimeng' : item.id === 'fal' ? 'fal' : (protocolInput?.value || 'openai'); item.base_url = selectedProtocol === 'jimeng' ? '' : baseInput.value.trim(); // 固定平台不从协议下拉读取 item.protocol = selectedProtocol; item.image_request_mode = normalizeImageRequestMode( - item.id === 'modelscope' || item.id === 'runninghub' || item.id === 'volcengine' || item.id === 'jimeng' + item.id === 'modelscope' || item.id === 'runninghub' || item.id === 'volcengine' || item.id === 'jimeng' || item.id === 'fal' ? 'openai' : (imageRequestModeInput?.value || item.image_request_mode) ); @@ -703,9 +711,9 @@ function ensureRunningHubLists(item){ } function updateProtocolFromInput(){ const item = provider(); - if(!item || !protocolInput || item.id === 'modelscope' || item.id === 'runninghub' || item.id === 'volcengine' || item.id === 'jimeng') return; + if(!item || !protocolInput || item.id === 'modelscope' || item.id === 'runninghub' || item.id === 'volcengine' || item.id === 'jimeng' || item.id === 'fal') return; const value = String(protocolInput.value || 'openai').toLowerCase(); - item.protocol = ['openai', 'apimart', 'gemini', 'volcengine', 'runninghub', 'jimeng'].includes(value) ? value : 'openai'; + item.protocol = ['openai', 'apimart', 'gemini', 'volcengine', 'runninghub', 'jimeng', 'fal'].includes(value) ? value : 'openai'; if(item.protocol === 'jimeng') item.base_url = ''; document.body.classList.toggle('show-jimeng', item.protocol === 'jimeng'); clearVerifyResult(); @@ -2147,7 +2155,7 @@ async function saveRecommendedApi(index){ if(ok) setStatus(trf('api.recommendSaved', {name:api.name})); } function sortedProviders(){ - const order = ['modelscope', 'runninghub', 'volcengine', 'lingjing']; + const order = ['modelscope', 'runninghub', 'volcengine', 'lingjing', 'fal']; return visibleProviders().sort((a, b) => { const ai = order.indexOf(a.id); const bi = order.indexOf(b.id); @@ -2290,13 +2298,13 @@ function renderEditor(){ baseInput.placeholder = EXAMPLE_BASE_URL; baseInput.value = item.base_url || ''; if(protocolInput){ - protocolInput.value = item.id === 'runninghub' ? 'runninghub' : item.id === 'volcengine' ? 'volcengine' : item.id === 'jimeng' ? 'jimeng' : (item.protocol || 'openai'); + protocolInput.value = item.id === 'runninghub' ? 'runninghub' : item.id === 'volcengine' ? 'volcengine' : item.id === 'jimeng' ? 'jimeng' : item.id === 'fal' ? 'fal' : (item.protocol || 'openai'); protocolInput.disabled = FIXED_PROTOCOL_PROVIDER_IDS.has(item.id); protocolInput.title = protocolInput.disabled ? '内置平台使用固定协议' : ''; } if(imageRequestModeInput){ imageRequestModeInput.value = normalizeImageRequestMode(item.image_request_mode); - imageRequestModeInput.disabled = item.id === 'modelscope' || item.id === 'runninghub' || item.id === 'volcengine' || item.id === 'jimeng'; + imageRequestModeInput.disabled = item.id === 'modelscope' || item.id === 'runninghub' || item.id === 'volcengine' || item.id === 'jimeng' || item.id === 'fal'; } keyInput.value = ''; keyInput.placeholder = item.has_key ? `${tr('api.keepCurrentKey')} ${item.key_preview || ''}` : tr('api.enterKey'); @@ -2348,6 +2356,15 @@ function renderEditor(){ keyInput.placeholder = '即梦 CLI 使用本机 dreamina login，无需 API Key'; keyHint.textContent = '请先在终端安装 dreamina CLI，并执行 dreamina login'; } + if(item.id === 'fal'){ + item.base_url = item.base_url || FAL_DEFAULT_BASE_URL; + item.protocol = 'fal'; + item.image_models = unique([...(item.image_models || []), ...FAL_DEFAULT_IMAGE_MODELS]); + item.video_models = []; + baseInput.value = item.base_url; + keyInput.placeholder = item.has_key ? `保持当前 fal.ai API Key ${item.key_preview || ''}` : '输入 fal.ai API Key'; + keyHint.textContent = item.has_key ? `${tr('api.keySaved')}${item.key_env || 'API/.env'}` : '尚未保存 fal.ai API Key'; + } document.body.classList.toggle('show-ms', isModelScope); document.body.classList.toggle('show-runninghub', isRunningHub); document.body.classList.toggle('show-volcengine', isVolcengine); @@ -2566,7 +2583,7 @@ function applyDetectedImageRequestMode(mode){ function applyDetectedProtocol(protocol){ const item = provider(); const detected = String(protocol || '').toLowerCase(); - if(!item || !protocolInput || !['openai', 'apimart', 'gemini', 'volcengine', 'runninghub', 'jimeng'].includes(detected)) return false; + if(!item || !protocolInput || !['openai', 'apimart', 'gemini', 'volcengine', 'runninghub', 'jimeng', 'fal'].includes(detected)) return false; if(String(protocolInput.value || '').toLowerCase() === detected && String(item.protocol || '').toLowerCase() === detected) return false; protocolInput.value = detected; item.protocol = detected; @@ -2664,7 +2681,7 @@ async function probeAsync(){ const detectedProtocol = String(data.protocol || '').toLowerCase(); const isAsync = data.ok === true && detectedProtocol === 'apimart'; const isOpenAiCompat = data.ok === true && detectedProtocol === 'openai'; - const keepManualProtocol = ['gemini', 'volcengine', 'jimeng'].includes(currentProtocol); + const keepManualProtocol = ['gemini', 'volcengine', 'jimeng', 'fal'].includes(currentProtocol); if(protocolInput && !keepManualProtocol){ applyDetectedProtocol(detectedProtocol || (isAsync ? 'apimart' : 'openai')); } @@ -2691,7 +2708,7 @@ async function probeAsync(){ 
${escapeHtml(rawJson)}
`); } catch(e){ - const keepManualProtocol = ['gemini', 'volcengine', 'jimeng'].includes(String(protocolInput?.value || item.protocol || '').toLowerCase()); + const keepManualProtocol = ['gemini', 'volcengine', 'jimeng', 'fal'].includes(String(protocolInput?.value || item.protocol || '').toLowerCase()); if(protocolInput && !keepManualProtocol){ protocolInput.value = 'openai'; protocolInput.dispatchEvent(new Event('change')); } const suffix = keepManualProtocol ? '，已保留当前手动选择的协议' : '，协议已设为 OpenAI 兼容'; showVerifyResult(`
⚠ ${escapeHtml(e.message || String(e))}${suffix}
`); @@ -2945,7 +2962,7 @@ async function clearKeyOnly(){ const ok = await saveProviders(); if(ok) keyInput.value = ''; } -const FIXED_PROTOCOL_PROVIDER_IDS = new Set(['modelscope', 'volcengine', 'jimeng', 'runninghub']); +const FIXED_PROTOCOL_PROVIDER_IDS = new Set(['modelscope', 'volcengine', 'jimeng', 'runninghub', 'fal']); function providerSupportsModelProtocol(item){ return Boolean(item) && !FIXED_PROTOCOL_PROVIDER_IDS.has(item.id); } @@ -3209,14 +3226,22 @@ async function saveProviders(){ ? 'volcengine' : item.id === 'jimeng' ? 'jimeng' - : ['openai', 'apimart', 'gemini', 'volcengine', 'runninghub', 'jimeng'].includes(String(item.protocol || '').toLowerCase()) ? String(item.protocol).toLowerCase() : 'openai'; + : item.id === 'fal' + ? 'fal' + : ['openai', 'apimart', 'gemini', 'volcengine', 'runninghub', 'jimeng', 'fal'].includes(String(item.protocol || '').toLowerCase()) ? String(item.protocol).toLowerCase() : 'openai'; item.image_request_mode = normalizeImageRequestMode( - item.id === 'modelscope' || item.id === 'runninghub' || item.id === 'volcengine' || item.id === 'jimeng' + item.id === 'modelscope' || item.id === 'runninghub' || item.id === 'volcengine' || item.id === 'jimeng' || item.id === 'fal' ? 'openai' : item.image_request_mode ); if(item.id === 'jimeng') item.base_url = ''; if(item.id === 'jimeng') item.video_models = unique([...(item.video_models || []).filter(model => !JIMENG_LEGACY_VIDEO_MODELS.has(String(model || '').trim())), ...JIMENG_DEFAULT_VIDEO_MODELS]); + if(item.id === 'fal'){ + item.base_url = item.base_url || FAL_DEFAULT_BASE_URL; + item.protocol = 'fal'; + item.image_models = unique([...(item.image_models || []), ...FAL_DEFAULT_IMAGE_MODELS]); + item.video_models = []; + } if(item.id === 'runninghub'){ item.base_url = item.base_url || RH_DEFAULT_BASE_URL; item.image_models = unique(item.image_models || []); @@ -3252,7 +3277,7 @@ async function saveProviders(){ id:item.id, name:item.name, base_url:item.base_url, - protocol:(item.id === 'modelscope') ? 'openai' : item.id === 'runninghub' ? 'runninghub' : item.id === 'volcengine' ? 'volcengine' : item.id === 'jimeng' ? 'jimeng' : (item.protocol || 'openai'), + protocol:(item.id === 'modelscope') ? 'openai' : item.id === 'runninghub' ? 'runninghub' : item.id === 'volcengine' ? 'volcengine' : item.id === 'jimeng' ? 'jimeng' : item.id === 'fal' ? 'fal' : (item.protocol || 'openai'), image_request_mode:item.image_request_mode || 'openai', image_generation_endpoint:item.image_generation_endpoint || '', image_edit_endpoint:item.image_edit_endpoint || '', From b2f6bc57f9e06cb9cf49db26afa5e1ccb79ada93 Mon Sep 17 00:00:00 2001 From: Glfcai-qiu Date: Thu, 25 Jun 2026 11:02:52 +0800 Subject: [PATCH 2/2] =?UTF-8?q?fix:=20=E4=BC=98=E5=8C=96=20fal.ai=20?= =?UTF-8?q?=E4=B8=8D=E5=90=8C=E6=A8=A1=E5=9E=8B=E7=9A=84=E5=8F=82=E6=95=B0?= =?UTF-8?q?=E9=80=82=E9=85=8D?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- main.py | 58 ++++++++++++++++++++++++++++++++++++++++++++++++++------- 1 file changed, 51 insertions(+), 7 deletions(-) diff --git a/main.py b/main.py index 8ccf7c7c3..c5311cbd3 100644 --- a/main.py +++ b/main.py @@ -884,7 +884,7 @@ def merge_default_api_providers(providers): current["base_url"] = fal_default["base_url"] current["protocol"] = "fal" current["image_request_mode"] = normalize_image_request_mode(current.get("image_request_mode")) - current["image_models"] = model_list_from_values([*(current.get("image_models") or []), *(fal_default.get("image_models") or [])]) + current["image_models"] = model_list_from_values(current.get("image_models") or []) current["chat_models"] = model_list_from_values(current.get("chat_models") or []) current["video_models"] = [] for current in merged: @@ -8713,14 +8713,58 @@ def fal_image_size(size): return "portrait_16_9" return {"width": width, "height": height} -def fal_reference_url(ref): - return reference_to_data_url(ref, max_size=1536) +FAL_KREA_ASPECT_RATIOS = ( + ("1:1", 1 / 1), + ("4:3", 4 / 3), + ("3:2", 3 / 2), + ("16:9", 16 / 9), + ("2.35:1", 2.35 / 1), + ("4:5", 4 / 5), + ("2:3", 2 / 3), + ("9:16", 9 / 16), +) -def fal_submit_payload(prompt, size, model, reference_images=None): +def fal_krea_aspect_ratio(width, height): + try: + width = float(width) + height = float(height) + except (TypeError, ValueError): + return "1:1" + if width <= 0 or height <= 0: + return "1:1" + ratio = width / height + return min(FAL_KREA_ASPECT_RATIOS, key=lambda item: abs(math.log(ratio / item[1])))[0] + +def build_fal_model_input(model, prompt, size, quality=None): + model_id = selected_model(model, FAL_DEFAULT_IMAGE_MODELS[0]).strip().strip("/") + model_lc = model_id.lower() + width, height = parse_size_pair(size) body = { "prompt": prompt, "image_size": fal_image_size(size), } + if model_lc.startswith("krea/v2/"): + body = { + "prompt": prompt, + "aspect_ratio": fal_krea_aspect_ratio(width, height), + } + quality_value = str(quality or "").strip().lower() + if quality_value in {"low", "medium", "high"}: + body["creativity"] = quality_value + for key in ("image_size", "width", "height", "negative_prompt", "num_inference_steps", "guidance_scale"): + body.pop(key, None) + elif model_lc == "fal-ai/flux-2-max": + if width and height: + body["image_size"] = {"width": width, "height": height} + for key in ("aspect_ratio", "creativity", "negative_prompt", "num_inference_steps", "guidance_scale"): + body.pop(key, None) + return body + +def fal_reference_url(ref): + return reference_to_data_url(ref, max_size=1536) + +def fal_submit_payload(prompt, size, model, reference_images=None, quality=None): + body = build_fal_model_input(model, prompt, size, quality) refs = [fal_reference_url(ref) for ref in (reference_images or [])[:ONLINE_IMAGE_REFERENCE_MAX]] refs = [url for url in refs if url] if refs: @@ -8782,9 +8826,9 @@ async def wait_for_fal_result(client, provider, model, submit_raw): result_response.raise_for_status() return result_response.json() if result_response.text else {} -async def generate_fal_provider_image(prompt, size, model, reference_images=None, provider=None): +async def generate_fal_provider_image(prompt, size, model, reference_images=None, provider=None, quality=None): provider = provider or get_api_provider_exact("fal") - body = fal_submit_payload(prompt, size, model, reference_images) + body = fal_submit_payload(prompt, size, model, reference_images, quality) async with httpx.AsyncClient(timeout=httpx.Timeout(connect=20.0, read=1800.0, write=120.0, pool=20.0)) as client: response = await client.post(fal_queue_url(provider, model), headers=fal_headers(provider), json=body) response.raise_for_status() @@ -8807,7 +8851,7 @@ async def generate_ai_image(prompt, size, quality, model, reference_images=None, if is_volcengine_provider(provider): return await generate_volcengine_provider_image(prompt, size, model, reference_images, provider) if is_fal_provider(provider): - return await generate_fal_provider_image(prompt, size, model, reference_images, provider) + return await generate_fal_provider_image(prompt, size, model, reference_images, provider, quality) is_gpt2 = is_gpt_image_2_model(model) is_apimart = is_apimart_provider(provider) # 不对 GPT 尺寸做任何缩小/拦截：用户选什么尺寸就原样发给上游；