Add HuggingFaceMedia provider with Video Generation

Add Support for Video Response in UI
Improve Support for Audio Response in UI
Fix ModelNotSupported errors in HuggingSpace providers

g4f/Provider/PollinationsAI.py

@@ -14,7 +14,8 @@ from ..image import to_data_uri, is_data_an_audio, to_input_audio
 from ..errors import ModelNotFoundError
 from ..requests.raise_for_status import raise_for_status
 from ..requests.aiohttp import get_connector
-from ..providers.response import ImageResponse, ImagePreview, FinishReason, Usage, Audio, ToolCalls
+from ..image.copy_images import save_response_media
+from ..providers.response import FinishReason, Usage, ToolCalls
 from .. import debug
 
 DEFAULT_HEADERS = {
@@ -239,8 +240,9 @@ class PollinationsAI(AsyncGeneratorProvider, ProviderModelMixin):
         async with ClientSession(headers=DEFAULT_HEADERS, connector=get_connector(proxy=proxy)) as session:
             async with session.get(url, allow_redirects=True) as response:
                 await raise_for_status(response)
-                image_url = str(response.url)
-                yield ImageResponse(image_url, prompt)
+                async for chunk in save_response_media(response, prompt):
+                    yield chunk
+                    return
 
     @classmethod
     async def _generate_text(
@@ -305,10 +307,10 @@ class PollinationsAI(AsyncGeneratorProvider, ProviderModelMixin):
             })
             async with session.post(url, json=data) as response:
                 await raise_for_status(response)
-                if response.headers["content-type"] == "audio/mpeg":
-                    yield Audio(await response.read())
+                async for chunk in save_response_media(response, messages[-1]["content"]):
+                    yield chunk
                     return
-                elif response.headers["content-type"].startswith("text/plain"):
+                if response.headers["content-type"].startswith("text/plain"):
                     yield await response.text()
                     return
                 elif response.headers["content-type"].startswith("text/event-stream"):

g4f/Provider/__init__.py

@@ -9,7 +9,7 @@ from .deprecated       import *
 from .needs_auth       import *
 from .not_working      import *
 from .local            import *
-from .hf               import HuggingFace, HuggingChat, HuggingFaceAPI, HuggingFaceInference
+from .hf               import HuggingFace, HuggingChat, HuggingFaceAPI, HuggingFaceInference, HuggingFaceMedia
 from .hf_space         import *
 from .mini_max         import HailuoAI, MiniMax
 from .template         import OpenaiTemplate, BackendApi

g4f/Provider/hf/HuggingChat.py

@@ -24,7 +24,7 @@ from ...requests import get_args_from_nodriver, DEFAULT_HEADERS
 from ...requests.raise_for_status import raise_for_status
 from ...providers.response import JsonConversation, ImageResponse, Sources, TitleGeneration, Reasoning, RequestLogin
 from ...cookies import get_cookies
-from .models import default_model, fallback_models, image_models, model_aliases, llama_models
+from .models import default_model, default_vision_model, fallback_models, image_models, model_aliases
 from ... import debug
 
 class Conversation(JsonConversation):
@@ -41,6 +41,7 @@ class HuggingChat(AsyncAuthedProvider, ProviderModelMixin):
     supports_stream = True
     needs_auth = True
     default_model = default_model
+    default_vision_model = default_vision_model
     model_aliases = model_aliases
     image_models = image_models
     text_models = fallback_models
@@ -107,8 +108,8 @@ class HuggingChat(AsyncAuthedProvider, ProviderModelMixin):
     ) -> AsyncResult:
         if not has_curl_cffi:
             raise MissingRequirementsError('Install "curl_cffi" package | pip install -U curl_cffi')
-        if model == llama_models["name"]:
-            model = llama_models["text"] if media is None else llama_models["vision"]
+        if not model and media is not None:
+            model = cls.default_vision_model
         model = cls.get_model(model)
 
         session = Session(**auth_result.get_dict())

g4f/Provider/hf/HuggingFaceAPI.py

@@ -6,27 +6,30 @@ from ...providers.types import Messages
 from ...typing import MediaListType
 from ...requests import StreamSession, raise_for_status
 from ...errors import ModelNotSupportedError
-from ...providers.helper import get_last_user_message
 from ...providers.response import ProviderInfo
 from ..template.OpenaiTemplate import OpenaiTemplate
-from .models import model_aliases, vision_models, default_vision_model, llama_models, text_models
+from .models import model_aliases, vision_models, default_llama_model, default_vision_model, text_models
 from ... import debug
 
 class HuggingFaceAPI(OpenaiTemplate):
-    label = "HuggingFace (Inference API)"
+    label = "HuggingFace (Text Generation)"
     parent = "HuggingFace"
     url = "https://api-inference.huggingface.com"
     api_base = "https://api-inference.huggingface.co/v1"
     working = True
     needs_auth = True
 
-    default_model = default_vision_model
+    default_model = default_llama_model
     default_vision_model = default_vision_model
     vision_models = vision_models
     model_aliases = model_aliases
     fallback_models = text_models + vision_models
 
-    provider_mapping: dict[str, dict] = {}
+    provider_mapping: dict[str, dict] = {
+        "google/gemma-3-27b-it": {
+            "hf-inference/models/google/gemma-3-27b-it": {
+                "task": "conversational",
+                "providerId": "google/gemma-3-27b-it"}}}
 
     @classmethod
     def get_model(cls, model: str, **kwargs) -> str:
@@ -47,7 +50,9 @@ class HuggingFaceAPI(OpenaiTemplate):
                     if [
                         provider
                         for provider in model.get("inferenceProviderMapping")
-                        if provider.get("task") == "conversational"]]
+                        if provider.get("status") == "live" and provider.get("task") == "conversational"
+                    ]
+                ] + list(cls.provider_mapping.keys())
             else:
                 cls.models = cls.fallback_models
         return cls.models
@@ -78,11 +83,12 @@ class HuggingFaceAPI(OpenaiTemplate):
         media: MediaListType = None,
         **kwargs
     ):
-        if model == llama_models["name"]:
-            model = llama_models["text"] if media is None else llama_models["vision"]
-        if model in cls.model_aliases:
-            model = cls.model_aliases[model]
+        if not model and media is not None:
+            model = cls.default_vision_model
+        model = cls.get_model(model)
         provider_mapping = await cls.get_mapping(model, api_key)
+        if not provider_mapping:
+            raise ModelNotSupportedError(f"Model is not supported: {model} in: {cls.__name__}")
         for provider_key in provider_mapping:
             api_path = provider_key if provider_key == "novita" else f"{provider_key}/v1"
             api_base = f"https://router.huggingface.co/{api_path}"

g4f/Provider/hf/HuggingFaceInference.py

@@ -10,6 +10,7 @@ from ..base_provider import AsyncGeneratorProvider, ProviderModelMixin, format_p
 from ...errors import ModelNotSupportedError, ResponseError
 from ...requests import StreamSession, raise_for_status
 from ...providers.response import FinishReason, ImageResponse
+from ...image.copy_images import save_response_media
 from ..helper import format_image_prompt, get_last_user_message
 from .models import default_model, default_image_model, model_aliases, text_models, image_models, vision_models
 from ... import debug
@@ -176,11 +177,9 @@ class HuggingFaceInference(AsyncGeneratorProvider, ProviderModelMixin):
                     debug.log(f"Special token: {is_special}")
                     yield FinishReason("stop" if is_special else "length")
                 else:
-                    if response.headers["content-type"].startswith("image/"):
-                        base64_data = base64.b64encode(b"".join([chunk async for chunk in response.iter_content()]))
-                        url = f"data:{response.headers['content-type']};base64,{base64_data.decode()}"
-                        yield ImageResponse(url, inputs)
-                    else:
+                    async for chunk in save_response_media(response, prompt):
+                        yield chunk
+                        return
                     yield (await response.json())[0]["generated_text"].strip()
 
 def format_prompt_mistral(messages: Messages, do_continue: bool = False) -> str:

g4f/Provider/hf/HuggingFaceMedia.py

@@ -0,0 +1,175 @@
+from __future__ import annotations
+
+import random
+import requests
+
+from ...providers.types import Messages
+from ...requests import StreamSession, raise_for_status
+from ...errors import ModelNotSupportedError
+from ...providers.helper import format_image_prompt
+from ...providers.base_provider import AsyncGeneratorProvider, ProviderModelMixin
+from ...providers.response import ProviderInfo, ImageResponse, VideoResponse
+from ...image.copy_images import save_response_media
+from ... import debug
+
+class HuggingFaceMedia(AsyncGeneratorProvider, ProviderModelMixin):
+    label = "HuggingFace (Image / Video Generation)"
+    parent = "HuggingFace"
+    url = "https://huggingface.co"
+    working = True
+    needs_auth = True
+
+    tasks = ["text-to-image", "text-to-video"]
+    provider_mapping: dict[str, dict] = {}
+    task_mapping: dict[str, str] = {}
+
+    @classmethod
+    def get_models(cls, **kwargs) -> list[str]:
+        if not cls.models:
+            url = "https://huggingface.co/api/models?inference=warm&expand[]=inferenceProviderMapping"
+            response = requests.get(url)
+            if response.ok:
+                models = response.json()
+                cls.models = [
+                    model["id"]
+                    for model in models
+                    if [
+                        provider
+                        for provider in model.get("inferenceProviderMapping")
+                        if provider.get("status") == "live" and provider.get("task") in cls.tasks
+                    ]
+                ]
+                cls.task_mapping = {
+                    model["id"]: [
+                        provider.get("task")
+                        for provider in model.get("inferenceProviderMapping")
+                    ].pop()
+                    for model in models
+                }
+            else:
+                cls.models = []
+        return cls.models
+
+    @classmethod
+    async def get_mapping(cls, model: str, api_key: str = None):
+        if model in cls.provider_mapping:
+            return cls.provider_mapping[model]
+        headers = {
+            'Content-Type': 'application/json',
+        }
+        if api_key is not None:
+            headers["Authorization"] = f"Bearer {api_key}"
+        async with StreamSession(
+            timeout=30,
+            headers=headers,
+        ) as session:
+            async with session.get(f"https://huggingface.co/api/models/{model}?expand[]=inferenceProviderMapping") as response:
+                await raise_for_status(response)
+                model_data = await response.json()
+                cls.provider_mapping[model] = {key: value for key, value in model_data.get("inferenceProviderMapping").items() if value["status"] == "live"}
+        return cls.provider_mapping[model]
+
+    @classmethod
+    async def create_async_generator(
+        cls,
+        model: str,
+        messages: Messages,
+        api_key: str = None,
+        extra_data: dict = {},
+        prompt: str = None,
+        proxy: str = None,
+        timeout: int = 0,
+        **kwargs
+    ):
+        provider_mapping = await cls.get_mapping(model, api_key)
+        headers = {
+            'Accept-Encoding': 'gzip, deflate',
+            'Content-Type': 'application/json',
+        }
+        new_mapping = {
+            "hf-free" if key == "hf-inference" else key: value for key, value in provider_mapping.items()
+            if key in ["replicate", "together", "hf-inference"]
+        }
+        provider_mapping = {**new_mapping, **provider_mapping}
+        last_response = None
+        for provider_key, provider in provider_mapping.items():
+            yield ProviderInfo(**{**cls.get_dict(), "label": f"HuggingFace ({provider_key})", "url": f"{cls.url}/{model}"})
+
+            api_base = f"https://router.huggingface.co/{provider_key}"
+            task = provider["task"]
+            provider_id = provider["providerId"]
+            if task not in cls.tasks:
+                raise ModelNotSupportedError(f"Model is not supported: {model} in: {cls.__name__} task: {task}")
+
+            prompt = format_image_prompt(messages, prompt)
+            if task == "text-to-video":
+                extra_data = {
+                    "num_inference_steps": 20,
+                    "video_size": "landscape_16_9",
+                    **extra_data
+                }
+            else:
+                extra_data = {
+                    "width": 1024,
+                    "height": 1024,
+                    **extra_data
+                }
+            if provider_key == "fal-ai":
+                url = f"{api_base}/{provider_id}"
+                data = {
+                    "prompt": prompt,
+                    "image_size": "square_hd",
+                    **extra_data
+                }
+            elif provider_key == "replicate":
+                url = f"{api_base}/v1/models/{provider_id}/prediction"
+                data = {
+                    "input": {
+                        "prompt": prompt,
+                        **extra_data
+                    }
+                }
+            elif provider_key in ("hf-inference", "hf-free"):
+                api_base = "https://api-inference.huggingface.co"
+                url = f"{api_base}/models/{provider_id}"
+                data = {
+                    "inputs": prompt,
+                    "parameters": {
+                        "seed": random.randint(0, 2**32),
+                        **extra_data
+                    }
+                }
+            elif task == "text-to-image":
+                url = f"{api_base}/v1/images/generations"
+                data = {
+                    "response_format": "url",
+                    "prompt": prompt,
+                    "model": provider_id,
+                    **extra_data
+                }
+
+            async with StreamSession(
+                headers=headers if provider_key == "free" or api_key is None else {**headers, "Authorization": f"Bearer {api_key}"},
+                proxy=proxy,
+                timeout=timeout
+            ) as session:
+                async with session.post(url, json=data) as response:
+                    if response.status in (400, 401, 402):
+                        last_response = response
+                        debug.error(f"{cls.__name__}: Error {response.status} with {provider_key} and {provider_id}")
+                        continue
+                    if response.status == 404:
+                        raise ModelNotSupportedError(f"Model is not supported: {model}")
+                    await raise_for_status(response)
+                    async for chunk in save_response_media(response, prompt):
+                        yield chunk
+                        return
+                    result = await response.json()
+                    if "video" in result:
+                        yield VideoResponse(result["video"]["url"], prompt)
+                    elif task == "text-to-image":
+                        yield ImageResponse([item["url"] for item in result.get("images", result.get("data"))], prompt)
+                    elif task == "text-to-video":
+                        yield VideoResponse(result["output"], prompt)
+                    return
+        await raise_for_status(last_response)

g4f/Provider/hf/__init__.py

@@ -9,6 +9,7 @@ from ..base_provider import AsyncGeneratorProvider, ProviderModelMixin
 from .HuggingChat import HuggingChat
 from .HuggingFaceAPI import HuggingFaceAPI
 from .HuggingFaceInference import HuggingFaceInference
+from .HuggingFaceMedia import HuggingFaceMedia
 from .models import model_aliases, vision_models, default_vision_model
 from ... import debug
 
@@ -51,6 +52,12 @@ class HuggingFace(AsyncGeneratorProvider, ProviderModelMixin):
                 debug.error(f"{cls.__name__} {type(e).__name__}; {e}")
         if not cls.image_models:
             cls.get_models()
+        try:
+            async for chunk in HuggingFaceMedia.create_async_generator(model, messages, **kwargs):
+                yield chunk
+            return
+        except ModelNotSupportedError:
+            pass
         if model in cls.image_models:
             if "api_key" not in kwargs:
                 async for chunk in HuggingChat.create_async_generator(model, messages, **kwargs):

g4f/Provider/hf/models.py

@@ -47,9 +47,5 @@ extra_models = [
     "NousResearch/Hermes-3-Llama-3.1-8B",
 ]
 default_vision_model = "meta-llama/Llama-3.2-11B-Vision-Instruct"
+default_llama_model = "meta-llama/Llama-3.3-70B-Instruct"
 vision_models = [default_vision_model, "Qwen/Qwen2-VL-7B-Instruct"]
-llama_models = {
-    "name": "llama-3",
-    "text": "meta-llama/Llama-3.3-70B-Instruct",
-    "vision": "meta-llama/Llama-3.2-11B-Vision-Instruct",
-}

g4f/Provider/hf_space/BlackForestLabs_Flux1Dev.py

@@ -67,7 +67,6 @@ class BlackForestLabs_Flux1Dev(AsyncGeneratorProvider, ProviderModelMixin):
         zerogpu_uuid: str = "[object Object]",
         **kwargs
     ) -> AsyncResult:
-        model = cls.get_model(model)
         async with StreamSession(impersonate="chrome", proxy=proxy) as session:
             prompt = format_image_prompt(messages, prompt)
             data = [prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps]

g4f/Provider/hf_space/BlackForestLabs_Flux1Schnell.py

@@ -37,8 +37,6 @@ class BlackForestLabs_Flux1Schnell(AsyncGeneratorProvider, ProviderModelMixin):
         randomize_seed: bool = True,
         **kwargs
     ) -> AsyncResult:
-
-        model = cls.get_model(model)
         width = max(32, width - (width % 8))
         height = max(32, height - (height % 8))
         prompt = format_image_prompt(messages, prompt)

g4f/Provider/hf_space/CohereForAI_C4AI_Command.py

@@ -24,9 +24,14 @@ class CohereForAI_C4AI_Command(AsyncGeneratorProvider, ProviderModelMixin):
         "command-r": "command-r",
         "command-r7b": "command-r7b-12-2024",
     }
-    
     models = list(model_aliases.keys())
 
+    @classmethod
+    def get_model(cls, model: str, **kwargs) -> str:
+        if model in cls.model_aliases.values():
+            return model
+        return super().get_model(model, **kwargs)
+
     @classmethod
     async def create_async_generator(
         cls, model: str, messages: Messages,

g4f/Provider/needs_auth/OpenaiChat.py

@@ -203,7 +203,21 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
         Returns:
             A list of messages with the user input and the image, if any
         """
-        # Create a message object with the user role and the content
+        # merged_messages = []
+        # last_message = None
+        # for message in messages:
+        #     current_message = last_message
+        #     if current_message is not None:
+        #         if current_message["role"] == message["role"]:
+        #             current_message["content"] += "\n" + message["content"]
+        #         else:
+        #             merged_messages.append(current_message)
+        #             last_message = message.copy()
+        #     else:
+        #         last_message = message.copy()
+        # if last_message is not None:
+        #     merged_messages.append(last_message)
+
         messages = [{
             "id": str(uuid.uuid4()),
             "author": {"role": message["role"]},

g4f/api/__init__.py

@@ -39,7 +39,7 @@ from g4f.client import AsyncClient, ChatCompletion, ImagesResponse, convert_to_p
 from g4f.providers.response import BaseConversation, JsonConversation
 from g4f.client.helper import filter_none
 from g4f.image import is_data_an_media
-from g4f.image.copy_images import images_dir, copy_images, get_source_url
+from g4f.image.copy_images import images_dir, copy_media, get_source_url
 from g4f.errors import ProviderNotFoundError, ModelNotFoundError, MissingAuthError, NoValidHarFileError
 from g4f.cookies import read_cookie_files, get_cookies_dir
 from g4f.Provider import ProviderType, ProviderUtils, __providers__
@@ -594,10 +594,10 @@ class Api:
                         ssl = False
                 if source_url is not None:
                     try:
-                        await copy_images(
+                        await copy_media(
                             [source_url],
                             target=target, ssl=ssl)
-                        debug.log(f"Image copied from {source_url}")
+                        debug.log(f"File copied from {source_url}")
                     except Exception as e:
                         debug.error(f"Download failed:  {source_url}\n{type(e).__name__}: {e}")
                         return RedirectResponse(url=source_url)

g4f/client/__init__.py

@@ -9,7 +9,7 @@ import aiohttp
 import base64
 from typing import Union, AsyncIterator, Iterator, Awaitable, Optional
 
-from ..image.copy_images import copy_images
+from ..image.copy_images import copy_media
 from ..typing import Messages, ImageType
 from ..providers.types import ProviderType, BaseRetryProvider
 from ..providers.response import *
@@ -532,7 +532,7 @@ class Images:
             images = await asyncio.gather(*[get_b64_from_url(image) for image in response.get_list()])
         else:
             # Save locally for None (default) case
-            images = await copy_images(response.get_list(), response.get("cookies"), proxy)
+            images = await copy_media(response.get_list(), response.get("cookies"), proxy)
             images = [Image.model_construct(url=f"/images/{os.path.basename(image)}", revised_prompt=response.alt) for image in images]
         
         return ImagesResponse.model_construct(

g4f/gui/client/qrcode.html

@@ -15,7 +15,7 @@
 </head>
 <body>
 
-    <h1>QR Scanner & QR Code Generator</h1>
+    <h1>QR Scanner & QR Code</h1>
     
     <h2>QR Code Scanner</h2>
     <video id="video"></video>
@@ -25,9 +25,8 @@
     <button id="toggleFlash">Toggle Flash</button>
     <p id="cam-status"></p>
 
-    <h2>Generate QR Code</h2>
+    <h2>QR Code</h2>
     <div id="qrcode"></div>
-    
     <button id="generateQRCode">Generate QR Code</button>
 
     <script type="module">

g4f/gui/client/static/js/chat.v1.js

@@ -80,7 +80,11 @@ if (window.markdownit) {
             .replaceAll('<code>', '<code class="language-plaintext">')
             .replaceAll('&lt;i class=&quot;', '<i class="')
             .replaceAll('&quot;&gt;&lt;/i&gt;', '"></i>')
-            .replaceAll('&lt;iframe type=&quot;text/html&quot; src=&quot;', '<iframe type="text/html" frameborder="0" allow="fullscreen" src="')
+            .replaceAll('&lt;video controls src=&quot;', '<video controls width="400" src="')
+            .replaceAll('&quot;&gt;&lt;/video&gt;', '"></video>')
+            .replaceAll('&lt;audio controls src=&quot;', '<audio controls src="')
+            .replaceAll('&quot;&gt;&lt;/audio&gt;', '"></audio>')
+            .replaceAll('&lt;iframe type=&quot;text/html&quot; src=&quot;', '<iframe type="text/html" frameborder="0" allow="fullscreen" height="390" width="640" src="')
             .replaceAll('&quot;&gt;&lt;/iframe&gt;', `?enablejsapi=1&origin=${new URL(location.href).origin}"></iframe>`)
     }
 }
@@ -229,7 +233,7 @@ function register_message_images() {
                     let seed = Math.floor(Date.now() / 1000);
                     newPath = `https://image.pollinations.ai/prompt/${newPath}?seed=${seed}&nologo=true`;
                     let downloadUrl = newPath;
-                    if (document.getElementById("download_images")?.checked) {
+                    if (document.getElementById("download_media")?.checked) {
                         downloadUrl = `/images/${filename}?url=${escapeHtml(newPath)}`;
                     }
                     const link = document.createElement("a");
@@ -862,12 +866,6 @@ async function add_message_chunk(message, message_id, provider, scroll, finish_m
             content_map.inner.innerHTML = markdown_render(message.preview);
             await register_message_images();
         }
-    } else if (message.type == "audio") {
-        audio = new Audio(message.audio);
-        audio.controls = true;   
-        content_map.inner.appendChild(audio);
-        audio.play();
-        reloadConversation = false;
     } else if (message.type == "content") {
         message_storage[message_id] += message.content;
         update_message(content_map, message_id, null, scroll);
@@ -1090,7 +1088,7 @@ const ask_gpt = async (message_id, message_index = -1, regenerate = false, provi
         } else {
             api_key = get_api_key_by_provider(provider);
         }
-        const download_images = document.getElementById("download_images")?.checked;
+        const download_media = document.getElementById("download_media")?.checked;
         let api_base;
         if (provider == "Custom") {
             api_base = document.getElementById("api_base")?.value;
@@ -1119,7 +1117,7 @@ const ask_gpt = async (message_id, message_index = -1, regenerate = false, provi
             provider: provider,
             messages: messages,
             action: action,
-            download_images: download_images,
+            download_media: download_media,
             api_key: api_key,
             api_base: api_base,
             ignored: ignored,
@@ -2723,6 +2721,9 @@ async function load_provider_models(provider=null) {
             option.value = model.model;
             option.dataset.label = model.model;
             option.text = `${model.model}${model.image ? " (Image Generation)" : ""}${model.vision ? " (Image Upload)" : ""}`;
+            if (model.task) {
+                option.text += ` (${model.task})`;
+            }
             modelProvider.appendChild(option);
             if (model.default) {
                 defaultIndex = i;

g4f/gui/server/api.py

@@ -8,7 +8,7 @@ from flask import send_from_directory
 from inspect import signature
 
 from ...errors import VersionNotFoundError
-from ...image.copy_images import copy_images, ensure_images_dir, images_dir
+from ...image.copy_images import copy_media, ensure_images_dir, images_dir
 from ...tools.run_tools import iter_run_tools
 from ...Provider import ProviderUtils, __providers__
 from ...providers.base_provider import ProviderModelMixin
@@ -53,6 +53,7 @@ class Api:
                         "default": model == provider.default_model,
                         "vision": getattr(provider, "default_vision_model", None) == model or model in getattr(provider, "vision_models", []),
                         "image": False if provider.image_models is None else model in provider.image_models,
+                        "task": None if not hasattr(provider, "task_mapping") else provider.task_mapping[model] if model in provider.task_mapping else None
                     }
                     for model in models
                 ]
@@ -127,7 +128,7 @@ class Api:
             **kwargs
         }
 
-    def _create_response_stream(self, kwargs: dict, conversation_id: str, provider: str, download_images: bool = True) -> Iterator:
+    def _create_response_stream(self, kwargs: dict, conversation_id: str, provider: str, download_media: bool = True) -> Iterator:
         def decorated_log(text: str, file = None):
             debug.logs.append(text)
             if debug.logging:
@@ -154,7 +155,7 @@ class Api:
             if hasattr(provider_handler, "get_parameters"):
                 yield self._format_json("parameters", provider_handler.get_parameters(as_json=True))
         try:
-            result = iter_run_tools(ChatCompletion.create, **{**kwargs, "model": model, "provider": provider_handler})
+            result = iter_run_tools(ChatCompletion.create, **{**kwargs, "model": model, "provider": provider_handler, "download_media": download_media})
             for chunk in result:
                 if isinstance(chunk, ProviderInfo):
                     yield self.handle_provider(chunk, model)
@@ -182,13 +183,13 @@ class Api:
                     yield self._format_json("preview", chunk.to_string())
                 elif isinstance(chunk, ImagePreview):
                     yield self._format_json("preview", chunk.to_string(), images=chunk.images, alt=chunk.alt)
-                elif isinstance(chunk, ImageResponse):
-                    images = chunk
-                    if download_images or chunk.get("cookies"):
+                elif isinstance(chunk, (ImageResponse, VideoResponse)):
+                    media = chunk
+                    if download_media or chunk.get("cookies"):
                         chunk.alt = format_image_prompt(kwargs.get("messages"), chunk.alt)
-                        images = asyncio.run(copy_images(chunk.get_list(), chunk.get("cookies"), chunk.get("headers"), proxy=proxy, alt=chunk.alt))
-                        images = ImageResponse(images, chunk.alt)
-                    yield self._format_json("content", str(images), images=chunk.get_list(), alt=chunk.alt)
+                        media = asyncio.run(copy_media(chunk.get_list(), chunk.get("cookies"), chunk.get("headers"), proxy=proxy, alt=chunk.alt))
+                        media = ImageResponse(media, chunk.alt) if isinstance(chunk, ImageResponse) else VideoResponse(media, chunk.alt)
+                    yield self._format_json("content", str(media), images=chunk.get_list(), alt=chunk.alt)
                 elif isinstance(chunk, SynthesizeData):
                     yield self._format_json("synthesize", chunk.get_dict())
                 elif isinstance(chunk, TitleGeneration):
@@ -205,8 +206,8 @@ class Api:
                     yield self._format_json("reasoning", **chunk.get_dict())
                 elif isinstance(chunk, YouTube):
                     yield self._format_json("content", chunk.to_string())
-                elif isinstance(chunk, Audio):
-                    yield self._format_json("audio", str(chunk))
+                elif isinstance(chunk, AudioResponse):
+                    yield self._format_json("content", str(chunk))
                 elif isinstance(chunk, DebugResponse):
                     yield self._format_json("log", chunk.log)
                 elif isinstance(chunk, RawResponse):

g4f/gui/server/backend_api.py

@@ -14,9 +14,8 @@ from typing import Generator
 from pathlib import Path
 from urllib.parse import quote_plus
 from hashlib import sha256
-from werkzeug.utils import secure_filename
 
-from ...image import is_allowed_extension, to_image
+from ...image import is_allowed_extension
 from ...client.service import convert_to_provider
 from ...providers.asyncio import to_sync_generator
 from ...client.helper import filter_markdown
@@ -25,6 +24,7 @@ from ...tools.run_tools import iter_run_tools
 from ...errors import ProviderNotFoundError
 from ...image import is_allowed_extension
 from ...cookies import get_cookies_dir
+from ...image.copy_images import secure_filename
 from ... import ChatCompletion
 from ... import models
 from .api import Api
@@ -130,16 +130,14 @@ class Backend_Api(Api):
                 if model != "default" and model in models.demo_models:
                     json_data["provider"] = random.choice(models.demo_models[model][1])
                 else:
-                    if not model or model == "default":
-                        json_data["model"] = models.demo_models["default"][0].name
-                    json_data["provider"] = random.choice(models.demo_models["default"][1])
+                    json_data["provider"] = models.HuggingFace
             kwargs = self._prepare_conversation_kwargs(json_data)
             return self.app.response_class(
                 self._create_response_stream(
                     kwargs,
                     json_data.get("conversation_id"),
                     json_data.get("provider"),
-                    json_data.get("download_images", True),
+                    json_data.get("download_media", True),
                 ),
                 mimetype='text/event-stream'
             )
@@ -331,21 +329,26 @@ class Backend_Api(Api):
                 [f.write(f"{filename}\n") for filename in filenames]
             return {"bucket_id": bucket_id, "files": filenames, "media": media}
 
-        @app.route('/backend-api/v2/files/<bucket_id>/media/<filename>', methods=['GET'])
-        def get_media(bucket_id, filename):
-            bucket_id = secure_filename(bucket_id)
-            bucket_dir = get_bucket_dir(bucket_id)
+        @app.route('/files/<bucket_id>/media/<filename>', methods=['GET'])
+        def get_media(bucket_id, filename, dirname: str = None):
+            bucket_dir = get_bucket_dir(secure_filename(bucket_id), secure_filename(dirname))
             media_dir = os.path.join(bucket_dir, "media")
             if os.path.exists(media_dir):
                 return send_from_directory(os.path.abspath(media_dir), filename)
-            return "File not found", 404
+            return "Not found", 404
+
+        @app.route('/files/<dirname>/<bucket_id>/media/<filename>', methods=['GET'])
+        def get_media_sub(dirname, bucket_id, filename):
+            return get_media(bucket_id, filename, dirname)
 
         @app.route('/backend-api/v2/files/<bucket_id>/<filename>', methods=['PUT'])
-        def upload_file(bucket_id, filename):
-            bucket_id = secure_filename(bucket_id)
-            bucket_dir = get_bucket_dir(bucket_id)
+        def upload_file(bucket_id, filename, dirname: str = None):
+            bucket_dir = secure_filename(bucket_id if dirname is None else dirname)
+            bucket_dir = get_bucket_dir(bucket_dir)
             filename = secure_filename(filename)
             bucket_path = Path(bucket_dir)
+            if dirname is not None:
+                bucket_path = bucket_path / secure_filename(bucket_id)
 
             if not supports_filename(filename):
                 return jsonify({"error": {"message": f"File type not allowed"}}), 400
@@ -366,6 +369,10 @@ class Backend_Api(Api):
             except Exception as e:
                 return jsonify({"error": {"message": f"Error uploading file: {str(e)}"}}), 500
     
+        @app.route('/backend-api/v2/files/<bucket_id>/<dirname>/<filename>', methods=['PUT'])
+        def upload_file_sub(bucket_id, filename, dirname):
+            return upload_file(bucket_id, filename, dirname)
+
         @app.route('/backend-api/v2/upload_cookies', methods=['POST'])
         def upload_cookies():
             file = None

g4f/image/copy_images.py

@@ -10,7 +10,10 @@ from urllib.parse import quote, unquote
 from aiohttp import ClientSession, ClientError
 
 from ..typing import Optional, Cookies
-from ..requests.aiohttp import get_connector
+from ..requests.aiohttp import get_connector, StreamResponse
+from ..image import EXTENSIONS_MAP
+from ..tools.files import get_bucket_dir
+from ..providers.response import ImageResponse, AudioResponse, VideoResponse
 from ..Provider.template import BackendApi
 from . import is_accepted_format, extract_data_uri
 from .. import debug
@@ -18,10 +21,10 @@ from .. import debug
 # Directory for storing generated images
 images_dir = "./generated_images"
 
-def get_image_extension(image: str) -> str:
+def get_media_extension(image: str) -> str:
     """Extract image extension from URL or filename, default to .jpg"""
-    match = re.search(r"\.(jpe?g|png|webp)$", image, re.IGNORECASE)
-    return f".{match.group(1).lower()}" if match else ".jpg"
+    match = re.search(r"\.(jpe?g|png|webp|mp4|mp3|wav)[?$]", image, re.IGNORECASE)
+    return f".{match.group(1).lower()}" if match else ""
 
 def ensure_images_dir():
     """Create images directory if it doesn't exist"""
@@ -35,7 +38,42 @@ def get_source_url(image: str, default: str = None) -> str:
             return decoded_url
     return default
 
-async def copy_images(
+def secure_filename(filename: str) -> str:
+    # Keep letters, numbers, basic punctuation and all Unicode chars
+    filename = re.sub(
+        r'[^\w.,_-]+',
+        '_', 
+        unquote(filename).strip(), 
+        flags=re.UNICODE
+    )
+    filename = filename[:100].strip(".,_-")
+    return filename
+
+async def save_response_media(response: StreamResponse, prompt: str):
+    content_type = response.headers["content-type"]
+    if content_type in EXTENSIONS_MAP or content_type.startswith("audio/"):
+        extension = EXTENSIONS_MAP[content_type] if content_type in EXTENSIONS_MAP else content_type[6:].replace("mpeg", "mp3")
+        bucket_id = str(uuid.uuid4())
+        dirname = str(int(time.time()))
+        bucket_dir = get_bucket_dir(bucket_id, dirname)
+        media_dir = os.path.join(bucket_dir, "media")
+        os.makedirs(media_dir, exist_ok=True)
+        filename = secure_filename(f"{content_type[0:5] if prompt is None else prompt}.{extension}")
+        newfile = os.path.join(media_dir, filename)
+        with open(newfile, 'wb') as f:
+            async for chunk in response.iter_content() if hasattr(response, "iter_content") else response.content.iter_any():
+                f.write(chunk)
+        media_url = f"/files/{dirname}/{bucket_id}/media/{filename}"
+        if response.method == "GET":
+            media_url = f"{media_url}?url={str(response.url)}"
+        if content_type.startswith("audio/"):
+            yield AudioResponse(media_url)
+        elif content_type.startswith("video/"):
+            yield VideoResponse(media_url, prompt)
+        else:
+            yield ImageResponse(media_url, prompt)
+
+async def copy_media(
     images: list[str],
     cookies: Optional[Cookies] = None,
     headers: Optional[dict] = None,
@@ -60,33 +98,18 @@ async def copy_images(
     ) as session:
         async def copy_image(image: str, target: str = None) -> str:
             """Process individual image and return its local URL"""
+            # Skip if image is already local
+            if image.startswith("/"):
+                return image
             target_path = target
             if target_path is None:
-                # Generate filename components
-                file_hash = hashlib.sha256(image.encode()).hexdigest()[:16]
-                timestamp = int(time.time())
-                
-                # Sanitize alt text for filename (Unicode-safe)
-                if alt:
-                    # Keep letters, numbers, basic punctuation and all Unicode chars
-                    clean_alt = re.sub(
-                        r'[^\w\s.-]',  # Allow all Unicode word chars
-                        '_', 
-                        unquote(alt).strip(), 
-                        flags=re.UNICODE
-                    )
-                    clean_alt = re.sub(r'[\s_]+', '_', clean_alt)[:100]
-                else:
-                    clean_alt = "image"
-
                 # Build safe filename with full Unicode support
-                extension = get_image_extension(image)
-                filename = (
-                    f"{timestamp}_"
-                    f"{clean_alt}_"
-                    f"{file_hash}"
-                    f"{extension}"
-                )
+                filename = secure_filename("".join((
+                    f"{int(time.time())}_",
+                    (f"{alt}_" if alt else ""),
+                    f"{hashlib.sha256(image.encode()).hexdigest()[:16]}",
+                    f"{get_media_extension(image)}"
+                )))
                 target_path = os.path.join(images_dir, filename)
             try:
                 # Handle different image types

g4f/models.py

@@ -234,11 +234,6 @@ llama_3_1_405b = Model(
 )
 
 # llama 3.2
-llama_3 = VisionModel(
-    name          = "llama-3",
-    base_provider = "Meta Llama",
-    best_provider = IterListProvider([HuggingChat, HuggingFace])
-)
 
 llama_3_2_1b = Model(
     name          = "llama-3.2-1b",
@@ -977,7 +972,6 @@ class ModelUtils:
 
 
 demo_models = {
-    "default": [llama_3, [HuggingFace]],
     llama_3_2_11b.name: [llama_3_2_11b, [HuggingChat]],
     qwen_2_vl_7b.name: [qwen_2_vl_7b, [HuggingFaceAPI]],
     deepseek_r1.name: [deepseek_r1, [HuggingFace, PollinationsAI]],

g4f/requests/raise_for_status.py

@@ -25,15 +25,12 @@ async def raise_for_status_async(response: Union[StreamResponse, ClientResponse]
         return
     if message is None:
         content_type = response.headers.get("content-type", "")
-        # if content_type.startswith("application/json"):
-        #     try:
-        #         data = await response.json()
-        #         message = data.get("error")
-        #         if isinstance(message, dict):
-        #                 message = data.get("message")
-        #     except Exception:
-        #         pass
-        # else:
+        if content_type.startswith("application/json"):
+            message = await response.json()
+            message = message.get("error", message)
+            if isinstance(message, dict):
+                message = message.get("message", message)
+        else:
             text = (await response.text()).strip()
             is_html = content_type.startswith("text/html") or text.startswith("<!DOCTYPE")
             message = "HTML content" if is_html else text
@@ -47,7 +44,9 @@ async def raise_for_status_async(response: Union[StreamResponse, ClientResponse]
     elif response.status == 403 and is_openai(text):
         raise ResponseStatusError(f"Response {response.status}: OpenAI Bot detected")
     elif response.status == 502:
-        raise ResponseStatusError(f"Response {response.status}: Bad gateway")
+        raise ResponseStatusError(f"Response {response.status}: Bad Gateway")
+    elif response.status == 504:
+        raise RateLimitError(f"Response {response.status}: Gateway Timeout ")
     else:
         raise ResponseStatusError(f"Response {response.status}: {message}")
 
@@ -70,6 +69,8 @@ def raise_for_status(response: Union[Response, StreamResponse, ClientResponse, R
     elif response.status_code == 403 and is_openai(response.text):
         raise ResponseStatusError(f"Response {response.status_code}: OpenAI Bot detected")
     elif response.status_code == 502:
-        raise ResponseStatusError(f"Response {response.status_code}: Bad gateway")
+        raise ResponseStatusError(f"Response {response.status_code}: Bad Gateway")
+    elif response.status_code == 504:
+        raise RateLimitError(f"Response {response.status_code}: Gateway Timeout ")
     else:
         raise ResponseStatusError(f"Response {response.status_code}: {message}")