返回提交历史
Modified
docs/media.md
+24
-0
Modified
g4f/Provider/PollinationsAI.py
+26
-20
Modified
g4f/Provider/audio/EdgeTTS.py
+10
-9
Modified
g4f/Provider/needs_auth/Gemini.py
+16
-2
Modified
g4f/api/__init__.py
+7
-1
Modified
g4f/client/__init__.py
+2
-1
Modified
g4f/client/stubs.py
+4
-0
Modified
g4f/image/copy_images.py
+1
-1
XFEstudio/gpt4free
feat: add audio generation support for multiple providers
- Added new examples for `client.media.generate` with `PollinationsAI`, `EdgeTTS`, and `Gemini` in `docs/media.md` - Modified `PollinationsAI.py` to default to `default_audio_model` when audio data is present - Adjusted `PollinationsAI.py` to conditionally construct message list from `prompt` when media is being generated - Rearranged `PollinationsAI.py` response handling to yield `save_response_media` after checking for non-JSON content types - Added support in `EdgeTTS.py` to use default values for `language`, `locale`, and `format` from class attributes - Improved voice selection logic in `EdgeTTS.py` to fallback to default locale or language when not explicitly provided - Updated `EdgeTTS.py` to yield `AudioResponse` with `text` field included - Modified `Gemini.py` to support `.ogx` audio generation when `model == "gemini-audio"` or `audio` is passed - Used `format_image_prompt` in `Gemini.py` to create audio prompt and saved audio file using `synthesize` - Appended `AudioResponse` to `Gemini.py` for audio generation flow - Added `save()` method to `Image` class in `stubs.py` to support saving `/media/` files locally - Changed `client/__init__.py` to fallback to `options["text"]` if `alt` is missing in `Images.create` - Ensured `AudioResponse` in `copy_images.py` includes the `text` (prompt) field - Added `Annotated` fallback definition in `api/__init__.py` for compatibility with older Python versions
b68b9ff6
代码差异
8 个文件
+90
-34
@@ -28,6 +28,30 @@ async def main():
28
28
asyncio.run(main())
29
29
```
30
30
31
#### **More examples for Generate Audio:**
32
33
```python
34
from g4f.client import Client
35
36
from g4f.Provider import EdgeTTS, Gemini, PollinationsAI
37
38
client = Client(provider=PollinationsAI)
39
response = client.media.generate("Hello", audio={"voice": "alloy", "format": "mp3"})
40
response.data[0].save("openai.mp3")
41
42
client = Client(provider=PollinationsAI)
43
response = client.media.generate("Hello", model="hypnosis-tracy")
44
response.data[0].save("hypnosis.mp3")
45
46
client = Client(provider=Gemini)
47
response = client.media.generate("Hello", model="gemini-audio")
48
response.data[0].save("gemini.ogx")
49
50
client = Client(provider=EdgeTTS)
51
response = client.media.generate("Hello", audio={"locale": "en-US"})
52
response.data[0].save("edge-tts.mp3")
53
```
54
31
55
#### **Transcribe an Audio File:**
32
56
33
57
Some providers in G4F support audio inputs in chat completions, allowing you to transcribe audio files by instructing the model accordingly. This example demonstrates how to use the `AsyncClient` to transcribe an audio file asynchronously:
@@ -177,7 +177,7 @@ class PollinationsAI(AsyncGeneratorProvider, ProviderModelMixin):
177
177
if is_data_an_audio(media_data, filename):
178
178
has_audio = True
179
179
break
180
model = next(iter(cls.audio_models)) if has_audio else model
180
model = cls.default_audio_model if has_audio else model
181
181
try:
182
182
model = cls.get_model(model)
183
183
except ModelNotFoundError:
@@ -202,6 +202,11 @@ class PollinationsAI(AsyncGeneratorProvider, ProviderModelMixin):
202
202
):
203
203
yield chunk
204
204
else:
205
if prompt is not None and len(messages) == 1:
206
messages = [{
207
"role": "user",
208
"content": prompt
209
}]
205
210
async for result in cls._generate_text(
206
211
model=model,
207
212
messages=messages,
@@ -315,9 +320,6 @@ class PollinationsAI(AsyncGeneratorProvider, ProviderModelMixin):
315
320
})
316
321
async with session.post(url, json=data) as response:
317
322
await raise_for_status(response)
318
async for chunk in save_response_media(response, format_image_prompt(messages), [model]):
319
yield chunk
320
return
321
323
if response.headers["content-type"].startswith("text/plain"):
322
324
yield await response.text()
323
325
return
@@ -339,20 +341,24 @@ class PollinationsAI(AsyncGeneratorProvider, ProviderModelMixin):
339
341
finish_reason = choice.get("finish_reason")
340
342
if finish_reason:
341
343
yield FinishReason(finish_reason)
342
return
343
result = await response.json()
344
if "choices" in result:
345
choice = result["choices"][0]
346
message = choice.get("message", {})
347
content = message.get("content", "")
348
if content:
349
yield content
350
if "tool_calls" in message:
351
yield ToolCalls(message["tool_calls"])
344
elif response.headers["content-type"].startswith("application/json"):
345
result = await response.json()
346
if "choices" in result:
347
choice = result["choices"][0]
348
message = choice.get("message", {})
349
content = message.get("content", "")
350
if content:
351
yield content
352
if "tool_calls" in message:
353
yield ToolCalls(message["tool_calls"])
354
else:
355
raise ResponseError(result)
356
if result.get("usage") is not None:
357
yield Usage(**result["usage"])
358
finish_reason = choice.get("finish_reason")
359
if finish_reason:
360
yield FinishReason(finish_reason)
352
361
else:
353
raise ResponseError(result)
354
if result.get("usage") is not None:
355
yield Usage(**result["usage"])
356
finish_reason = choice.get("finish_reason")
357
if finish_reason:
358
yield FinishReason(finish_reason)
362
async for chunk in save_response_media(response, format_image_prompt(messages), [model]):
363
yield chunk
364
return
@@ -20,8 +20,9 @@ from ..helper import format_image_prompt
20
20
class EdgeTTS(AsyncGeneratorProvider, ProviderModelMixin):
21
21
label = "Edge TTS"
22
22
working = has_edge_tts
23
default_model = "edge-tts"
23
default_language = "en"
24
24
default_locale = "en-US"
25
default_format = "mp3"
25
26
26
27
@classmethod
27
28
def get_models(cls) -> list[str]:
@@ -38,29 +39,29 @@ class EdgeTTS(AsyncGeneratorProvider, ProviderModelMixin):
38
39
messages: Messages,
39
40
proxy: str = None,
40
41
prompt: str = None,
41
audio: dict = {"voice": None, "format": "mp3"},
42
audio: dict = {},
42
43
**kwargs
43
44
) -> AsyncResult:
44
45
prompt = format_image_prompt(messages, prompt)
45
46
if not prompt:
46
47
raise ValueError("Prompt is empty.")
47
voice = audio.get("voice", model)
48
voice = audio.get("voice", model if model and model != "edge-tts" else None)
48
49
if not voice:
49
50
voices = await VoicesManager.create()
50
51
if "locale" in audio:
51
52
voices = voices.find(Locale=audio["locale"])
52
elif "language" in audio:
53
if "-" in audio["language"]:
54
voices = voices.find(Locale=audio["language"])
53
elif audio.get("language", cls.default_language) != cls.default_language:
54
if "-" in audio.get("language"):
55
voices = voices.find(Locale=audio.get("language"))
55
56
else:
56
voices = voices.find(Language=audio["language"])
57
voices = voices.find(Language=audio.get("language"))
57
58
else:
58
59
voices = voices.find(Locale=cls.default_locale)
59
60
if not voices:
60
61
raise ValueError(f"No voices found for language '{audio.get('language')}' and locale '{audio.get('locale')}'.")
61
62
voice = random.choice(voices)["Name"]
62
63
63
format = audio.get("format", "mp3")
64
format = audio.get("format", cls.default_format)
64
65
filename = get_filename([cls.default_model], prompt, f".{format}", prompt)
65
66
target_path = os.path.join(get_media_dir(), filename)
66
67
ensure_media_dir()
@@ -69,4 +70,4 @@ class EdgeTTS(AsyncGeneratorProvider, ProviderModelMixin):
69
70
communicate = edge_tts.Communicate(prompt, voice=voice, proxy=proxy, **extra_parameters)
70
71
71
72
await communicate.save(target_path)
72
yield AudioResponse(f"/media/{filename}", voice=voice, prompt=prompt)
73
yield AudioResponse(f"/media/{filename}", voice=voice, text=prompt)
@@ -20,16 +20,17 @@ except ImportError:
20
20
21
21
from ... import debug
22
22
from ...typing import Messages, Cookies, MediaListType, AsyncResult, AsyncIterator
23
from ...providers.response import JsonConversation, Reasoning, RequestLogin, ImageResponse, YouTube
23
from ...providers.response import JsonConversation, Reasoning, RequestLogin, ImageResponse, YouTube, AudioResponse
24
24
from ...requests.raise_for_status import raise_for_status
25
25
from ...requests.aiohttp import get_connector
26
26
from ...requests import get_nodriver
27
from ...image.copy_images import get_filename, get_media_dir, ensure_media_dir
27
28
from ...errors import MissingAuthError
28
29
from ...image import to_bytes
29
30
from ...cookies import get_cookies_dir
30
31
from ...tools.media import merge_media
31
32
from ..base_provider import AsyncGeneratorProvider, ProviderModelMixin
32
from ..helper import format_prompt, get_cookies, get_last_user_message
33
from ..helper import format_prompt, get_cookies, get_last_user_message, format_image_prompt
33
34
from ... import debug
34
35
35
36
REQUEST_HEADERS = {
@@ -68,6 +69,7 @@ models = {
68
69
"gemini-2.0-flash-exp": {"x-goog-ext-525001261-jspb": '[null,null,null,null,"f299729663a2343f"]'},
69
70
"gemini-2.0-flash-thinking": {"x-goog-ext-525001261-jspb": '[null,null,null,null,"9c17b1863f581b8a"]'},
70
71
"gemini-2.0-flash-thinking-with-apps": {"x-goog-ext-525001261-jspb": '[null,null,null,null,"f8f8f5ea629f5d37"]'},
72
"gemini-audio": {}
71
73
}
72
74
73
75
class Gemini(AsyncGeneratorProvider, ProviderModelMixin):
@@ -153,8 +155,20 @@ class Gemini(AsyncGeneratorProvider, ProviderModelMixin):
153
155
return_conversation: bool = False,
154
156
conversation: Conversation = None,
155
157
language: str = "en",
158
prompt: str = None,
159
audio: dict = None,
156
160
**kwargs
157
161
) -> AsyncResult:
162
if audio is not None or model == "gemini-audio":
163
prompt = format_image_prompt(messages, prompt)
164
filename = get_filename(["gemini"], prompt, ".ogx", prompt)
165
ensure_media_dir()
166
path = os.path.join(get_media_dir(), filename)
167
with open(path, "wb") as f:
168
async for chunk in cls.synthesize({"text": prompt}, proxy):
169
f.write(chunk)
170
yield AudioResponse(f"/media/{filename}", text=prompt)
171
return
158
172
cls._cookies = cookies or cls._cookies or get_cookies(GOOGLE_COOKIE_DOMAIN, False, True)
159
173
if conversation is not None and getattr(conversation, "model", None) != model:
160
174
conversation = None
@@ -33,6 +33,12 @@ from starlette.responses import FileResponse
33
33
from types import SimpleNamespace
34
34
from typing import Union, Optional, List
35
35
36
try:
37
from typing import Annotated
38
except ImportError:
39
class Annotated:
40
pass
41
36
42
import g4f
37
43
import g4f.Provider
38
44
import g4f.debug
@@ -52,7 +58,7 @@ from .stubs import (
52
58
ChatCompletionsConfig, ImageGenerationConfig,
53
59
ProviderResponseModel, ModelResponseModel,
54
60
ErrorResponseModel, ProviderResponseDetailModel,
55
FileResponseModel, UploadResponseModel, Annotated
61
FileResponseModel, UploadResponseModel
56
62
)
57
63
from g4f import debug
58
64
@@ -463,7 +463,8 @@ class Images:
463
463
urls.extend(item.urls)
464
464
if not urls:
465
465
return None
466
return MediaResponse(urls, items[0].alt, items[0].options)
466
alt = getattr(items[0], "alt", items[0].options.get("text"))
467
return MediaResponse(urls, alt, items[0].options)
467
468
468
469
def create_variation(
469
470
self,
@@ -217,6 +217,10 @@ class Image(BaseModel):
217
217
revised_prompt=revised_prompt
218
218
))
219
219
220
def save(self, path: str):
221
if self.url is not None and self.url.startswith("/media/"):
222
os.rename(self.url.replace("/media", get_media_dir()), path)
223
220
224
class ImagesResponse(BaseModel):
221
225
data: List[Image]
222
226
model: str
@@ -70,7 +70,7 @@ async def save_response_media(response: StreamResponse, prompt: str, tags: list[
70
70
if response.method == "GET":
71
71
media_url = f"{media_url}?url={str(response.url)}"
72
72
if content_type.startswith("audio/"):
73
yield AudioResponse(media_url)
73
yield AudioResponse(media_url, text=prompt)
74
74
elif content_type.startswith("video/"):
75
75
yield VideoResponse(media_url, prompt)
76
76
else: