Add new Client API with Docs

Use object urls for the preview of image uploads.
Fix upload images in You provider
Fix create image. It's now a single image.
Improve system message for create images.

.github/workflows/unittest.yml

@@ -24,7 +24,7 @@ jobs:
       run: pip install -r requirements-min.txt
     - name: Run tests
       run: python -m etc.unittest
-    - name: Set up Python 3.11
+    - name: Set up Python 3.12
       uses: actions/setup-python@v4
       with:
         python-version: "3.12"

README.md

@@ -226,6 +226,23 @@ docker-compose down
 
 ## 💡 Usage
 
+### New Client with Image Generation
+```python
+from g4f.client import Client
+
+client = Client()
+response = client.images.generate(
+  model="gemini",
+  prompt="a white siamese cat",
+  ...
+)
+image_url = response.data[0].url
+```
+Result:
+[![Image with cat](/docs/cat.jpeg)](/docs/client.md)
+
+[to the client API](/docs/client.md)
+
 ### The Web UI
 
 To start the web interface, type the following codes in the command line.

docs/client.md

@@ -0,0 +1,71 @@
+### Client API
+##### from g4f (beta)
+
+#### Start
+This new client could:
+
+```python
+from g4f.client import Client
+```
+replaces this:
+
+```python
+from openai import OpenAI
+```
+in your Python Code.
+
+New client have the same API as OpenAI.
+
+#### Client
+
+Create the client with custom providers:
+
+```python
+from g4f.client import Client
+from g4f.Provider import BingCreateImages, OpenaiChat, Gemini
+
+client = Client(
+    provider=OpenaiChat,
+    image_provider=Gemini,
+    proxies=None
+)
+```
+
+#### Examples
+
+Use the ChatCompletions:
+
+```python
+stream = client.chat.completions.create(
+    model="gpt-4",
+    messages=[{"role": "user", "content": "Say this is a test"}],
+    stream=True,
+)
+for chunk in stream:
+    if chunk.choices[0].delta.content is not None:
+        print(chunk.choices[0].delta.content, end="")
+```
+
+Or use it for creating a image:
+```python
+response = client.images.generate(
+  model="dall-e-3",
+  prompt="a white siamese cat",
+  ...
+)
+
+image_url = response.data[0].url
+```
+
+Also this works with the client:
+```python
+response = client.images.create_variation(
+  image=open('cat.jpg')
+  model="bing",
+  ...
+)
+
+image_url = response.data[0].url
+```
+
+[to Home](/docs/client.md)

g4f/Provider/BingCreateImages.py

@@ -1,60 +1,22 @@
 from __future__ import annotations
 
 import asyncio
-import time
 import os
 from typing import Generator
 
 from ..cookies import get_cookies
-from ..webdriver import WebDriver, get_driver_cookies, get_browser
 from ..image import ImageResponse
 from ..errors import MissingRequirementsError, MissingAuthError
-from .bing.create_images import BING_URL, create_images, create_session
+from .bing.create_images import create_images, create_session, get_cookies_from_browser
 
-BING_URL = "https://www.bing.com"
-TIMEOUT_LOGIN = 1200
-
-def wait_for_login(driver: WebDriver, timeout: int = TIMEOUT_LOGIN) -> None:
-    """
-    Waits for the user to log in within a given timeout period.
-
-    Args:
-        driver (WebDriver): Webdriver for browser automation.
-        timeout (int): Maximum waiting time in seconds.
-
-    Raises:
-        RuntimeError: If the login process exceeds the timeout.
-    """
-    driver.get(f"{BING_URL}/")
-    start_time = time.time()
-    while not driver.get_cookie("_U"):
-        if time.time() - start_time > timeout:
-            raise RuntimeError("Timeout error")
-        time.sleep(0.5)
-
-def get_cookies_from_browser(proxy: str = None) -> dict[str, str]:
-    """
-    Retrieves cookies from the browser using webdriver.
-
-    Args:
-        proxy (str, optional): Proxy configuration.
-
-    Returns:
-        dict[str, str]: Retrieved cookies.
-    """
-    with get_browser(proxy=proxy) as driver:
-        wait_for_login(driver)
-        time.sleep(1)
-        return get_driver_cookies(driver)
-
-class CreateImagesBing:
+class BingCreateImages:
     """A class for creating images using Bing."""
 
     def __init__(self, cookies: dict[str, str] = {}, proxy: str = None) -> None:
         self.cookies = cookies
         self.proxy = proxy
 
-    def create_completion(self, prompt: str) -> Generator[ImageResponse, None, None]:
+    def create(self, prompt: str) -> Generator[ImageResponse, None, None]:
         """
         Generator for creating imagecompletion based on a prompt.
 
@@ -91,4 +53,4 @@ class CreateImagesBing:
         proxy = self.proxy or os.environ.get("G4F_PROXY")
         async with create_session(cookies, proxy) as session:
             images = await create_images(session, prompt, proxy)
-            return ImageResponse(images, prompt, {"preview": "{image}?w=200&h=200"})
+            return ImageResponse(images, prompt, {"preview": "{image}?w=200&h=200"} if len(images) > 1 else {})

g4f/Provider/You.py

@@ -58,9 +58,14 @@ class You(AsyncGeneratorProvider):
                 "selectedChatMode": chat_mode,
                 #"chat": json.dumps(chat),
             }
+            params = {
+                "userFiles": upload,
+                "selectedChatMode": chat_mode,
+            }
             async with (client.post if chat_mode == "default" else client.get)(
                 f"{cls.url}/api/streamingSearch",
                 data=data,
+                params=params,
                 headers=headers,
                 cookies=cookies
             ) as response:

g4f/Provider/__init__.py

@@ -53,7 +53,7 @@ from .Vercel          import Vercel
 from .Ylokh           import Ylokh
 from .You             import You
 
-from .CreateImagesBing import CreateImagesBing
+from .BingCreateImages import BingCreateImages
 
 import sys
 

g4f/Provider/bing/create_images.py

@@ -21,8 +21,10 @@ from ..create_images import CreateImagesProvider
 from ..helper import get_connector
 from ...base_provider import ProviderType
 from ...errors import MissingRequirementsError
+from ...webdriver import WebDriver, get_driver_cookies, get_browser
 
 BING_URL = "https://www.bing.com"
+TIMEOUT_LOGIN = 1200
 TIMEOUT_IMAGE_CREATION = 300
 ERRORS = [
     "this prompt is being reviewed",
@@ -35,6 +37,39 @@ BAD_IMAGES = [
     "https://r.bing.com/rp/TX9QuO3WzcCJz1uaaSwQAz39Kb0.jpg",
 ]
 
+def wait_for_login(driver: WebDriver, timeout: int = TIMEOUT_LOGIN) -> None:
+    """
+    Waits for the user to log in within a given timeout period.
+
+    Args:
+        driver (WebDriver): Webdriver for browser automation.
+        timeout (int): Maximum waiting time in seconds.
+
+    Raises:
+        RuntimeError: If the login process exceeds the timeout.
+    """
+    driver.get(f"{BING_URL}/")
+    start_time = time.time()
+    while not driver.get_cookie("_U"):
+        if time.time() - start_time > timeout:
+            raise RuntimeError("Timeout error")
+        time.sleep(0.5)
+
+def get_cookies_from_browser(proxy: str = None) -> dict[str, str]:
+    """
+    Retrieves cookies from the browser using webdriver.
+
+    Args:
+        proxy (str, optional): Proxy configuration.
+
+    Returns:
+        dict[str, str]: Retrieved cookies.
+    """
+    with get_browser(proxy=proxy) as driver:
+        wait_for_login(driver)
+        time.sleep(1)
+        return get_driver_cookies(driver)
+
 def create_session(cookies: Dict[str, str], proxy: str = None, connector: BaseConnector = None) -> ClientSession:
     """
     Creates a new client session with specified cookies and headers.
@@ -141,6 +176,8 @@ def read_images(html_content: str) -> List[str]:
     """
     soup = BeautifulSoup(html_content, "html.parser")
     tags = soup.find_all("img", class_="mimg")
+    if not tags:
+        tags = soup.find_all("img", class_="gir_mmimg")
     images = [img["src"].split("?w=")[0] for img in tags]
     if any(im in BAD_IMAGES for im in images):
         raise RuntimeError("Bad images found")
@@ -158,10 +195,10 @@ def patch_provider(provider: ProviderType) -> CreateImagesProvider:
     Returns:
         CreateImagesProvider: The patched provider with image creation capabilities.
     """
-    from ..CreateImagesBing import CreateImagesBing
-    service = CreateImagesBing()
+    from ..BingCreateImages import BingCreateImages
+    service = BingCreateImages()
     return CreateImagesProvider(
         provider,
-        service.create_completion,
+        service.create,
         service.create_async
     )

g4f/Provider/create_images.py

@@ -7,10 +7,14 @@ from ..typing import CreateResult, Messages
 from ..base_provider import BaseProvider, ProviderType
 
 system_message = """
-You can generate custom images with the DALL-E 3 image generator.
+You can generate images, pictures, photos or img with the DALL-E 3 image generator.
 To generate an image with a prompt, do this:
+
 <img data-prompt=\"keywords for the image\">
-Don't use images with data uri. It is important to use a prompt instead.
+
+Never use own image links. Don't wrap it in backticks.
+It is important to use a only a img tag with a prompt.
+
 <img data-prompt=\"image caption\">
 """
 

g4f/Provider/needs_auth/OpenaiChat.py

@@ -386,50 +386,47 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin):
                 ) as response:
                     if not response.ok:
                         raise RuntimeError(f"Response {response.status_code}: {await response.text()}")
-                    try:
-                        last_message: int = 0
-                        async for line in response.iter_lines():
-                            if not line.startswith(b"data: "):
-                                continue
-                            elif line.startswith(b"data: [DONE]"):
-                                break
-                            try:
-                                line = json.loads(line[6:])
-                            except:
-                                continue
-                            if "message" not in line:
-                                continue
-                            if "error" in line and line["error"]:
-                                raise RuntimeError(line["error"])
-                            if "message_type" not in line["message"]["metadata"]:
-                                continue
-                            try:
-                                image_response = await cls.get_generated_image(session, auth_headers, line)
-                                if image_response:
-                                    yield image_response
-                            except Exception as e:
-                                yield e
-                            if line["message"]["author"]["role"] != "assistant":
-                                continue
-                            if line["message"]["content"]["content_type"] != "text":
-                                continue
-                            if line["message"]["metadata"]["message_type"] not in ("next", "continue", "variant"):
-                                continue
-                            conversation_id = line["conversation_id"]
-                            parent_id = line["message"]["id"]
-                            if response_fields:
-                                response_fields = False
-                                yield ResponseFields(conversation_id, parent_id, end_turn)
-                            if "parts" in line["message"]["content"]:
-                                new_message = line["message"]["content"]["parts"][0]
-                                if len(new_message) > last_message:
-                                    yield new_message[last_message:]
-                                last_message = len(new_message)
-                            if "finish_details" in line["message"]["metadata"]:
-                                if line["message"]["metadata"]["finish_details"]["type"] == "stop":
-                                    end_turn.end()
-                    except Exception as e:
-                        raise e
+                    last_message: int = 0
+                    async for line in response.iter_lines():
+                        if not line.startswith(b"data: "):
+                            continue
+                        elif line.startswith(b"data: [DONE]"):
+                            break
+                        try:
+                            line = json.loads(line[6:])
+                        except:
+                            continue
+                        if "message" not in line:
+                            continue
+                        if "error" in line and line["error"]:
+                            raise RuntimeError(line["error"])
+                        if "message_type" not in line["message"]["metadata"]:
+                            continue
+                        try:
+                            image_response = await cls.get_generated_image(session, auth_headers, line)
+                            if image_response:
+                                yield image_response
+                        except Exception as e:
+                            yield e
+                        if line["message"]["author"]["role"] != "assistant":
+                            continue
+                        if line["message"]["content"]["content_type"] != "text":
+                            continue
+                        if line["message"]["metadata"]["message_type"] not in ("next", "continue", "variant"):
+                            continue
+                        conversation_id = line["conversation_id"]
+                        parent_id = line["message"]["id"]
+                        if response_fields:
+                            response_fields = False
+                            yield ResponseFields(conversation_id, parent_id, end_turn)
+                        if "parts" in line["message"]["content"]:
+                            new_message = line["message"]["content"]["parts"][0]
+                            if len(new_message) > last_message:
+                                yield new_message[last_message:]
+                            last_message = len(new_message)
+                        if "finish_details" in line["message"]["metadata"]:
+                            if line["message"]["metadata"]["finish_details"]["type"] == "stop":
+                                end_turn.end()
                 if not auto_continue:
                     break
                 action = "continue"

g4f/__init__.py

@@ -16,7 +16,8 @@ def get_model_and_provider(model    : Union[Model, str],
                            stream   : bool,
                            ignored  : list[str] = None,
                            ignore_working: bool = False,
-                           ignore_stream: bool = False) -> tuple[str, ProviderType]:
+                           ignore_stream: bool = False,
+                           **kwargs) -> tuple[str, ProviderType]:
     """
     Retrieves the model and provider based on input parameters.
 

g4f/client.py

@@ -0,0 +1,267 @@
+from __future__ import annotations
+
+import re
+
+from .typing import Union, Generator, AsyncGenerator, Messages, ImageType
+from .base_provider import BaseProvider, ProviderType
+from .Provider.base_provider import AsyncGeneratorProvider
+from .image import ImageResponse as ImageProviderResponse
+from .Provider import BingCreateImages, Gemini, OpenaiChat
+from .errors import NoImageResponseError
+from . import get_model_and_provider
+
+ImageProvider = Union[BaseProvider, object]
+Proxies = Union[dict, str]
+
+def read_json(text: str) -> dict:
+    """
+    Parses JSON code block from a string.
+
+    Args:
+        text (str): A string containing a JSON code block.
+
+    Returns:
+        dict: A dictionary parsed from the JSON code block.
+    """
+    match = re.search(r"```(json|)\n(?P<code>[\S\s]+?)\n```", text)
+    if match:
+        return match.group("code")
+    return text
+
+def iter_response(
+    response: iter,
+    stream: bool,
+    response_format: dict = None,
+    max_tokens: int = None,
+    stop: list = None
+) -> Generator:
+    content = ""
+    idx = 1
+    chunk = None
+    finish_reason = "stop"
+    for idx, chunk in enumerate(response):
+        content += str(chunk)
+        if max_tokens is not None and idx > max_tokens:
+            finish_reason = "max_tokens"
+            break
+        first = -1
+        word = None
+        if stop is not None:
+            for word in list(stop):
+                first = content.find(word)
+                if first != -1:
+                    content = content[:first]
+                    break
+            if stream:
+                if first != -1:
+                    first = chunk.find(word)
+                    if first != -1:
+                        chunk = chunk[:first]
+                    else:
+                        first = 0
+            yield ChatCompletionChunk([ChatCompletionDeltaChoice(ChatCompletionDelta(chunk))])
+        if first != -1:
+            break
+    if not stream:
+        if response_format is not None and "type" in response_format:
+            if response_format["type"] == "json_object":
+                response = read_json(response)
+        yield ChatCompletion([ChatCompletionChoice(ChatCompletionMessage(response, finish_reason))])
+
+async def aiter_response(
+    response: aiter,
+    stream: bool,
+    response_format: dict = None,
+    max_tokens: int = None,
+    stop: list = None
+) -> AsyncGenerator:
+    content = ""
+    try:
+        idx = 0
+        chunk = None
+        async for chunk in response:
+            content += str(chunk)
+            if max_tokens is not None and idx > max_tokens:
+                break
+            first = -1
+            word = None
+            if stop is not None:
+                for word in list(stop):
+                    first = content.find(word)
+                    if first != -1:
+                        content = content[:first]
+                        break
+            if stream:
+                if first != -1:
+                    first = chunk.find(word)
+                    if first != -1:
+                        chunk = chunk[:first]
+                    else:
+                        first = 0
+                yield ChatCompletionChunk([ChatCompletionDeltaChoice(ChatCompletionDelta(chunk))])
+            if first != -1:
+                break
+            idx += 1
+    except:
+        ...
+    if not stream:
+        if response_format is not None and "type" in response_format:
+            if response_format["type"] == "json_object":
+                response = read_json(response)
+        yield ChatCompletion([ChatCompletionChoice(ChatCompletionMessage(response))])
+
+class Model():
+    def __getitem__(self, item):
+        return getattr(self, item)
+
+class ChatCompletion(Model):
+    def __init__(self, choices: list):
+        self.choices = choices
+        
+class ChatCompletionChunk(Model):
+    def __init__(self, choices: list):
+        self.choices = choices
+
+class ChatCompletionChoice(Model):
+    def __init__(self, message: ChatCompletionMessage):
+        self.message = message
+        
+class ChatCompletionMessage(Model):
+    def __init__(self, content: str, finish_reason: str):
+        self.content = content
+        self.finish_reason = finish_reason
+        self.index = 0
+        self.logprobs = None
+        
+class ChatCompletionDelta(Model):
+    def __init__(self, content: str):
+        self.content = content
+        
+class ChatCompletionDeltaChoice(Model):
+    def __init__(self, delta: ChatCompletionDelta):
+        self.delta = delta
+
+class Client():
+    proxies: Proxies = None
+    chat: Chat
+
+    def __init__(
+        self,
+        provider: ProviderType = None,
+        image_provider: ImageProvider = None,
+        proxies: Proxies = None,
+        **kwargs
+    ) -> None:
+        self.proxies: Proxies = proxies
+        self.images = Images(self, image_provider)
+        self.chat = Chat(self, provider)
+
+    def get_proxy(self) -> Union[str, None]:
+        if isinstance(self.proxies, str) or self.proxies is None:
+            return self.proxies
+        elif "all" in self.proxies:
+            return self.proxies["all"]
+        elif "https" in self.proxies:
+            return self.proxies["https"]
+        return None
+
+class Completions():
+    def __init__(self, client: Client, provider: ProviderType = None):
+        self.client: Client = client
+        self.provider: ProviderType = provider
+
+    def create(
+        self,
+        messages: Messages,
+        model: str,
+        provider: ProviderType = None,
+        stream: bool = False,
+        response_format: dict = None,
+        max_tokens: int = None,
+        stop: list = None,
+        **kwargs
+    ) -> Union[dict, Generator]:
+        if max_tokens is not None:
+            kwargs["max_tokens"] = max_tokens
+        if stop:
+            kwargs["stop"] = list(stop)
+        model, provider = get_model_and_provider(
+            model,
+            self.provider if provider is None else provider,
+            stream,
+            **kwargs
+        )
+        response = provider.create_completion(model, messages, stream=stream, **kwargs)
+        if isinstance(provider, type) and issubclass(provider, AsyncGeneratorProvider):
+            response = iter_response(response, stream, response_format) # max_tokens, stop
+        else:
+            response = iter_response(response, stream, response_format, max_tokens, stop)
+        return response if stream else next(response)
+
+class Chat():
+    completions: Completions
+
+    def __init__(self, client: Client, provider: ProviderType = None):
+        self.completions = Completions(client, provider)
+        
+class ImageModels():
+    gemini = Gemini
+    openai = OpenaiChat
+
+    def __init__(self, client: Client) -> None:
+        self.client = client
+        self.default = BingCreateImages(proxy=self.client.get_proxy())
+
+    def get(self, name: str) -> ImageProvider:
+        return getattr(self, name) if hasattr(self, name) else self.default
+
+class ImagesResponse(Model):
+    data: list[Image]
+
+    def __init__(self, data: list) -> None:
+        self.data = data
+    
+class Image(Model):
+    url: str
+
+    def __init__(self, url: str) -> None:
+        self.url = url
+    
+class Images():
+    def __init__(self, client: Client, provider: ImageProvider = None):
+        self.client: Client = client
+        self.provider: ImageProvider = provider
+        self.models: ImageModels = ImageModels(client)
+
+    def generate(self, prompt, model: str = None, **kwargs):
+        provider = self.models.get(model) if model else self.provider or self.models.get(model)
+        if isinstance(provider, BaseProvider) or isinstance(provider, type) and issubclass(provider, BaseProvider):
+            prompt = f"create a image: {prompt}"
+            response = provider.create_completion(
+                "",
+                [{"role": "user", "content": prompt}],
+                True,
+                proxy=self.client.get_proxy()
+            )
+        else:
+            response = provider.create(prompt)
+            
+        for chunk in response:
+            if isinstance(chunk, ImageProviderResponse):
+                return ImagesResponse([Image(image)for image in list(chunk.images)])
+        raise NoImageResponseError()
+
+    def create_variation(self, image: ImageType, model: str = None, **kwargs):
+        provider = self.models.get(model) if model else self.provider
+        if isinstance(provider, BaseProvider):
+            response = provider.create_completion(
+                "",
+                [{"role": "user", "content": "create a image like this"}],
+                True,
+                image=image,
+                proxy=self.client.get_proxy()
+            )
+            for chunk in response:
+                if isinstance(chunk, ImageProviderResponse):
+                    return ImagesResponse([Image(image)for image in list(chunk.images)])
+        raise NoImageResponseError()

g4f/errors.py

@@ -1,35 +1,38 @@
 class ProviderNotFoundError(Exception):
-    pass
+    ...
 
 class ProviderNotWorkingError(Exception):
-    pass
+    ...
 
 class StreamNotSupportedError(Exception):
-    pass
+    ...
 
 class ModelNotFoundError(Exception):
-    pass
+    ...
 
 class ModelNotAllowedError(Exception):
-    pass
+    ...
 
 class RetryProviderError(Exception):
-    pass
+    ...
 
 class RetryNoProviderError(Exception):
-    pass
+    ...
 
 class VersionNotFoundError(Exception):
-    pass
+    ...
 
 class NestAsyncioError(Exception):
-    pass
+    ...
 
 class ModelNotSupportedError(Exception):
-    pass
+    ...
 
 class MissingRequirementsError(Exception):
-    pass
+    ...
 
 class MissingAuthError(Exception):
-    pass
+    ...
+
+class NoImageResponseError(Exception):
+    ...

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

@@ -52,6 +52,12 @@ const handle_ask = async () => {
         }
         await add_message(window.conversation_id, "user", message);
         window.token = message_id();
+
+        if (imageInput.dataset.src) URL.revokeObjectURL(imageInput.dataset.src);
+        const input = imageInput && imageInput.files.length > 0 ? imageInput : cameraInput
+        if (input.files.length > 0) imageInput.dataset.src = URL.createObjectURL(input.files[0]);
+        else delete imageInput.dataset.src
+
         message_box.innerHTML += `
             <div class="message">
                 <div class="user">
@@ -64,10 +70,6 @@ const handle_ask = async () => {
                         ? '<img src="' + imageInput.dataset.src + '" alt="Image upload">'
                         : ''
                     }
-                    ${cameraInput.dataset.src
-                        ? '<img src="' + cameraInput.dataset.src + '" alt="Image capture">'
-                        : ''
-                    }
                 </div>
             </div>
         `;
@@ -683,24 +685,13 @@ observer.observe(message_input, { attributes: true });
     document.getElementById("version_text").innerHTML = text
 })()
 for (const el of [imageInput, cameraInput]) {
-    console.log(el.files);
     el.addEventListener('click', async () => {
         el.value = '';
-        delete el.dataset.src;
-    });
-    do_load = async () => {
-        if (el.files.length) {
-            delete imageInput.dataset.src;
-            delete cameraInput.dataset.src;
-            const reader = new FileReader();
-            reader.addEventListener('load', (event) => {
-                el.dataset.src = event.target.result;
-            });
-            reader.readAsDataURL(el.files[0]);
+        if (imageInput.dataset.src) {
+            URL.revokeObjectURL(imageInput.dataset.src);
+            delete imageInput.dataset.src
         }
-    }
-    do_load()
-    el.addEventListener('change', do_load);
+    });
 }
 fileInput.addEventListener('click', async (event) => {
     fileInput.value = '';