ATH-MaaS/Pixelle-Video · error · RuntimeError

未在响应中找到 url 或 b64_json

Error message

未在响应中找到 url 或 b64_json

What it means

generate_image raises RuntimeError when the API response contains image data whose items expose neither a usable url nor b64_json. The code handles b64_json and url cases; anything else (e.g. revised_prompt only, or unexpected relay formats) falls through to this error.

Source

Thrown at pixelle_video/services/api_services/image_gpt.py:128

                        file_name = f"gpt_{int(time.time())}_{uuid.uuid4().hex[:6]}.png"
                        file_path = os.path.join(save_dir, file_name)
                        with open(file_path, "wb") as f:
                            f.write(base64.b64decode(img_data.b64_json))
                        return file_path
                    return img_data.b64_json

                # 2. 处理 URL 格式
                elif hasattr(img_data, 'url') and img_data.url:
                    url = img_data.url
                    if save_dir:
                        os.makedirs(save_dir, exist_ok=True)
                        file_name = f"gpt_{int(time.time())}_{uuid.uuid4().hex[:6]}.png"
                        file_path = os.path.join(save_dir, file_name)
                        if self.image_processor.download_image(url, file_path):
                            return file_path
                        return url
                
                raise RuntimeError("未在响应中找到 url 或 b64_json")
            except Exception as e:
                last_error = e
                msg = str(e)
                # Other errors: wait before retry
                print(f"Image generation error: {e}. Retrying in 10 seconds.")
                time.sleep(10)
                break  # Break inner loop to retry all models
            attempts += 1
        raise Exception(f"Max attempts reached, failed to generate image. Last error: {last_error}")

    def generate_images(self, prompt, count=4, size="1024x1024", quality="standard", model=None):
        """Generate multiple image URLs by calling Images API 'count' times."""
        urls = []
        for _ in range(count):
            url = self.generate_image(prompt=prompt, size=size, quality=quality, model=model)
            urls.append(url)
        return urls

View on GitHub (pinned to 848b054e4f)

Solutions

1. Log response.data[0] to see what fields the provider actually returns and adapt parsing.
2. Set response_format explicitly ('b64_json' is safest with relays) or update the client SDK so attributes match.
3. Switch to a relay/endpoint that returns standard OpenAI response shapes.
4. Rely on the built-in retry loop (it retries across models) but fix the endpoint for a permanent solution.

Example fix

// before
resp = client.images.generate(model="dall-e-3", prompt=prompt, n=1)

// after
resp = client.images.generate(model="dall-e-3", prompt=prompt, n=1,
                              response_format="b64_json")

# After the API call, before consuming the image:
img = response.data[0]
if not (getattr(img, 'url', None) or getattr(img, 'b64_json', None)):
    print('unexpected response item:', img)

def extract_image_payload(item):
    b64 = getattr(item, "b64_json", None)
    if b64:
        return ("b64", b64)
    url = getattr(item, "url", None)
    if url:
        return ("url", url)
    return None

try:
    out = client.generate_image(prompt)
except RuntimeError as e:
    if "未在响应中找到 url 或 b64_json" in str(e):
        out = client.generate_image(prompt, model="dall-e-3")  # try another model
    else:
        raise

When it happens

Trigger: The Images API response data[0] lacks both b64_json and url attributes/values — e.g. a relay returning a non-standard response object, or response_format mismatch (url requested but provider returns only b64, with b64 empty).

Common situations: Third-party OpenAI-compatible relays with non-standard response shapes, requesting response_format='url' from a model that only returns base64, SDK version changes altering response attribute names.

Related errors

• OpenAI API 返回数据为空
• Max attempts reached, failed to generate image. Last error:
• str(e)
• API image generation returned no result: provider={provider}
• DashScope generation failed: {e}

AI-assisted analysis of ATH-MaaS/Pixelle-Video@848b054e4f (2026-08-30). Data as JSON: /api/errors/fc9c43cb2b549308. Report an issue: GitHub.