linkfox-aigc-imagegen

AI image generation

Generate images based on prompts and reference images, support multiple AI models, create tasks asynchronously and poll for results.

Core Features

• Multiple model options: 7 models available, each with different strengths and resolution support.
• Image-to-image: pass in a reference image + prompt to generate a new image.
• Batch output: generate up to 10 images in a single run.
• Asynchronous mode: creating a task immediately returns a taskId, and the script automatically polls until completion (10-minute timeout).

Model Description

Model (provider)	Description	Features
BANANA	LFBanana (basic version)	Only supports 1K resolution
BANANA_2	LFBanana2	Supports higher resolution
BANANA_PRO	LFBanana Pro (default)	Best overall results
GPT_2_IMAGE	Img2	Supports the quality parameter to control quality
AIDRAW_EDIT	Linkfox-Image-1	Suitable for editing scenarios
WAN2_7	Wan 2.7	Wanxiang model
SEEDREAM5	Seedream 5.0	Only supports 2K

Parameter Overview

• Required fields：prompt、imageUrls、outputNum(default 1）、resolution(default 1K）、quality(default high, only GPT_2_IMAGE)

For the complete parameter table, response field structure, and error codes, see references/api.md。

How to Call

• Create task：POST /aigc/imageGenAsync → returns {taskId}
• Poll result：POST /aigc/taskQuery → pass in {taskId} → returns status and result
• Python script：python scripts/aigc_imagegen.py '<JSON 参数>'(the script automatically completes creation + polling internally)

Asynchronous flow：

1. The script calls /aigc/imageGenAsync to create a task and obtain taskId
2. Poll /aigc/taskQuery, initial interval 10 seconds, decreasing by 1 second each time to a minimum of 5 seconds, maximum wait 10 minutes
3. After success, automatically download the image to the session media/ directory

File storage path convention(follows the CLAUDE.md specification):

All files are stored under <cwd>/linkfox/<YYYY-MM-DD>/<session>/ , divided into directories by type:

Content type	Directory	Path format	Description
Generated image	media/	<session>/media/linkfox-aigc-imagegen-<ts_ms>.<ext>	such as media/linkfox-aigc-imagegen-1718000000000.png
Raw API response	data/	<session>/data/linkfox-aigc-imagegen-<ts>.json	contains the complete taskId, status, temporary URL, etc.

Output strategy (default script behavior)：

1. On successful image generation(regardless of how many):

   • stdout outputs Saved full response: <路径数组> format lines, for example:
     • 1 image:Saved full response: ["/path/to/media/a.png"]
     • Multiple images:Saved full response: ["/path/to/media/a.png", "/path/to/media/b.png"]
   • The raw API response (including temporary URLs, etc.) is silently written to disk at data/。
   • The frontend UI can parseall Saved full response: lines and render the subsequent JSON arrays.
   • Each script call usually outputs one line; when called multiple times, there will be multiple lines.

2. When there is no image output(failure/no result):

   • stdout outputs the raw response path:Saved full response: /path/to/data/xxx.json

Data reading recommendations：Do not Read the content of the image file saved by Read (to avoid base64 entering the context). Simply provide the local path after Saved full response: to the user.

Resolve authentication and compute quota issues

When the following exceptions occur, use references/onboarding.md to guide problem resolution:

Exceptions

• API Key not configured: The environment variable is not configured LINKFOX_AGENT_API_KEY, and LINKFOXAGENT_API_KEY。
• Response 401 or 402 status code
• Response indicates insufficient compute quota or balance: The message contains "insufficient compute quota/insufficient billing/insufficient balance/quota exceeded/insufficient balance/package expired/recharge required/please recharge", or content with similar meaning.

Usage guidelines

1. Model selection: For general scenarios use BANANA_PRO(default); for high-quality image quality control use GPT_2_IMAGE; to edit existing images use AIDRAW_EDIT。
2. Prompts: The more specific the description, the better the result. English prompts are recommended for better results.
3. Reference image: At least one reference image URL must be provided.
4. Resolution: Supports 1K/2K/4K, default 1K.

Examples

1. Generate a product scene image with the default model

{"imageUrls": ["https://example.com/product.jpg"], "prompt": "product photography on white background, studio lighting", "provider": "BANANA_PRO", "outputNum": 1, "resolution": "1K", "aspectRatio": "1:1", "quality": "high"}

2. Generate high-quality images with GPT_2_IMAGE

{"imageUrls": ["https://example.com/ref.jpg"], "prompt": "modern minimalist product showcase", "provider": "GPT_2_IMAGE", "outputNum": 1, "resolution": "1K", "aspectRatio": "16:9", "quality": "high"}

3. Generate multiple images in batch

{"imageUrls": ["https://example.com/ref.jpg"], "prompt": "creative product ad design", "provider": "BANANA_PRO", "outputNum": 4, "resolution": "2K", "aspectRatio": "1:1", "quality": "high"}

Limitations

• At least one reference image URL must be provided.
• Generate up to 10 images per request.
• Generation time varies by model and resolution, typically 10-120 seconds. The script automatically polls, with a 10-minute timeout.
• Do not retry on failure (only 1 attempt).

Not applicable

Not applicable：

• Text-only generation → linkfox-aigc-textgen
• Video generation → linkfox-aigc-videogen

Feedback

See references/api.md。

Execute instructions

Important: After receiving the ARGUMENTS: field at the end of this document, you must immediately call python scripts/aigc_imagegen.py '<JSON>' to execute,do not wait for user confirmation or additional input。