Skip to content

image-gen.md

Latest commit

 

History

History
142 lines (113 loc) · 6.98 KB

image-gen.md

File metadata and controls

142 lines (113 loc) · 6.98 KB

exhub-image-gen

The exhub-image-gen module provides MCP-based AI image generation using the Gitee AI image generation API (OpenAI-compatible).

Setup

Configuration

The image gen server requires a Gitee AI API key stored in SecretVault:

mix scr.insert dev giteeai_api_key "your-gitee-ai-api-key"

If you already configured giteeai_api_key for exhub-web-tools, no additional setup is needed — the same key is shared.

The server starts automatically with the Exhub application.

MCP Endpoint

POST /image-gen/mcp

Tool: image_gen

Generate a high-quality image from a text description.

Parameters

Parameter Type Required Default Description
prompt string ✓ — Text description of the image to generate. Be specific and detailed.
model string Qwen-Image Model to use (see table below)
size string 1024x1024 Output image dimensions (see sizes below)
negative_prompt string Standard quality negative prompt Elements to avoid. Not supported by Kolors.
guidance_scale float Model default How closely the model follows the prompt. Not supported by Qwen-Image.
num_inference_steps integer Model default Denoising steps. Higher = better quality but slower.

Supported Models

Model negative_prompt guidance_scale num_inference_steps Default steps Default scale
Qwen-Image (default) ✓ — ✓ 30 —
Qwen-Image-2512 ✓ — ✓ 30 —
Kolors — ✓ ✓ 25 7.5
GLM-Image ✓ ✓ ✓ 30 1.5
FLUX.1-schnell — ✓ ✓ 4 0.0
FLUX.1-dev ✓ ✓ ✓ 28 3.5
FLUX.1-Krea-dev ✓ ✓ ✓ 28 3.5
FLUX.2-dev ✓ ✓ ✓ 20 7.5
FLUX.2-klein-9B — ✓ ✓ 8 3.5
FLUX.2-klein-4B — ✓ ✓ 8 3.5
HunyuanDiT-v1.2 ✓ ✓ ✓ 25 5.0
stable-diffusion-xl-base-1.0 ✓ ✓ ✓ 30 7.5
stable-diffusion-3.5-large-turbo ✓ ✓ ✓ 8 1.0
stable-diffusion-3-medium ✓ ✓ ✓ 28 7.0
CogView4-6B ✓ ✓ ✓ 50 7.5
HiDream-I1-Full ✓ ✓ ✓ 50 7.0
z-image-turbo — ✓ ✓ 8 3.5
Z-Image ✓ ✓ ✓ 28 5.0
LongCat-Image ✓ ✓ ✓ 28 5.0

Supported Sizes

Size Aspect Ratio
256x256 1:1 small
512x512 1:1
1024x1024 (default) 1:1
1024x576 16:9 landscape
576x1024 9:16 portrait
1024x768 4:3 landscape
768x1024 3:4 portrait
1024x640 16:10 landscape
640x1024 10:16 portrait
2048x2048 1:1 high-res

Response Format

{
  "image_url": "https://ai.gitee.com/...",
  "model": "Qwen-Image",
  "size": "1024x1024",
  "prompt": "a cat sitting on a mountain at sunset",
  "params": {
    "response_format": "url",
    "negative_prompt": "...",
    "num_inference_steps": 30
  }
}

Display the image with markdown: ![Generated Image](image_url)

Usage Examples

Basic generation (default model)

{
  "prompt": "a serene mountain lake at golden hour, photorealistic, 8k"
}

Kolors with custom guidance

{
  "prompt": "a futuristic city skyline at night, neon lights, cyberpunk style",
  "model": "Kolors",
  "size": "1024x576",
  "guidance_scale": 10.0,
  "num_inference_steps": 28
}

FLUX.2-dev portrait

{
  "prompt": "portrait of a young woman with flowing red hair, soft studio lighting, detailed",
  "model": "FLUX.2-dev",
  "size": "768x1024",
  "negative_prompt": "blurry, low quality, distorted face",
  "num_inference_steps": 30,
  "guidance_scale": 9.0
}

Quality Tuning Tips

Problem Solution
Image quality is low or lacks detail Increase num_inference_steps (e.g. 25 → 35)
Image ignores prompt details Increase guidance_scale (e.g. 7.5 → 15)
Image is oversaturated or distorted Decrease guidance_scale
Need a specific aspect ratio Choose the appropriate size
Want to avoid specific elements Use negative_prompt (where supported)