Img2Img Techniques
Image-to-image generation - using an existing image as a starting point for AI generation. The output is influenced by both the input image and the text prompt.
Img2img uses a reference image to guide generation, giving you more control over composition and structure than text-to-image alone.
How It Works
- Input image is encoded to latent space
- Noise is added based on denoising strength
- Model denoises guided by your prompt
- Result combines image structure with prompt guidance
Denoising Strength
The key parameter controlling input influence:
| Strength | Result |
|---|---|
| 0.0-0.3 | Minor style changes, keeps structure |
| 0.3-0.5 | Moderate changes, recognizable source |
| 0.5-0.7 | Significant changes, some structure |
| 0.7-1.0 | Major transformation, little from source |
Use Cases
Style Transfer
Convert photos to paintings, anime, etc.
- Use low-medium denoising
- Describe target style in prompt
Sketch to Image
Turn rough drawings into finished art
- Medium-high denoising
- Describe desired details
Photo Enhancement
Improve or modify photos
- Low denoising for subtle changes
- Specific prompts for modifications
Iteration
Refine AI-generated images
- Feed output back as input
- Gradually improve with each pass
Tips
- Start with lower denoising and increase
- Match prompt to what’s in the image
- Resolution should match input image
- Combine with ControlNet for precise control