What is img2img? Complete Guide to Image-to-Image AI Generation

Text-to-image gets all the attention, but img2img is often more useful in practice. Instead of generating from nothing, you start with an existing image and transform it. This guide explains exactly how img2img works and when to use it.

Quick Answer: img2img (image-to-image) is an AI generation mode where you provide an input image that gets partially noised, then denoised according to your text prompt. The result keeps structural elements from the original while applying new styles, details, or modifications. Denoise strength controls how much the AI changes: low values keep the original mostly intact, high values allow dramatic transformation.

What img2img Actually Does

Text-to-Image vs Image-to-Image

Text-to-image (txt2img):

• Starts from pure noise
• AI has total creative freedom
• Results entirely determined by prompt and seed
• No structural reference

Image-to-image (img2img):

• Starts from your input image
• AI works within existing structure
• Results guided by both image AND prompt
• Maintains composition, changes details

Think of txt2img as asking an artist to paint from imagination. img2img is showing them a rough sketch and asking them to finish it.

The Technical Process

1. Input image loaded: Your source image is processed
2. Noise added: Random noise partially covers the image (controlled by denoise strength)
3. Denoising: AI removes noise while following your prompt
4. Output: Result maintains structure from input but applies prompt guidance

The magic is in step 2: less noise means more original image preserved. More noise means more freedom for the AI to change things.

Understanding Denoise Strength

Denoise strength is the single most important img2img parameter. It controls how much the AI can change.

Denoise Strength Values

0.0-0.2 (Very Low):

• Minimal change
• Subtle style adjustments
• Colors shift slightly
• Structure completely preserved
• Use for: minor corrections, color grading

0.3-0.4 (Low):

• Noticeable but controlled changes
• Details refined
• Style applied gently
• Core composition intact
• Use for: gentle style transfer, refinement passes

0.5-0.6 (Medium):

• Significant transformation
• Major style application
• New details emerge
• Basic shapes preserved
• Use for: style transfer, artistic interpretation

0.7-0.8 (High):

• Dramatic changes
• Original barely recognizable
• AI has creative freedom
• Only rough composition preserved
• Use for: complete reimagining, major modifications

0.9-1.0 (Very High):

• Near-complete regeneration
• Original mainly affects composition
• Essentially txt2img with slight guidance
• Use for: variations, when you want something similar but different

Finding the Right Strength

Start at 0.5 and adjust:

• Result too similar to input? Increase strength.
• Result losing important details? Decrease strength.
• Sweet spot varies by use case and model.

Common img2img Use Cases

1. Style Transfer

Transform photos into different artistic styles.

Example workflow:

1. Input: Photo of a landscape
2. Prompt: "oil painting, impressionist style, vibrant colors"
3. Denoise: 0.5-0.7
4. Result: Photo composition with painted aesthetic

Best practices:

• Higher denoise for dramatic style change
• Include style keywords in prompt
• Test multiple denoise values

2. Image Refinement

Improve or fix generated images.

Example workflow:

1. Input: txt2img result with minor issues
2. Prompt: Same or enhanced prompt
3. Denoise: 0.3-0.5
4. Result: Refined version with fixes

Best practices:

• Lower denoise to preserve what works
• Focus prompt on areas needing improvement
• Multiple passes can compound improvements

3. Upscaling Enhancement

Improve quality while increasing resolution.

Example workflow:

1. Input: Low-resolution image (upscaled traditionally first)
2. Prompt: Detailed description of content
3. Denoise: 0.3-0.5
4. Result: Higher quality with added detail

This differs from simple upscaling, which can't add true detail. img2img actually generates new details appropriate to the content.

4. Character Variations

Create variations of characters or scenes.

Example workflow:

1. Input: Character image you like
2. Prompt: Modified description (different outfit, pose adjustment)
3. Denoise: 0.4-0.6
4. Result: Same character with requested changes

Best practices:

• Lower denoise preserves identity better
• Be specific about desired changes
• Use with LoRAs for better consistency

5. Photo to Illustration

Convert photos to illustrated styles.

Example workflow:

1. Input: Real photograph
2. Prompt: "anime style illustration" or "digital art"
3. Denoise: 0.6-0.8
4. Result: Illustrated version of photo

6. Sketch to Finished Art

Complete rough sketches or drawings.

Example workflow:

1. Input: Hand-drawn sketch or rough concept
2. Prompt: Detailed description of desired output
3. Denoise: 0.7-0.9
4. Result: Finished artwork following sketch structure

This is powerful for artists who sketch traditionally but want AI to assist with finishing.

img2img in Different Tools

ComfyUI

In ComfyUI, img2img is achieved through the standard workflow with modifications:

1. Load your input image
2. Encode it with VAE Encode
3. Connect to KSampler with appropriate denoise value
4. The denoise parameter directly controls transformation amount

Key nodes:

• Load Image: Brings in your source
• VAE Encode: Converts to latent space
• KSampler: denoise parameter is your strength control

For detailed ComfyUI workflows, see our ComfyUI workflow organization guide.

AUTOMATIC1111

In the img2img tab:

1. Upload image to input
2. Write prompt
3. Adjust "Denoising strength" slider
4. Generate

Apatero

Apatero.com provides img2img through the image input options. Upload your source image and adjust transformation strength in the interface.

Advanced img2img Techniques

Multi-Pass Refinement

Instead of one high-denoise pass, do multiple low-denoise passes.

Example:

• Pass 1: 0.3 denoise, general refinement
• Pass 2: 0.3 denoise, detail enhancement
• Pass 3: 0.2 denoise, final polish

Result: More controlled transformation than single 0.8 pass.

Progressive Enhancement

Start with high denoise, gradually reduce.

Example:

• Pass 1: 0.7 denoise, major style change
• Pass 2: 0.4 denoise, refine new style
• Pass 3: 0.2 denoise, final details

This applies dramatic changes while maintaining control.

Regional img2img

Different denoise for different areas.

In ComfyUI: Use masking to apply different denoise values to different regions. High denoise on areas you want changed, low on areas to preserve.

Use case: Changing a character's outfit while keeping their face identical.

Combining with ControlNet

img2img provides content, ControlNet provides structure.

Example workflow:

1. Input image sets base content
2. ControlNet (edge detection) locks composition
3. img2img transforms within those constraints

This gives maximum control over results.

img2img vs Inpainting

People often confuse these:

img2img:

• Transforms entire image
• Global changes
• Denoise affects everything

Inpainting:

• Transforms only masked region
• Selective changes
• Rest of image untouched

When to use each:

• Want to change overall style? img2img
• Want to fix specific area? Inpainting
• Want variation of whole image? img2img
• Want to add/remove something specific? Inpainting

Both are essential techniques. Learn more about inpainting in our AI image editing guides.

Common Problems and Solutions

Problem: Result Looks Nothing Like Input

Cause: Denoise too high

Solution: Reduce denoise to 0.4-0.5, increase gradually if needed

Problem: Changes Barely Visible

Cause: Denoise too low

Solution: Increase denoise, ensure prompt clearly describes desired changes

Problem: Faces Getting Distorted

Cause: Face area receiving too much noise

Solutions:

• Reduce overall denoise
• Use inpainting on face only
• Apply FaceDetailer after img2img
• Use lower strength for portrait work

Problem: Losing Important Details

Cause: Denoise erasing features prompt doesn't mention

Solution: Include all important features in prompt, even if keeping them. "A woman with red hair and blue eyes" if you want to keep those specific features.

Problem: Colors Going Wrong

Cause: Model interpreting prompt color differently than input

Solutions:

• Specify exact colors in prompt
• Use color-specific terms
• Try different denoise values
• Some models handle color better than others

Problem: Inconsistent Results

Cause: High denoise creates variation between generations

Solutions:

• Lower denoise for more consistency
• Lock seed for reproducible results
• Use ControlNet for structural consistency

Best Practices for Production Work

Start Conservative

Begin with lower denoise (0.3-0.4) and increase. Easier to add change than remove it.

Match Resolution

Input image resolution affects output. For best results:

• Use input resolution close to model's native resolution
• Or resize input to target dimensions first

Prompt Completeness

Unlike txt2img, img2img prompts should describe EVERYTHING in the final image, not just changes. The AI uses the prompt as guidance for what to preserve and what to modify.

Test Before Committing

Before processing hundreds of images:

1. Test on single image
2. Try 3-5 denoise values
3. Find optimal settings
4. Then batch process

Preserve Originals

Always keep original images. img2img is destructive to originals if overwritten. Work on copies.

img2img for AI Influencer Workflows

img2img is valuable for AI influencer content creation:

Outfit Changes

1. Generate base character image
2. img2img with clothing description changes
3. Low denoise (0.3-0.4) preserves face
4. Result: Same person, different outfit

Pose Refinement

1. Generate character in approximate pose
2. img2img to refine pose details
3. Medium denoise (0.4-0.5)
4. Result: Better pose, same character

Style Consistency

1. Take best character image
2. img2img new images to match style
3. Creates consistent aesthetic across images

For complete AI influencer workflows, see our comprehensive tutorial.

Frequently Asked Questions

What's the best denoise strength?

There's no universal best. 0.5 is a good starting point. Adjust based on how much change you want.

Can I img2img any image?

Yes, any image works as input. Quality and resolution affect results.

Does img2img work with all models?

Most diffusion models support img2img. Workflow details vary between interfaces.

Is img2img faster than txt2img?

Similar speed at same step count. Sometimes faster because lower denoise needs fewer steps.

Can I convert photos to specific character LoRAs?

Yes, img2img + character LoRA can transform photos while applying learned character features.

What resolution should my input be?

Match the model's native resolution or your target output resolution.

Why does my img2img look worse than the input?

Denoise too high, prompt doesn't describe important features, or model limitations. Try lower denoise.

Can I batch process img2img?

Yes, most interfaces support batch img2img with consistent settings.

Does seed matter for img2img?

Yes. Same seed + same settings = reproducible results. Different seeds create variations.

How is img2img different from ControlNet?

img2img uses image content directly. ControlNet extracts structural information (edges, depth) as guidance. They can be combined.

Wrapping Up

img2img transforms AI image generation from one-shot creation to iterative refinement. Understanding denoise strength gives you precise control over how much the AI changes.

Key concepts:

• Denoise strength controls transformation amount
• Lower denoise = more preservation
• Higher denoise = more creativity
• Prompt describes entire final image
• Multiple passes give more control than single high-denoise pass

Master img2img and you'll find yourself using it more than txt2img for production work. The ability to guide the AI with an existing image opens possibilities that text alone cannot achieve.

For hands-on practice, Apatero.com offers img2img functionality without local setup. For advanced workflows in ComfyUI, explore our ComfyUI guides.

Quick Reference: Denoise Settings

Next Steps

Once comfortable with basic img2img:

1. Combine with ControlNet for structure + content control
2. Learn inpainting for selective modifications
3. Explore multi-pass workflows for complex transformations
4. Use IPAdapter for reference-based generation alongside img2img

The img2img technique is foundational. Every advanced workflow builds on understanding how denoise strength affects the balance between input preservation and AI creativity.