Skip to main content

What are reference images?

Reference images are photos or illustrations you upload to guide the AI during image generation. They help the AI understand:
  • Style: Artistic style, color palette, mood
  • Composition: Layout, framing, perspective
  • Subject details: Character appearance, object features, textures
  • Atmosphere: Lighting, environment, overall feel

When to use reference images

Reference images are particularly useful for:
  • Character consistency: Maintaining the same character across multiple images
  • Style matching: Replicating a specific artistic style
  • Composition guidance: Achieving a particular layout or framing
  • Color palette: Matching specific color schemes
  • Texture and detail: Preserving specific textures or details

Uploading reference images

1

Prepare your image

Choose a clear, high-quality reference image:
  • Good resolution (at least 512x512 pixels recommended)
  • Clear subject matter
  • Relevant to your prompt
  • Appropriate file format (JPEG, PNG)
2

Upload in the interface

In the image generation interface:
  1. Open the References panel in your session
  2. Drag-and-drop or click to upload (up to five stored per session)
  3. Wait for upload to complete
  4. Verify the image appears and select up to three before generating
3

Write your prompt

Combine your reference image with a descriptive prompt:
  • Describe what you want to change or modify
  • Specify elements to keep from the reference
  • Add details not present in the reference
4

Generate

Click generate. The AI will use both your prompt and reference image to create the result.

Best practices

Choosing reference images

High quality: Use clear, well-lit reference images for best results.
Relevant content: Choose references that match what you’re trying to create.
Simple compositions: References with clear, uncluttered compositions work best.

Writing prompts with references

Be specific: Describe what to keep from the reference and what to change.
Mention the reference: Explicitly call out elements you want preserved.
Add new details: Use the prompt to add elements not in the reference image.

Use case examples

Character consistency

Goal: Generate multiple images of the same character
  1. Upload a clear character reference
  2. Use prompts like: “Same character in [new scenario], maintaining appearance from reference”
  3. Generate multiple variations

Style matching

Goal: Match a specific artistic style
  1. Upload an image in the desired style
  2. Prompt: “New subject in the style of the reference image”
  3. The AI will adapt the style to your new subject

Composition guidance

Goal: Achieve a specific layout
  1. Upload reference with desired composition
  2. Prompt: “Different subject but same composition and framing as reference”
  3. Generate with new subject in similar layout

Limits and behavior

  • Uploads per session: Store up to five reference images in a session.
  • Selections per run: Attach up to three references to a single generation request (platform limit).
  • Reusing outputs: Recently generated images appear alongside uploads so you can immediately reuse them as references.
  • No strength sliders: References act as guidance only; there are no blending or weight controls in the UI.

Limitations and considerations

Not exact copying: Reference images guide the AI but don’t produce exact copies. The AI interprets and adapts the reference.
Copyright: Only use reference images you have rights to use. Don’t use copyrighted material without permission.
Quality matters: Low-quality or unclear reference images may not produce good results.

Tips for better results

Combine multiple references: Use up to three to balance pose, style, or composition cues.
Iterate: Try different reference images or combinations to find what works best.
Use with prompts: Reference images work best when combined with descriptive prompts.

Troubleshooting

  • Verify the image uploaded successfully
  • Check that the model supports reference images
  • Ensure your prompt references the image
  • Try a more specific prompt
  • Use a clearer reference image
  • Adjust reference strength if available
  • Experiment with different models
  • Check file format (JPEG, PNG supported)
  • Verify file size isn’t too large
  • Try a different image
  • Check your internet connection