Image Prompts
Attach reference images to text prompts—their role, the upload flow, and how many references you can use at once.
What image prompts do
Image prompts add visual context the model uses to guide:
- Subject composition — pose, placement, interaction with environment
- Style and materials — texture feel, grain, finish
- Character consistency — reference face or outfit across generations
Text prompts and image prompts work together. Image prompts are anchors; text refines intent.
How many references at once
The editor supports up to 14 reference images per generation—the same ceiling Google documents for Gemini 3 image workflows when blending many inputs.
How Google splits the 14 (API mental model)
Per Gemini image generation — up to 14 reference images, the roles differ by model tier:
| Model (doc names) | Object-like refs | Character refs |
|---|---|---|
| Gemini 3.1 Flash Image preview | Up to 10 high-consistency object images | Up to 4 character images |
| Gemini 3 Pro Image preview | Up to 6 high-fidelity object images | Up to 5 character images |
Use that table when planning who is identity-locked vs what is a prop or style anchor.
Practical counts in-app
| Use case | Recommended count |
|---|---|
| Single style reference | 1 |
| Character + scene reference | 2 |
| Style + lighting + subject | 3–4 |
| Multi-image synthesis | 5+ |
More references dilute unless each one is distinct and purposeful.
Rights and safety
Google’s editing section reminds you to only upload images you have rights to use and to avoid deceptive or harmful outputs. Same rules apply here: reference photos should be yours, licensed, or otherwise cleared.
Upload flow
- Click the image attach icon in the prompt area.
- Drag or browse to add images.
- Reorder if order matters (some features read first image differently).
- Add text to guide how references should blend.
What to check before you upload
- Crop to the key feature (face, product, texture) so noise is minimal.
- Use consistent lighting across multiple refs to reduce conflicts.
- Label refs internally if you share prompts with a team (for example “ref_a_front_product.jpg”).
Next
- Style reference for the most common use case.
- Character reference for portrait and identity work.