Plain text prompts can't always describe what you want. "A modern living room in the style of THIS image" is much easier when you can show the image. KoalaImages supports up to 5 reference images per generation across vision-capable models.
How it works
- Upload reference images: drag and drop or paste up to 5 images.
- Write a text prompt describing what you want generated.
- Pick a vision-capable model: gpt-image-2 (Fast / Quality / Ultra) or Nano Banana 2.
- Generate: the model conditions output on your reference images plus the text prompt.
Common use cases
- Style transfer: "Generate a portrait in this art style" with the reference being a painting.
- Character consistency: keep the same character across multiple scenes by referencing prior generations.
- Scene matching: generate a hero image that matches the visual style of your existing brand assets.
- Product variations: show your product in different settings while keeping the product itself accurate.
- Visual brief expansion: paste a moodboard image, get original generated images that match its feel.
Supported formats
- JPEG
- PNG
- GIF (first frame used)
- WebP
Model selection
- GPT Image 2 Fast: 50 words / 1 credit. Cheapest path to reference-image generation.
- GPT Image 2 Quality: 200 words / 8 credits. Higher quality.
- GPT Image 2 Ultra: 20 credits (KoalaImages only). Premium quality with the strongest adherence to references.
- Nano Banana 2: 100 words / 5 credits. Alternative model with reference support.
Tips for best results
- Be specific in the text prompt about what to keep from the reference vs. what to change.
- For style transfer, write the subject in text + reference for style.
- For character consistency, reference the same character across generations; describe pose/setting in text.
- For product shots, reference the product itself; describe the environment.
- Use higher-quality models (Medium / Ultra) when adherence to references matters.
API
Reference images are passed to /api/image-generation/ as the referenceImages array. See the KoalaImages API docs.
Key Benefits
- Up to 5 reference images per generation
- JPEG, PNG, GIF, WebP supported
- Vision-capable models: gpt-image-2 + nano-banana-2
- Use for style transfer, character consistency, scene matching
- Combine with text prompt for fine-grained control
Related Features
Ready to try this feature?
Experience the power of Koala AI's platform with a free trial.
Get Started for FreeView Pricing Plans