Image to image: controlled variants from a reference photo
If you need image to image editing, start by uploading a reference, then pick a model and settings. The workspace changes style, backgrounds, or blends references from pixels you already have — not a blank text-only canvas.
Reference count by model (up to 10, 8, or 1)
Nano Banana is built for fast multi-reference trials — up to 10 images. Nano Banana Pro and Flux Kontext take up to 8 for finer edits. GPT-4o Image accepts only one reference; use it when a single strong photo plus a detailed prompt is enough. Do not force a multi-image batch into GPT-4o Image.
Formats: JPEG, PNG, WEBP in — flexible ratios out
Uploads accept JPEG, PNG, and WEBP. Aspect ratio and resolution options follow the selected model (Nano Banana family covers many ratios; some models offer 1K/2K/4K). Pick the output ratio before you generate to avoid a second crop.
Models and credits: trade speed for detail
- Nano Banana — 3 credits, fast image-to-image drafts
- Flux Kontext — 4 credits, flexible restyling
- GPT-4o Image — 6 credits, strong prompt following on one reference
- Nano Banana Pro — from 6 credits (12 at 4K) for higher detail
Three steps for one image-to-image run
- Upload references within the current model’s file limit, using JPEG, PNG, or WEBP.
- Describe what should change and what must stay; pick aspect ratio and resolution.
- Generate, compare, then tighten the prompt or switch models for the next pass.
Image to image FAQ
How is image to image different from text-only generation?
Image to image starts from a photo you upload, so composition and subject stay closer to the source while your prompt changes style or background. Text-only generation invents the scene from words alone.
How many reference images can I upload?
It depends on the model: Nano Banana accepts up to 10; Nano Banana Pro and Flux Kontext up to 8; GPT-4o Image accepts only 1. Check the selected model before uploading a batch.
Which image formats are supported?
Upload JPEG, PNG, or WEBP references. Outputs use the high-quality formats offered in the playground for download and further editing.
How many credits does each model use?
Nano Banana costs 3 credits per run; Flux Kontext 4; GPT-4o Image 6; Nano Banana Pro 6 by default (12 at 4K). Always trust the cost shown in the workspace.
How do I keep a person or product consistent?
Use a sharp, well-framed reference. State what must stay (face, bottle silhouette, logo) in the prompt, and avoid changing too many elements in one pass.