Portrait photo restyled as a textured oil painting
Image to Image AI (img2img)
Upload a photo, say what should change and what must stay, and an instruction-following image model returns the edited or restyled version - up to 4 results per run.
Who uses image to image, and for what?
Designers, marketers, ecommerce teams and photographers use image to image when the composition already exists: a product shot that needs a new backdrop, a portrait in another medium, a sketch that needs finishing. Text to image is the blank-canvas alternative.
Restyle
Keep the subject and layout, change the medium: photo to watercolor, anime, oil paint or 3D render. Name what must not move, such as the face or the pose, because the model keeps only what the prompt protects.
Transformations You Can Run
Real image-to-image moves - anchor a source image and let the prompt push it somewhere new.
Same subject, new medium
Photo to Art Style
Turn a plain photo into anime, watercolor, oil painting, or 3D render while keeping the original composition intact.
Rough lines to finished art
Sketch to Render
Feed a rough sketch and let the generator fill in lighting, texture, and detail to produce a polished render.
Same product, new setting
Backdrop and Label Swaps
Ask GPT Image 2 or Nano Banana Pro to replace the backdrop or the text on a package and keep the product. Check edges and small print in the result: areas the prompt did not mention can still change.
How image to image works
Upload a source image, write what should change and what must stay, pick a model that takes as many references as you have, then generate up to 4 results and compare them.
1. Upload an Image
Your starting point
Drop in a JPEG, PNG or WebP file. Most models accept up to 10 MB per image; GPT Image 2, GPT Image 2.5 and Nano Banana 2 Lite accept up to 30 MB.
2. Write a Prompt
Say what changes and what stays
Name the change ('make it a watercolor', 'replace the sky with a sunset') and the parts to keep ('same face, same pose'). The source anchors the result only as far as the prompt protects it.
3. Pick a Model
One reference or ten
Flux Kontext Pro and Midjourney take one image, Nano Banana and Nano Banana Pro take 4, and Seedream, GPT Image 2 and Nano Banana 2 Lite take up to 10. Nano Banana, Nano Banana 2, Nano Banana 2 Lite and WAN 2.7 Image are on every plan.
4. Generate & Compare
Iterate until it lands
Run up to 4 results at once, keep the closest, and refine the prompt instead of re-uploading. The form shows the credit cost before each run.
What a run costs and returns
Available in the form today, each priced in the same credits.
Per run, depending on the model and length chosen.
Includes a commercial licence for what you make.
The longest prompt every model on this page accepts.
Image to Image FAQ
What img2img means, why there is no denoising slider here, which files you can upload, free credits and installs - the essentials before you upload your first image.
Transform Your First Image
Drop in an image, say what should change and what must stay, and compare up to 4 results side by side - no install, no GPU.