Image to image AI, by typing.

Image to image means you start from a photo instead of a blank page. You upload a picture, type the change you want in plain English, and the model generates a new image built on your original — same scene, new instruction applied. It is the same idea Stable Diffusion users call img2img, without the settings panel.

There is no denoise slider, no strength dial, no node graph. The instruction is the control surface: say what should change and what should stay. “Make it a cyberpunk night scene” rebuilds the mood of the whole frame; “change only the jacket to red leather” leaves everything else alone. The more specific the sentence, the closer the result lands to what you pictured.

It runs online in the browser — nothing to install. Uploads can be JPG, PNG, or WebP, up to 4 photos in one edit, which is how you combine references or apply one photo’s look to another. Results usually arrive in about a minute; complex prompts can take a few minutes. And the download carries no watermark, free tier included — the same policy as our AI image generator with no watermark.

There are three steps and no settings to learn. The editor takes an image and an instruction, and everything you would normally control with sliders — strength, style, composition — you control by being specific in the sentence you type. Here is the whole flow, from upload to download.

1. Upload a photo. Open the AI photo editor and add your picture — JPG, PNG, or WebP. You can add up to 4 photos when the edit needs references.
2. Type the instruction. One plain sentence: the style you want, the background to swap, the object to remove. Name what must stay the same if it matters — “keep the pose and framing”.
3. Generate and download. The result arrives in about a minute — a few minutes for complex prompts — and the download is the full-resolution file, no watermark, no signup.

Both pairs below were made the way you would make them: one uploaded photo, one typed instruction, one result. The left image is the original upload; the right is the download. The captions under the results are the exact instructions, unedited — copy either one into the editor.

Three families of edits cover most of what people type into the editor: full restyles that redraw the whole photo in a new medium, targeted edits that change one thing and leave the rest, and background or scene swaps that move the subject somewhere else. All three take one upload and one sentence.

Restyles are the classic image-to-image move — watercolor, oil paint, anime, film photography. The walkthrough in turn a photo into another style shows how to phrase them and what each medium does to detail, and the restyle prompt library has copy-paste instructions with before-and-after examples.

Targeted edits and swaps read the same way: remove the crowd in the background, change the t-shirt to a navy suit, put the subject on a beach at golden hour. If you would rather start from a blank page than a photo, that is a job for the AI image generator — this page is for images you already have.