Free AI inpainting: paint it, describe it, done
Inpainting means repainting one part of an image while leaving the rest untouched, and this free AI inpainting tool makes it a two-step job: brush over the area, then write what should be there. A bare wall gets a framed painting, an empty table gets a vase of flowers, a cloudy patch of sky turns clear, a stain on a shirt disappears. It differs from the free AI object remover, where you describe an object in words and the AI finds and erases it anywhere in the picture. Here you decide exactly where the change happens, and you can add something new instead of only removing.
Leave the description empty and the tool acts as an eraser for the painted area, filling it with the background that naturally continues around it. That makes it a precise generative fill for the cases where an object remover would grab the wrong thing.
Typical jobs include adding a missing object to a product shot, replacing a distracting sign with plain wall, fixing a gap in a fence, giving a person a different hat, changing a single flower into a different variety, or filling a hole left after cropping out a photobomber. Each one comes down to the same two steps: paint the spot, then say what it should become.
How the generative fill keeps the rest of your photo intact
Qwen Image Edit, the open model that does the painting, edits a whole picture at about one megapixel and takes no mask of its own. So the tool works around it in your browser. It cuts out the painted area with some room around it for context, marks the painted part on that crop, asks the model to replace exactly that marked area with your description, and then pastes the model's new content back into your original over the painted area, with a thin soft edge just outside it so the two blend. Everything beyond that edge is your original file, pixel for pixel, at its full resolution.
Because only the crop goes to the model, a small edit on a large photo gets the model's full resolution instead of a shrunken copy of the whole frame. The only part that comes from the model is the painted area. A mask that covers most of a big photo is sent whole, and then the filled region can look slightly softer than its surroundings. Small and medium masks blend best.
The soft edge sits just outside the painted area, so whatever the model draws right up to your brush stroke stays crisp. It is a few pixels wide: enough to hide the seam, narrow enough that a face or a line of text next to the mask is barely touched. If something new looks cut off, paint the area slightly larger and run it again.
Tips for a clean image filler result
Paint slightly past the edges of what you want to change; a tight mask leaves a thin outline of the old content. Use the brush size slider for small details, and Undo for a stray stroke. Describe the new content plainly and include how it should sit in the scene, for example "a green potted plant on the floor, lit from the window". The model blends lighting and perspective on its own, but naming them helps.
Keep each fill to one idea. To change two separate areas, run the tool twice, downloading in between. For edits that touch the whole image, such as a new style or new lighting, the free AI image editor takes a freeform instruction, and to add new area beyond the borders of a photo, use the free AI image extender.
Some content is harder than others. Natural textures like grass, water, sky, and wood fill most convincingly, and simple objects like plants, lamps, and frames are reliable additions. Hands, faces, and readable text are the hardest things to paint into a small area; describe them simply and expect to run them more than once. Each run uses a fresh random seed, so repeated tries give you real variety to choose from.
Limits and privacy
The tool takes JPG, PNG, and WebP up to 4096 pixels per side and downloads the result as a PNG at your original size. Every visitor gets a daily allowance on shared free GPUs without an account, and connecting a free Hugging Face account raises it. Descriptions go through the same safety filter as every free prompt tool, and results are checked before they are shown. Images are processed transiently and deleted automatically.