How to change one part of an AI-generated image
Updated · 5 min read
Short answer
An AI-generated image is one flat bitmap, so regenerating with a new prompt usually changes the parts you liked too. A steadier approach is to split the image into layers first: keep what works as is, move, remove, or redraw only the layer that needs it, then export PNG or PSD.
Why is it so hard to change one thing in an AI image?
Image models redraw the whole picture every time. Change one word or the seed and the composition, lighting, and details can all shift, so it is hard to touch only the part you meant to.
Inpainting can redraw what is inside a selected area, but it cannot move things. Nudging the product to the right, swapping the headline to the other side, or removing an element and filling the gap cleanly are outside what it does.
There is also the familiar problem of garbled or misspelled text in AI images, which regenerating does not reliably fix.
What you can do once it is in layers
- Move and resize: subject, headline, and decor are on separate layers, and the background is already filled, so nothing leaves a gap.
- Remove extras: hide or delete a layer and the rest of the image stays the same.
- Redraw one layer: select it, choose “Edit” (改图), and describe the change. The result sits right above the original, which is hidden so you can switch back any time.
- Add something new: “Generate asset” (生成素材) creates a transparent element from a prompt and drops it onto the image.
- Fix bad text: select the text layer and choose “Hide old text, write new” (隐藏旧字，原位写新字) to put new, editable text in the same spot.
Splitting an AI image in Delayer
Upload, or generate
Drop in the image you already generated. No image yet? Write a description on the canvas and generate one first.
Split with named elements
AI images tend to have lots of small details. Name the few elements you actually want to change, like “person, headline, flowers bottom left”, and leave the rest in the background for cleaner layers.
Edit only that layer
Double-click to open the full-screen editor. Move, scale, or hide layers, or select one and choose Edit to change how it looks.
Export
Export a flattened PNG, a layered PSD, or any single layer as a transparent PNG.
Regenerate, inpaint, or edit layers?
| Approach | Good for | Not great for |
|---|---|---|
| Regenerate | Exploring directions and seeing more options | Images you mostly like already |
| Inpainting | Replacing what is inside one area | Moving elements or changing the layout |
| Split and edit layers | Moving or removing elements, redrawing one element, extracting assets, fixing text | A completely new style or composition |
A few tips
- Split the original first, then make changes. The more you edit before splitting, the harder it is to tell where a problem came from.
- Filled-in background is inferred from its surroundings; large areas of fine texture may differ slightly, so check after moving the subject.
- Edit regenerates a layer using the original as a reference, so details will not match exactly. The original layer stays, so you can always switch back.
FAQ
Does it work with images from Midjourney, Stable Diffusion, and other tools?
Yes. Delayer only looks at the image itself. Any PNG, JPG, or WebP up to 30MB can be uploaded and split, whichever tool made it.
Does splitting reduce image quality?
Layers are output at the original size, and untouched areas match the original. Background fill and edges are produced by a model, so fine details may differ slightly—hide layers any time to compare.
How many credits do Edit and Generate asset use?
Generating, editing, and splitting are priced separately. Each button shows its cost before you submit, and failed jobs are refunded.