Question 1

How is this different from Remove Watermark?

Accepted Answer

Technically, it's the same underlying pipeline (paint a mask, send to gpt-image-1 edit). The difference is intent: /watermark-remover uses a fixed, expert-crafted prompt optimized for erasing overlays, while this tool lets you write your own prompt to describe any edit. If you know you want to remove a watermark, use that tool — you don't have to think. If you want to change a sofa color, add an object, or replace a background, use this tool and describe the change.

Question 2

What kinds of edits does this work well for?

Accepted Answer

Very well: object removal (with clean backgrounds to infer from), color changes, clothing swaps, adding objects that fit naturally in the scene, background replacements, lighting adjustments on a region. Less well: very precise geometric edits (moving something by exact pixels), perfect color matches across large regions, or edits that require understanding content outside the masked area that the AI can't infer.

Question 3

Does the prompt need to be long and detailed?

Accepted Answer

Not necessarily, but detail helps. 'Remove this' is a valid prompt and often works. 'Replace with a mid-century modern armchair in burnt orange linen, matching the room's warm afternoon lighting' will give you much more control over the result. The example prompt pills show you what kind of detail works — copy one and tweak it for your image.

Question 4

Why do I need to paint a mask at all?

Accepted Answer

Because this is a LOCAL edit, not a global one. Without a mask, gpt-image-1 would rewrite your entire image based on the prompt — you'd lose everything, including the parts you wanted to keep. The mask tells the model 'only change these pixels, preserve everything else exactly'. If you want full-image regeneration from a prompt, use the main Studio at /create — that's what it's built for.

Question 5

How much does each run cost?

Accepted Answer

Each run costs 4 credits (~$0.06 on a Basic plan). That covers the real gpt-image-1 edit provider cost plus a small margin. New accounts get 10 credits on signup so you can try the tool twice. For frequent use, a subscription plan bundles credits at a lower effective rate.

Question 6

Can I chain multiple edits on the same image?

Accepted Answer

Yes — after a successful edit, click Start over to go back to upload mode, re-upload the result image from the download, and run another edit. Each run is independent so you can iteratively refine an image (remove an object, then change a color, then add something) without losing any detail across passes.

Question 7

What if the edit doesn't match my prompt?

Accepted Answer

Usually this means one of three things: (1) your prompt was too vague — try being more specific about materials, colors, lighting, and style; (2) your mask was too small to give gpt-image-1 room to work — try expanding the mask area; or (3) the AI model genuinely struggles with the exact concept you want. For (3), try rephrasing the prompt from a different angle, or open the main Studio for full-image generation instead.

Question 8

Are my uploads private?

Accepted Answer

Yes. Your uploads and generated edits are stored in your private account workspace, never shared or used for training. OpenAI receives the image + mask + prompt to process the edit but doesn't retain them per their API data policy. Delete any upload from your account at any time.

涂抹区域，描述改动

Generative fill for any image

Pixel-precise mask

Plain-English prompts

Lossless non-masked regions

Works on anything

Three steps from idea to edit

Upload your image

Paint the edit region

Describe and generate

常见问题解答

想要更精细的控制？