
AI image editing at original resolution up to 8K
Precise Edit is an image editing feature from Ideogram that uses a context window to make localized edits to high-resolution images (up to 8K) without downscaling. Edits are applied at original pixels within the context window, while everything outside stays untouched, preserving the source resolution with no seams or color mismatch.
Mark a region of an image and edit only that area at its original pixels. Everything outside the context window stays untouched, so the image keeps its original resolution with no seams and no color mismatch.
Edits images at any resolution up to 8K at their original pixels — demonstrated on images up to 7,360 × 4,121 px (~30.3 MP) — removing the need to downscale to edit or stitch crops back together with border and color artifacts.
Recolors products, repairs logos, replaces objects, and removes them — with examples across ecommerce, real estate, detail repair, interior design, and architecture.
Reference images can be sent alongside a mask and context window to guide an edit (e.g., replacing label text or swapping in a product from a reference photo); the reference image itself is never changed.
A context window can be specified manually (coordinates) or fitted automatically ('auto') around the edit area.
The context window renders small text, logos, and fine detail more sharply, such as repairing debossed lettering on hang tags or printed label text at original pixels.
Precise Edit is available via an API, with documentation provided.
Users send an image with a mask identifying the edit area, plus either a manually specified context window (e.g., coordinates like 1664,3900,2688,5436) or an auto-fitted context window. An optional reference image can also be provided to guide the edit.
Only the area inside the context window is edited — up to 2K in size — at its original pixels. The returned image is the same resolution as the source, with the edited region stitched back with no seams or artifacts.
It edits images at any resolution up to 8K, editing the targeted area at its original pixels. Examples on the page include sources from ~2K up to 7,360 × 4,121 px (~30.3 MP).
No. With the context window, only the window is edited at original pixels and the rest of the image stays untouched, so the result keeps the original resolution — previously, editing required scaling down (e.g., a 4K image came back at 2K).
You can edit any area up to 2K at its original pixels within the context window; everything outside the window remains unchanged.
The page states that the edited region is stitched back with no seams, no border artifacts, and no color mismatch, with everything outside the context window untouched.
Yes. Reference images can be sent (as reference_images) alongside a mask and context window to guide the edit — for example, to replace label text or swap in a specific product — and the reference image itself is never changed.
You send a mask identifying the edit area plus a context window, which can be given as explicit coordinates (e.g., 1664,3900,2688,5436) or fitted automatically using the 'auto' option.
It recolors products, repairs logos and text, replaces objects (matching pose and angle from a reference), and removes objects entirely — filling in the background behind them.
Yes, Precise Edit is available via an API, and API documentation is linked from the product page.