Precise Edit with context window
Make surgical edits with context window in images up to 8K. You no longer have to downscale to edit, or stitch crops back with border and color artifacts.
You can edit any area up to 2K at its original pixels. Everything outside the context window stays untouched, so your image keeps its original resolution, with no seam and no color mismatch.
With and without a context window
The context window also improves editing quality in fine detail.
Without a context window
The 4K original is scaled down to edit, so the result comes back at 2K.
With a context window
Only the window is edited, so the result keeps the original 4K resolution.
Prompt
Repair the lowercase carhartt lettering on the woven label so every letter is separate and legible in dark brown thread, like the reference image.
Reference images


Localized edits in Precise Edit
Precise Edit recolors products, repairs logos, replaces objects, and removes them.
Ecommerce


4,016 × 6,016 px · 24.2 MP · full image


5,504 × 3,072 px · 16.9 MP · full image


4,096 × 4,096 px · 16.8 MP · full image


4,000 × 4,000 px · 16.0 MP · full image


5,504 × 3,072 px · 16.9 MP · full image


3,584 × 4,480 px · 16.1 MP · full image


3,584 × 4,480 px · 16.1 MP · full image
Real estate


5,504 × 3,072 px · 16.9 MP · full image


2,496 × 1,664 px · 4.2 MP · full image


5,056 × 3,392 px · 17.1 MP · full image


7,360 × 4,121 px · 30.3 MP · full image


5,504 × 3,072 px · 16.9 MP · full image
Fix detail


2,048 × 2,048 px · 4.2 MP · full image


2,048 × 2,048 px · 4.2 MP · full image


2,048 × 2,048 px · 4.2 MP · full image


3,712 × 4,608 px · 17.1 MP · full image


3,712 × 4,608 px · 17.1 MP · full image
Interior design


5,504 × 3,072 px · 16.9 MP · full image


5,504 × 3,072 px · 16.9 MP · full image


5,504 × 3,072 px · 16.9 MP · full image


5,504 × 3,072 px · 16.9 MP · full image


5,504 × 3,072 px · 16.9 MP · full image


5,504 × 3,072 px · 16.9 MP · full image
Architecture


14,495 × 4,568 px · 66.2 MP · full image


5,504 × 3,072 px · 16.9 MP · full image


5,056 × 3,392 px · 17.1 MP · full image


5,056 × 3,392 px · 17.1 MP · full image
API
One field sets where the model works.
context_window names a region of the image. The model edits that region at its
original pixels and places it back, and every pixel outside it returns exactly as sent. A
mask inside the window narrows the edit to the pixels that may change.
-
Precise Edit
POST /v2/image/precise-edit/ideogram-4-5Send the image and a prompt, with a mask, a context_window, or both. The response is the full image at its own size.
-
The context window
-
context_window=none - Edit the whole image. Over 2048 × 2048 pixels of area, it comes back scaled down to fit. This is the default.
-
context_window=auto - Fit the window around the mask, so the edit keeps the image's full resolution.
-
context_window=y_min,x_min,y_max,x_max - Set the window yourself, by its corner pixels, rows first. Each side is at least 256 px, the area is at most 2048 × 2048, and the aspect ratio stays within 6:1.
-
Price per image (USD)
| Quality | Price |
|---|---|
| Very low | $0.008 |
| Low | $0.03 |
| Medium | $0.06 |
| High | $0.22 |
Frequently asked questions
What is Precise Edit in Ideogram?
Does editing one detail change the rest of my image?
What resolution do I get back?
context_window=auto or by naming the coordinates yourself, the result is returned at the width and height you uploaded. The default none mode behaves differently: it edits the whole image and downscales a large one to fit, so the result comes back smaller than you sent it.Do I need to supply a mask?
context_window=auto you send a mask and Ideogram fits the working region around it. The mask is the same width and height as your image, black over the area to change and white everywhere else. If you name the region by coordinates instead, a mask is optional, but if you send one its black area must sit inside that region.How small an area can I edit?
What is the difference between the mask and the context window?
Can I use Precise Edit from the API?
context_window field. See the API reference for the full request and response shape.Precise Edit






