# Ideogram 4.5 Precise Edit

Edit an image guided by a text prompt with Ideogram 4.5 precise editing: only what the prompt asks for changes, untouched pixels stay identical, and the output keeps the size of image 1. Image 1 is the image to edit and up to 4 further images can be connected as references. A single region of image 1 can also be edited on its own, selected by a mask or by bounding boxes, which returns the image at full size.

## Inputs

| Parameter | Description | Data Type | Required | Range |
|-----------|-------------|-----------|----------|-------|
| `model` | Model to use. (default: `"ideogram-4.5"`) | DYNAMIC_COMBO | Yes | `"ideogram-4.5"` |
| `images` | Growable slot: image 1 is the image to edit and images 2 to 5 are optional references (`image_1` ... `image_5`). Refer to them in the prompt as @Image1, @Image2, ...; a batched input counts once per image. Each image must have an aspect ratio between 1:6 and 6:1, except that image 1 is not aspect-checked while a context window is in use. | IMAGE | Yes | 1 to 5 images |
| `prompt` | Editing instructions. Supports @Image1-style references to the connected images. (default: empty string) | STRING | Yes | 1 to 10000 characters |
| `quality` | Quality tier. Higher tiers cost more and take longer. (default: `"medium"`) | COMBO | Yes | `"very_low"`<br>`"low"`<br>`"medium"`<br>`"high"` |
| `seed` | Seed for generation. The same images, prompt, settings and seed give the same result. (default: 42) | INT | Yes | 0 to 2147483647 |
| `context_window` | Region of image 1 to edit. `"none"` edits the whole image. `"auto"` edits a region around `context_window_mask` and `"custom"` the region from `context_window_bboxes`; both keep image 1 at full size and leave the rest of the image unchanged. (default: `"none"`) | COMBO | No | `"none"`<br>`"auto"`<br>`"custom"` |
| `context_window_bboxes` | Region of image 1 to edit, for example the output of Create Bounding Boxes. Required when `context_window` is `"custom"`; not used otherwise. Several boxes are merged into one region, which is grown to at least 256 pixels per side and an aspect ratio of 6:1. The region can be up to 4608 pixels per side and about 4 MP (e.g. 2048x2048). | BOUNDING_BOX | No | - |
| `context_window_mask` | White marks what to edit; the edit region is placed around it. Required when `context_window` is `"auto"`; not used otherwise. The mask must contain both white (edit) and black (keep) areas. | MASK | No | - |

### Parameter Constraints

- **Image count:** at least 1 and at most 5 images; a batched input counts once per image. Image 1 is the image to edit and images 2 to 5 are references. With `context_window_mask` at most 4 images are accepted, that is image 1 and 3 references.
- **Output size:** the result keeps the size of image 1, so this node has no size, width or height inputs. That also holds when a region is edited, where the full image comes back with only the region changed.
- **Image aspect ratio:** every image must be no wider than 6x its height and no taller than 6x its width (between 1:6 and 6:1). Image 1 is exempt while `context_window` is `"auto"` or `"custom"`, because only its selected region is sent.
- **Region source size:** with a context window in use, image 1 may be up to 8192x8192 pixels (about 67 MP); larger sources raise an error. Without a context window, images larger than about 4 MP or 4608 px on the long side are scaled down before they are sent.
- **Region limits:** a region from `context_window_bboxes` can be up to 4608 pixels per side and about 4 MP; boxes that lie entirely outside image 1 raise an error.
- **Prompt tags:** `@ImageN` is matched case-insensitively and must not exceed the number of connected images; the prompt must be non-whitespace and at most 10000 characters.
- **Upload limit:** image 1, the mask and the references together must fit in Ideogram's 50 MB upload; image 1 is re-encoded as JPEG when needed and the node raises an error if it still does not fit.
- **Content safety:** if Ideogram's content safety filter blocks the result, the node raises an error instead of returning an image.

## Outputs

| Output Name | Description | Data Type |
|-------------|-------------|-----------|
| `IMAGE` | The edited image as a batch, at the size of image 1. | IMAGE |

> This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! [Edit on GitHub](https://github.com/Comfy-Org/embedded-docs/blob/main/comfyui_embedded_docs/docs/IdeogramPreciseEditApi/en.md)

---
**Source fingerprint (SHA-256):** `84f042739a8d7bc1a4b268f6175be6a598b67d32b9311fdf204a923472f5fcc9`
