Ideogram 4.5 in ComfyUI: Text to Image and Edit

Run the Ideogram 4.5 partner nodes in ComfyUI for text to image, image edit, and precise edit that keeps untouched pixels identical

Ideogram 4.5 is the newer model in the family and Ideogram’s most precise edit model: it changes only what the prompt asks for, so repeated edits keep colors, textures, and details in place. It ships in ComfyUI as three partner nodes that are billed per image through the same Comfy account as the rest of the partner nodes. Ideogram 4.0 keeps its own page: Ideogram 4.0 in ComfyUI.

To use the Partner Nodes, you need to ensure that you are logged in properly and using a permitted network environment. Please refer to the Partner Nodes Overview section of the documentation to understand the specific requirements for using the Partner Nodes.

Local users

Make sure your ComfyUI is updated.

• Download ComfyUI
• Update Guide

Workflows in this guide can be found in the Workflow Templates. If you can’t find them in the template, your ComfyUI may be outdated.

If nodes are missing when loading a workflow, possible reasons:

1. You are not using the latest ComfyUI version (Nightly version)
2. Some nodes failed to import at startup

Cloud users

• Cloud will update after ComfyUI stable release.

So, if you find any core node missing in this document, it might be because the new core nodes have not yet been released in the latest stable version. Please wait for the next stable release.

Text to Image

Generate an image from a text prompt, at a canvas size and quality tier you choose.

Run on Comfy Cloud

Open in Comfy Cloud

Download Workflow

Download JSON or search “Ideogram 4.5: Text to Image” in Template Library

The Ideogram 4.5 Text to Image node takes:

Input	Values	Effect
prompt	up to 10000 characters	The image description. A structured JSON caption works here too, for example a final_prompt from an earlier run
size	auto or a 1K or 2K preset	auto lets the model pick a canvas that suits the prompt. Presets cover both 1K and 2K and run from 1:1 to 23:9, for example (2K) 2048x2048 (1:1) and (1K) 1280x720 (16:9)
quality	low, medium (default), high	Higher tiers cost more and take longer
magic_prompt	auto (default), on, off	Rewrites the prompt into a detailed structured caption before generating. off keeps your wording as literal as possible
seed	integer	Text to image is not reproducible from the seed alone, because the prompt is rewritten on every run

The node also outputs final_prompt, the structured caption the image was generated from. To reproduce an image, feed that caption back with magic_prompt set to off and the same seed.

Image Edit

Edit or combine up to five images with one instruction. image_1 is the image being edited; images 2 to 5 supply design, material, or color references. Address them in the prompt as @Image1, @Image2, and so on.

Run on Comfy Cloud

Open in Comfy Cloud

Download Workflow

Download JSON or search “Ideogram 4.5: Image Edit” in Template Library

Input material

interior_orange_sofa.png

Load into the image_1 slot on the Ideogram 4.5 Edit node

The template builds its instruction with a Build JSON Prompt node, so you can mark a region of the image and caption it, for example “remove it”. Turn on Nodes 2.0 if the editor does not appear.

The Ideogram 4.5 Edit node adds:

Input	Values	Effect
size	auto (default), source, a preset, or custom	auto picks a canvas of about 2K from the images and the prompt. source keeps the size of image_1. A preset with a different aspect ratio recomposes the scene. custom uses the width and height below
width, height	256 to 4608, step 32	Used only when size is custom. The two dimensions together must stay within 4194304 pixels (2048x2048) and within an aspect ratio of 6:1
quality	very_low, low, medium (default), high	Higher tiers cost more and take longer
seed	integer	The same images, prompt, settings, and seed give the same result

Up to five images can be connected in total. Inputs above roughly 4 MP are scaled down before they are uploaded, and each image must sit between 1:6 and 6:1.

Edit re-renders the whole image, so it can shift the framing and the untouched areas along with the edit. Use Precise Edit when the rest of the picture has to stay exactly as it is.

Precise Edit

Change only what the prompt asks for. Untouched pixels stay identical. Accepts up to four reference images beyond the image being edited.

Run on Comfy Cloud

Open in Comfy Cloud

Download Workflow

Download JSON or search “Ideogram 4.5: Precise Edit” in Template Library

Input material

redhair_girl_portrait.png

Load into the image_1 slot on the Ideogram 4.5 Precise Edit node

Precise Edit shares the reference slots, @Image1 addressing, quality, and seed controls with Edit. It has no size control. With a region, the output keeps the size of image_1. Without a region (context_window at the default none), sources above about 4 MP are scaled down first, so the output may be smaller. The template draws its instruction with the Build JSON Prompt node as well.

Edit a single region

Without a region the node re-renders the whole of image_1, and anything above about 4 MP (2048x2048) is scaled down first. Set context_window to edit one region instead: only that region is regenerated, and the output keeps the original size of image_1, from a source of up to 8192x8192 pixels. Outside the region, pixels are copied through unchanged while image_1 stays a PNG. If image_1 is converted to JPEG because the PNG exceeds the upload budget, pixels outside the region can change slightly.

context_window	Extra input	Region that is edited
none (default)	none	The whole image, scaled down above about 4 MP
auto	context_window_mask	Placed around the white area of the mask
custom	context_window_bboxes	One enclosing rectangle around all connected boxes (space between separated boxes is included and may be edited), for example from the Create Bounding Boxes node

The region can be 256 to 4608 pixels per side, up to about 4 MP, and at most 6:1. Regions or boxes under 256 pixels per side are padded to 256. For auto, keep the context_window_mask region within those same limits. A larger white area is not rejected: Ideogram may fall back to a whole-image edit, and a large image may be scaled down. Only the region has to be valid, so image_1 itself can have any aspect ratio, and with context_window_mask you can connect three reference images instead of four.

Use it for product recolors, material swaps, and touch-ups that must leave the lighting, background, and framing of the rest of the photo alone.