Single sampling node (UIT Sampler) for HiDream-O1-Image .
- Single node covering the full generation pipeline for UiT model.
- img2img support — input images are automatically rescaled to 4 MP via Lanczos
- Up to 2 reference images attachable to conditioning
step_imagesoutput — all intermediate denoising steps in one batch tensor
- Comfy-Org (Recommended)
- HiDream-O1-Image_clear (Finetuned model for clear outputs)
- HiDream-O1-Image (Official)
- HiDream-O1-Image-Dev (Official)
- HiDream-O1-Image-Dev-2604 (Official)
These images contain ComfyUI workflow.
Category: sampling/uit
| Name | Type | Required | Description |
|---|---|---|---|
model |
MODEL | ✅ | HiDream-O1 model loaded via Load Checkpoint |
clip |
CLIP | ✅ | CLIP / text encoder (dummy connection) |
vae |
VAE | ✅ | VAE (dummy connection) |
input_image |
IMAGE | optional | Source image for img2img; auto-rescaled to 4 MP |
reference_image1 |
IMAGE | optional | Reference image 1 |
reference_image2 |
IMAGE | optional | Reference image 2 |
positive_prompt |
STRING | optional | Positive text prompt |
negative_prompt |
STRING | optional | Negative text prompt |
HiDream-O1 does not use an external VAE or CLIP, but due to the specifications of ComfyUI, a dummy connection is required.
| Name | Type | Description |
|---|---|---|
width |
INT | Output width in pixels (ignored when input_image is connected) |
height |
INT | Output height in pixels (ignored when input_image is connected) |
seed |
INT | Random seed |
cfg |
FLOAT | Classifier-free guidance scale |
sampler |
ENUM | Sampler name (e.g. euler) |
scheduler |
ENUM | Scheduler name (e.g. normal) |
steps |
INT | Number of sampling steps |
denoise |
FLOAT | Denoising strength (1.0 = full generation) |
noise_scale |
FLOAT | Noise scale — base: 8.0, dev: 7.5 |
Default resolution is 2048×2048 (4 MP), matching the HiDream-O1 training resolution. When input_image is connected, the resolution is derived from the rescaled image.
| Name | Type | Description |
|---|---|---|
image |
IMAGE | Final generated image |
step_images |
IMAGE | All intermediate step images stacked into one batch |
- Clone this repository into your ComfyUI
custom_nodesfolder:
cd ComfyUI/custom_nodes
git clone https://github.com/easygoing0114/ComfyUI-uit-hidream-o1.git- Restart ComfyUI. The UIT Sampler node should now appear in the node search under
sampling/uit.
- HiDream-O1-Image and UiT architecture: English | Japanese
- Image Generative AI model License: English | Japanese
This project is licensed under the MIT License.
- Migrated to the comfyui v3 schema.
- Fixed minor issues.
- Fixed an issue where
reference_imagewas also applied tonegative_cond.
- Initial release: UIT Sampler for HiDream-O1-Image
