Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.comfy.org/llms.txt

Use this file to discover all available pages before exploring further.

Make sure your ComfyUI is updated.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 If you can’t find them in the template, your ComfyUI may be outdated. (Desktop version’s update will delay sometime)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
HiDream-O1-Image Demo HiDream-O1-Image is a unified image generative foundation model open-sourced by HiDream-ai on May 8, 2026. Built on a Pixel-level Unified Transformer (UiT), it operates without external VAEs or disjoint text encoders, supporting text-to-image, instruction-based image editing, subject-driven personalization, and storyboard generation at up to 2,048 × 2,048 resolution. HiDream-O1-Image is released under the MIT License and natively supported in ComfyUI (PR #13817). HiDream-O1-Image - GitHub

​
Model Versions

ModelInference StepsRepository
HiDream-O1-Image (Full)50🤗 HiDream-O1-Image
HiDream-O1-Image-Dev28🤗 HiDream-O1-Image-Dev

​
HiDream-O1-Image Full Workflow

​
1. Download Workflow

Update your ComfyUI to the latest version, then go to Workflow -> Browse Templates -> Image and find “HiDream O1 Full: Image generation”. HiDream-O1-Image Full Workflow

Download JSON Workflow File

Download workflow

Run on Comfy Cloud

Open in cloud

​
2. Download Models

Checkpoint — re-packaged and quantized. All use bf16 for worst outliers, unused deepstack layers removed: Text Encoder (prompt enhancement) — shared across all versions: LoRA (optional) — the Dev distillation can also be applied to the Full model as a LoRA, allowing you to adjust the distillation strength (by Kijai): 
📂 ComfyUI/
├── 📂 models/
│   ├── 📂 checkpoints/
│   │   ├── hidream_o1_image_fp8_scaled.safetensors
│   │   ├── hidream_o1_image_mxfp8.safetensors
│   │   └── hidream_o1_image_bf16.safetensors
│   ├── 📂 loras/
│   │   └── hidream_o1_dev_lora_rank_64_bf16.safetensors
│   └── 📂 text_encoders/
│       └── gemma4_e4b_it_fp8_scaled.safetensors

​
3. Using the Workflow

  • Make sure CheckpointLoaderSimple loads your chosen checkpoint
  • Enter your prompt in the CLIPTextEncode node
  • Text-to-Image mode: set the “Switch to Image Edit” toggle to off (default). The sampler will use your text prompt directly.
  • Image Edit mode: set “Switch to Image Edit” to on, then upload a reference image in the Load Image node and connect it to HiDreamO1ReferenceImages.
Note: The O1 sampler outputs latent samples that need to be decoded through a VAEDecode node using the VAE loaded by CheckpointLoaderSimple.

​
HiDream-O1-Image Dev Workflow

​
1. Download Workflow

Go to Workflow -> Browse Templates -> Image and find “HiDream O1 Dev”. HiDream-O1-Image Dev Workflow

Download JSON Workflow File

Download workflow

Run on Comfy Cloud

Open in cloud

​
2. Download Models

Checkpoint (Dev) — re-packaged and quantized. All use bf16 for worst outliers, unused deepstack layers removed: Text Encoder (prompt enhancement) — shared across all versions: 
📂 ComfyUI/
├── 📂 models/
│   ├── 📂 checkpoints/
│   │   ├── hidream_o1_image_dev_fp8_scaled.safetensors
│   │   ├── hidream_o1_image_dev_mxfp8.safetensors
│   │   └── hidream_o1_image_dev_bf16.safetensors
│   └── 📂 text_encoders/
│       └── gemma4_e4b_it_fp8_scaled.safetensors

​
3. Using the Workflow

  • Make sure CheckpointLoaderSimple loads hidream_o1_image_dev_fp8_scaled.safetensors
  • The Dev version uses 28 steps with CFG=1.0 — no negative prompt needed
  • Text-to-Image mode: set the “Switch to Image Edit” toggle to off (default)
  • Image Edit mode: set “Switch to Image Edit” to on, upload a reference image in Load Image, and connect it to HiDreamO1ReferenceImages

​
Additional Notes

  • Long Text Rendering: HiDream-O1-Image excels at rendering text within images. Specify text content, font style, color, and positioning in your prompt for best results. Long-Text Rendering & Layout Control
  • Image Scaling: If your input reference image is too large, enable the ImageScaleToTotalPixels node (Ctrl+B) to scale it to 4MP.
  • Patch Seam Smoothing (Experimental): The HiDreamO1PatchSeamSmoothing node reduces visible seam artifacts during sampling. Enabled by default in the Full workflow.