🎯 Project Overview

HiDream-I1 is a 17B parameter text-to-image foundational model open-sourced by Zhixiang Future in April 2025. It excels in multiple benchmarks, surpassing FLUX and becoming a new benchmark in open-source image generation.

💡 Recommended Solution for 16GB VRAM

1. First Choice: FP8 Version — Best quality, VRAM just enough
2. Alternative: GGUF Q8 — Leaves more VRAM headroom
3. Alternative: GGUF Q6 — Allows running other programs simultaneously

Step 1: Install ComfyUI

# Clone ComfyUI
git clone https://github.com/comfyanonymous/ComfyUI
cd ComfyUI

Create virtual environment
python -m venv venv
source venv/bin/activate  # Linux/Mac
or venv\Scripts\activate  # Windows
Install dependencies
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
pip install -r requirements.txt

Step 2: Download Model Files (FP8 Version)

# Enter model directory
cd ComfyUI/models/diffusion_models/

Download FP8 Dev version (recommended)
wget https://huggingface.co/Comfy-Org/HiDream-I1_ComfyUI_repackaged/resolve/main/hidream_i1_dev_fp8.safetensors
Or GGUF quantized version (requires ComfyUI-GGUF plugin)
wget https://huggingface.co/city96/HiDream-I1-Dev-gguf/resolve/main/hidream-i1-dev-Q8_0.gguf

Step 3: Download Text Encoders

cd ComfyUI/models/clip/

Download 4 text encoders (~10GB)
wget https://huggingface.co/Comfy-Org/HiDream-I1_ComfyUI_repackaged/resolve/main/text_encoders/clip_l.safetensors
wget https://huggingface.co/Comfy-Org/HiDream-I1_ComfyUI_repackaged/resolve/main/text_encoders/t5xxl_fp8.safetensors
wget https://huggingface.co/Comfy-Org/HiDream-I1_ComfyUI_repackaged/resolve/main/text_encoders/llama_3.1_encoder_fp8.safetensors

Step 4: Launch and Use

# Launch ComfyUI
python main.py --listen 0.0.0.0 --port 8188

Access via browser at http://localhost:8188
Drag in workflow JSON or image to start generating

# Install dependencies
pip install torch diffusers transformers accelerate sentencepiece
pip install git+https://github.com/huggingface/diffusers

Python inference code
import torch
from diffusers import HiDreamImagePipeline
pipe = HiDreamImagePipeline.from_pretrained(
    "HiDream-ai/HiDream-I1-Dev",
    torch_dtype=torch.float16,
    variant="fp8"
)
pipe.to("cuda")
prompt = "A majestic lion in a mystical forest, ultra detailed, 8k"
image = pipe(prompt, guidance_scale=5.0, num_inference_steps=28).images[0]
image.save("output.png")

✨ Summary

RTX 4060Ti 16GB can smoothly run HiDream-I1!
Recommended to use FP8 Dev Version + ComfyUI
For the best image quality and generation experience 🎨