hofixD/comfyui-hidream-l1-full-img2img

🌟 HiDream Img2Img ComfyUI Workflow

Advanced image-to-image generation with HiDream model suite and Florence-2 prompt generator

📋 Overview

This workflow combines the power of HiDream diffusion models with Florence-2 captioning for enhanced image-to-image generation in ComfyUI:

• ✨ Image-to-image generation with the state-of-the-art HiDream diffusion model
• 🔮 Optional Florence-2 intelligent prompt generation and image captioning
• 🖼️ VAE encoding/decoding and advanced CLIP-based text encoding
• 🚫 Customizable negative prompts for artifact reduction
• 💻 Low VRAM mode available for systems with limited resources

🚀 Try It Now!

You can test this workflow directly on Replicate:
▶️ Run on Replicate

📥 Required Models & Setup

🎨 Diffusion Model

The workflow supports two HiDream model variants:

Full Model (Default)

• hidream_i1_full_fp16.safetensors
  📁 Place in: ComfyUI/models/diffusion_models
  📦 Download

Dev Model (Alternative)

• hidream_i1_dev_bf16.safetensors
  📁 Place in: ComfyUI/models/diffusion_models
  📦 Download

Credit: HiDream.ai

📝 Text Encoders

📁 Place all in: ComfyUI/models/text_encoders

• clip_g_hidream.safetensors
  📦 Download

• clip_l_hidream.safetensors
  📦 Download

• llama_3.1_8b_instruct_fp8_scaled.safetensors
  📦 Download

• t5xxl_fp8_e4m3fn_scaled.safetensors
  📦 Download

🖼️ VAE

• ae.safetensors
  📁 Place in: ComfyUI/models/vae
  📦 Download

🔍 Florence-2 Prompt Generator

• Florence-2-large
  ⚡ Automatic download at runtime
  📦 Microsoft Florence-2

  Credit: MiaoshouAI for the optimized implementation

💡 Usage Guide

1. Download all required models and place them in the correct directories as listed above
2. Import the workflow into ComfyUI
3. Load your input image, adjust settings as needed
4. Choose whether to use Florence-2 automatic captioning:
   • With Florence-2: Provide a brief prefix that will be combined with the AI-generated caption
   • Without Florence-2: Enter your complete custom prompt directly
5. Customize the negative prompt to avoid unwanted elements
6. Generate new images with enhanced quality

💻 Low VRAM Mode (< 24GB VRAM)

For systems with limited VRAM, use this alternative setup:

1. Install city96/ComfyUI-GGUF custom node
2. Replace the standard Diffusion Model Loader with the Unet LOADER node
3. Download the optimized HiDream-I1 Full or DEV GGUF model:
   • 📦 HiDream-I1-Full-gguf
   • 📦 HiDream-I1-DEV-gguf
   • 📁 Place in: ComfyUI/models/unet

📊 Workflow Diagram

🙏 Acknowledgements

• HiDream.ai for the remarkable diffusion model and encoders
• Microsoft for the Florence-2 vision-language model
• MiaoshouAI for the Florence-2 prompt generator implementation
• ComfyUI team for the intuitive workflow engine
• city96 for the GGUF optimization for low VRAM systems

---

⭐ If you find this workflow useful, please consider starring the repository! ⭐