This repository provides a custom ComfyUI node for generating images using the HiDream AI model. It targets users of ComfyUI looking to leverage HiDream's capabilities, offering configurable generation parameters and support for various model types and quantization levels to manage VRAM usage.

How It Works

The node integrates the HiDream AI model into the ComfyUI workflow, allowing users to generate images via text prompts. It supports "full," "dev," and "fast" model types and utilizes 4-bit quantization (NF4) to reduce VRAM requirements, making it accessible on hardware with approximately 15GB VRAM, though full models may require up to 27GB.

Quick Start & Requirements

• Install: Clone the repository into ComfyUI/custom_nodes/.
• Requirements: Install dependencies via pip install -r requirements.txt. Requires PyTorch 2+ for SDPA. Triton installation is also necessary for SageAttention.
• Models: Download HiDream models and four text encoders from Hugging Face (e.g., azaneko/HiDream-I1-Full-nf4, hugging-quants/Meta-Llama-3.1-8B-Instruct-GPTQ-INT4).
• GPU: A GPU is required for inference; 16GB+ VRAM recommended.
• Docs: Workflow folder

Highlighted Details

• Supports 4-bit quantization (NF4) for reduced VRAM usage.
• Offers configurable resolution and inference steps.
• Includes support for multiple image generation (Img2img not yet supported).
• Merged with SanDiegoDude/ComfyUI-HiDream-Sampler for a cleaner implementation.

Maintenance & Community

The project is maintained by lum3on, with contributions from SanDiegoDude. The author notes they are an illustrator, not a developer, and welcomes contributions and PRs. A list of planned features and improvements is available in the README.

Licensing & Compatibility

The repository does not explicitly state a license in the provided README. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The project is in active development, with the author indicating they will update it at their "pace and capabilities." Some features are explicitly marked as "not yet supported" (e.g., multi-image for Img2img). Installation can be challenging for some users.