Universal ComfyUI workflow for HunyuanImage-3.0-Instruct using the HunyuanImage3 Low-Memory BF16 custom nodes.
Designed for low-VRAM systems with BF16 CUDA / CPU / SSD offload support.
Features:
• Text-to-Image
• Image-to-Image
• Diffusion preview
• Preserve mask
• Final RGB pixel preservation
• Tiled refinement
• CUDA / CPU / SSD offload
• Low-memory BF16 inference
Required custom node:
HunyuanImage3 Low-Memory BF16
Node ID:
hunyuanimage3-low-memory
Recommended version:
0.4.0
GitHub:
https://github.com/foxidermist/hunyuanimage3-low-memory
Workflow:
HunyuanImage3_Instruct_Universal_0.4.0.json
The workflow is intended for HunyuanImage-3.0-Instruct and provides a single universal graph for both T2I and I2I generation.
Description
v0.4.0 is the first public release of the universal workflow.
Includes:
- T2I and I2I in one workflow
- Low-VRAM BF16 operation
- CUDA / CPU / SSD offload
- Diffusion preview
- Preserve mask
- Final pixel preservation
- Tiled refinement
Requires:
HunyuanImage3 Low-Memory BF16 custom nodes v0.4.0
