NVIDIA's PiD Pixel Diffusion is the best AI image model for high resolution images. Free and open source!
Requirements:
- The "Minimum" Threshold (12GB VRAM): A card with 12GB VRAM (e.g., RTX 3060 12GB, RTX 4070/5070 Ti) sits right on the edge. Since the official test used 13GB, you would likely need to use lower resolution outputs or optimization tricks to fit within the 12GB limit.
· Architecture: An NVIDIA GPU with CUDA support is required.
· Software: PyTorch (with CUDA), diffusers>=0.37, and transformers>=4.57.
https://github.com/Comfy-Org/ComfyUI/pull/14103
https://huggingface.co/Comfy-Org/PixelDiT/tree/main/text_encoders
https://huggingface.co/Comfy-Org/PixelDiT/tree/main/text_encoders
00:00 - PiD intro
00:19 - Personal demos
02:00 - More demos and how it works
02:52 - PiD vs SeedVR2
04:14 - How to install
05:49 - Workflows
06:24 - PiD upscaler workflow
11:46 - Higgsfield
13:16 - Before after slider
14:06 - Zimage to PiD
19:55 - PiD text to image