ComfyUI - VAE Decode를 대체하는 PiD 리뷰

I tested whether PiD (Pixel Diffusion Decoder) in ComfyUI can replace VAE Decode and produce higher-resolution, sharper results. 00:00 Intro 00:10 PiD project overview 00:46 What PiD does 01:07 ComfyUI-PiD integration and workflow 01:49 PiD Conditioning node 02:12 Downloading the Hugging Face model 02:40 Basic PiD workflow test 03:40 Z image Turbo workflow 04:01 512 x 512 input and 2048 output 04:58 VAE Decode vs PiD structure 06:08 Comparing results with Image Compare node 06:37 Seed change test 06:54 4096 high-resolution output test 07:34 FLUX model selection error 08:22 FLUX2 VAE test 09:13 Color issues and current limits 09:49 Extra test with FLUX1 VAE 10:20 Wide aspect ratio output issue 10:45 Training resolution and input size limits 11:26 Basic Scheduler and sigma adjustment 11:49 Comparison with existing upscaling methods 12:18 Current evaluation of PiD 12:48 Wrap-up 🚀 Deno AI HUB https://deno2026.github.io/Deno-AI-HUB/ 💻 PC Spec 9950X3D 256GB RAM RTX5090