Smittie's PiD Upscaler Workflow
Description
With this workflow you can upscale low-resolution images (512, 1K) to much higher resolutions (2K, 4K) using Nvidia's PiD (Pixel Diffusion). It is a simple image to image (I2I) workflow.
Install Instructions
1. Update ComfyUI to latest build with PiD support
2. Install all missing custom nodes via the ComfyUI Manager
3. Download all needed models
Diffusion model
For Z-Image or Flux.1 like images in 1K resolution use pid_flux1_1024_to_4096_4step_bf16.safetensors
Text encoder
VAE
Video Instructions
If you don't know what to do, you can watch the Nvidia PiD Tutorial by AI Search. He works with a different workflow, which I have built upon.
Acknowledgement
Most credits goes to the Nvidia, ComfyUI and CivitAI team, as well as kijai and AI Search. Thank you guys!
Description
initial release
FAQ
Comments (8)
Бесполезный воркфлоу, нерабочий. Не работает ни в одной версии комфи.
It works. Just update the portable ComfyUI version with the "update_comfyui.bat", then install the missing nodes.
Perhaps you can share an error message you have in the console?
Аналогично, зависает на сабграфе PiD Sampler.
Там столько всего внутри, он что, майнингом занимается?)
What do you mean with "зависает", what freezes there? It should run, if you have the hardware to run it.
No, there is no mining. You can inspect the graphs and the code from them, if you are uncertain.
@Smittie It reaches the SamplerCustom node and freezes. The GPU and VRAM are fully occupied, but the process does not continue. You can't undo it either. Only restarting Comfy works.
@serch2k Ok, the only thing I can do from distance is basically to ask you these questions:
1. Did you update ComfyUI to the latest version?
2. Is not only your VRAM, but also you RAM exceeded?
3. Are all downloaded models error free (no CRC-Error caused by download)?
@Smittie 1. Yes, always
2. no, 16vram+64ram
3. everything is fine. everything works with other workflow
@serch2k mmh, then I really don't know why it got stuck for you during the SamplerCustom node. You could try to swap the SamplerCustom with a SamplerCustomAdvanced node (or the easier KSampler) and hook it up correctly, perhaps this one will work for you?