Nvidia PiD 4K Image Upscaler Workflow (Z-Image, Flux.1, Flux.2, SD3 compatible) by Smittie

Use this image-to-image workflow to upscale low-resolution 512 or 1K images into 2K or 4K images with Nvidia's PiD (Pixel Diffusion); its input is a low-resolution image, and its output is a higher-resolution image.

The graph includes image loading and resizing, PiD conditioning, sampling, VAE encoding and decoding, previewing, and saving.

Alongside the input image, the workflow uses a diffusion model, text encoder, and VAE.

PreviewImage and SaveImage nodes support viewing and saving the resulting images.

Setup requires a ComfyUI build with PiD support, missing custom nodes installed through ComfyUI Manager, and the needed models downloaded.

Listed node packs include comfyui-easy-use, comfyui-kjnodes, and rgthree-comfy.

ComfyMath is also named; verify the intended pack before setup.

Suggestion · non vérifié

Listed model files are pid_flux1_1024_to_4096_4step_bf16.safetensors, gemma_2_2b_it_elm_bf16.safetensors, and ae.safetensors.

For Z-Image or Flux.1-like images at 1K, the instructions specify pid_flux1_1024_to_4096_4step_bf16.safetensors as the diffusion model.

The workflow is labeled compatible with Z-Image, Flux.1, Flux.2, and SD3.

Confirm that the three listed model files are available before use.

Suggestion · non vérifié

No model-use license is listed; verify the applicable usage terms before using it.

Suggestion · non vérifié