1024 to 4096px in Seconds — Z-Image Turbo + NVIDIA PiD in ComfyUI

In this video, I'll show you how to generate ultra-high-resolution 4K images using Z-Image Turbo combined with NVIDIA's PiD (Pixel Diffusion Decoder) inside ComfyUI — a two-stage pipeline that first generates a 1024px image at lightning speed, then decodes it to stunning 4096px quality in just 4 steps. This is one of the most powerful open-source image upscaling workflows available right now.

Featuring NVIDIA PiD (Pixel Diffusion Decoder)huggingface.co

What you'll learn

• - NVIDIA P-Model (Pixel Diffusion Decoder) — How it differs from standard image generation by working directly in pixel space.
• - One-Step High-Resolution Generation: The process of creating 2K and 4K images in a single step to save time.
• - Workflow Setup in ComfyUI: Steps to integrate P-model nodes and update your configuration.
• - VRAM Optimization: Differences between FP8 and BF16 model versions for different hardware.
• - Generative vs. Latent Upscaling — A comparison of how P-model enhances fine details like hair and skin texture compared to standard methods.
• - Z-Image Turbo Integration: How to use base models to generate latent images for upscaling.

Workflowdrive.google.com

Model links

Z-Image Turbo (Stage 1)

models/diffusion_models

BF16huggingface.co NVFP4huggingface.co

models/text_encoders

qwen_3_4b.safetensorshuggingface.co

models/vae

ae.safetensorshuggingface.co

PiD Decoder (Stage 2)

models/diffusion_models

pid_flux1_1024_to_4096_4step_bf16.safetensorshuggingface.co

models/text_encoders

gemma_2_2b_it_elm_bf16.safetensorshuggingface.co

• 👍 Support the Channel If you found this helpful:

• 👍 Like the video
• 🔔 Subscribe for more AI workflows
• 💬 Comment what you want to see next

#ZImageTurbo#NvidiaPiD#ComfyUI#AIImage#GenerativeAI#4KAI#PixelDiffusion#AIUpscale