Skip to content
WorkflowListedExternal
Open source

Z Upscaler — v1.0

It needs less vram than flux. Higher resolutions may require more vram. The higher denoise strength - the better the image quality and the less it resembles the original.

TaskMultiCapabilitiesImage to image · UpscalingFamilyZ-ImageEstimated VRAMNot yet characterizedBgrade

Runtime profile

Access
External workflow
Status
Listed
Source
Civitai
Updated
Aug 18, 2026

Source description

It needs less vram than flux. Higher resolutions may require more vram. The higher denoise strength - the better the image quality and the less it resembles the original.

Recommendations:

denoise level 0.2-0.3

Real-ESRGANx2 and 4xNMKD-Siax

All required models download links:

Z-image-turbo-quantized

Qwen3-4B-quantinized-text-encoder

vae

download any upscale model you like from here upscalers

Required custom nodes:

ComfyUI-Essentials

ComfyUI-KJnodes

What you need

Hardware and download-size estimates are not available yet.

Requirements

Models

6
  • 4x_NMKD-Siax_200k.pth

    Not resolved

    Upscaler · Unknown

  • ae.safetensors

    File unverified

    VAE · Hugging Face · Comfy-Org/z_image_turbo

  • qwen_3_4b.safetensors

    File unverified

    Text encoder · Hugging Face · Comfy-Org/z_image_turbo

  • Checkpoint · Hugging Face · Comfy-Org/z_image_turbo

  • zImageTurboQuantized_fp8ScaledE4m3fnKJ.safetensors

    Not resolved

    Checkpoint · Unknown

  • zImageTurboQuantized_textEncoderFp8Scaled.safetensors

    Not resolved

    Text encoder · Unknown

Node packs

2

AI-generated commentary

AI-generated explanation based on source and configuration details. Suggestions are clearly labeled.

Purpose

  • Interpret Z Upscaler as an image-upscaling workflow: it appears intended to take an image and produce an upscaled image, and its listing gives denoise and upscaler guidance.

    Suggestion · not verified

Inputs and outputs

  • The description does not specify accepted image file types or the output format.

Setup

  • The listed base model is ZImageTurbo.

Dependencies

  • The required custom nodes are ComfyUI-Essentials and ComfyUI-KJnodes.

  • Listed model files include ae.safetensors, qwen_3_4b.safetensors, and z_image_turbo_bf16.safetensors.

  • Other listed files are 4x_NMKD-Siax_200k.pth, zImageTurboQuantized_fp8ScaledE4m3fnKJ.safetensors, and zImageTurboQuantized_textEncoderFp8Scaled.safetensors.

Configuration

  • Try a denoise level of 0.2–0.3, as the listing recommends.

    Suggestion · not verified

  • The listing suggests Real-ESRGANx2 or 4xNMKD-Siax as upscaler choices.

    Suggestion · not verified

  • The description associates higher denoise strength with better image quality and less resemblance to the original.

Cautions

  • The listing reports lower VRAM use than Flux and says higher resolutions may require more VRAM.

  • The guide names Z-image-turbo-quantized and Qwen3-4B-quantinized-text-encoder, while the listed files include zImageTurboQuantized_fp8ScaledE4m3fnKJ.safetensors and zImageTurboQuantized_textEncoderFp8Scaled.safetensors; check the names before downloading.

    Suggestion

Does this need to be edited?

Sign in to send an edit request.

Sources

1 source
CivitaiPrimary
civitai.com/models/2241420

Verified Jul 19

Source excerpts

1 excerpt

Source context: 660 downloads · Type Workflows · Base model ZImageTurbo