WorkflowListedExternal

FLUX.1 SmartUpscaler-[2Pass Upscaler]. ControlNet + SD UPscale

This workflow designed to restore, upscale and color fix low resolution realistic images and portraits (blury and non-blurry aswell) using FLUX.1 models. Allows you to output 4Megapixel+ images.

Sourcecivitai.com/models/888017

TaskMultiCapabilitiesImage to image · UpscalingFamilyFLUXEstimated VRAMEstimate unavailableBgrade

Runtime profile

Access: External workflow
Status: Listed
Source: Civitai
Updated: Jul 20, 2026

Source description

This workflow designed to restore, upscale and color fix low resolution realistic images and portraits (blury and non-blurry aswell) using FLUX.1 models. Allows you to output 4Megapixel+ images.

Important Usage notes:

jasperai/Flux.1-dev-Controlnet-Upscaler · Hugging Face By default any image you load (even 256x256 image) will be resized to match 1Megapixel, which is base resolution for FLUX.1 model. This setting can be altered in - "Image Input Resizer" nodes. After that, this resized image will be upscaled by whichever value you have in Ultimate SD Upscale node. Prompt is not mandatory, but can help if you get unstable results, as well as denoise values. You can still use value "Image Upscale_by = 1", in that case upscaler will act as an image refiner.

For celebrity images or characters it's better to use celebrity/character designated Lora to preserve accurate facial features.

AI-generated commentary

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

Purpose

FLUX.1 SmartUpscaler-[2Pass Upscaler]. ControlNet + SD UPscale is for restoring, upscaling, and color-fixing low-resolution realistic images and portraits. It takes a loaded image, can optionally use a prompt, and can produce images at 4 megapixels or more.

Flow

By default, a loaded image—including one that is 256x256—is resized to 1 megapixel, then upscaled by the value set in Ultimate SD Upscale.

Inputs and outputs

You can leave the prompt empty; the description says a prompt may help with unstable results, as can denoise values.
Suggestion · not verified
Set Image Upscale_by = 1 if you want the upscaler to act as an image refiner.
Suggestion · not verified

Setup

It is published by Nxssar and uses Flux.1 D as its base model.
An AI-app option is linked at https://www.seaart.ai/workFlowAppDetail/cti9f9le878c73dh27ig.

Dependencies

The workflow names these model files: 4x_NMKD-Siax_200k.pth, clip_l.safetensors, Flux.1-dev-Controlnet-Upscaler.safetensors, and flux1-dev-Q6_K.gguf.
It also names t5xxl_fp8_e4m3fn.safetensors and thwri/CogFlorence-2.2-Large.
Named nodes include Anything Everywhere3; CLIPTextEncodeFlux; ColorMatch; ConditioningZeroOut; ControlNetApplyAdvanced; ControlNetLoader; CR Image Input Switch; CR Prompt Text; CR Text Input Switch; DownloadAndLoadFlorence2Model; DualCLIPLoaderGGUF; easy imageSize; Fast Groups Bっ

Configuration

Change the 1-megapixel target in the “Image Input Resizer” nodes when you want a different base resolution.
Suggestion · not verified
For celebrity images or characters, consider a designated Lora to help preserve accurate facial features.
Suggestion · not verified

Cautions

Before use, verify that your setup includes the named model files and nodes.
Suggestion · not verified
No license is specified for this workflow.

Does this need to be edited?

Sources

1 source

CivitaiPrimaryVersion 1188527

civitai.com/models/888017

Verified Jul 20

Source excerpts

1 excerpt

NoteJul 20civitai.com/models/888017

Source context: 2678 downloads · Type Workflows · Base model Flux.1 D