WorkflowListedExternal

FLUX-Advanced-Img2Img workflow. Rectified flow Inversion [RF inversion] [Ref2Img] - V2

This is an Advanced Img2Img workflow for ComfyUI based on google's rectified flow inversion. It allows you to change your image , transfer style, inpaint, face swap etc... only by using prompts . No manual inpainting...

Sourcecivitai.com/models/886496

TaskMultiCapabilitiesImage to image · UpscalingFamilyFLUXEstimated VRAMEstimate unavailableBgrade

Runtime profile

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

Source description

To simplify: imagine basic img2img workflow, where you send input image -> make random noise from it -> send it to sampler and get output image, but in that case you can control what kind of noise you are getting from input image, depending on your settings. For instance, you can change either, part of image(inpainting) or transfer image style(ip-adapter).

So it looks like this: input image -> unsample image to noise (with additional "textual attention mask" a.k.a prompt input for img2noise unsampler. It should work good enough even without prompt, but you can specify what part of image you want to get "masked" for denoising) -> noise from unsampler goes to the sampler and you get output image edited with unsampled noise.

AI-generated commentary

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

Purpose

This ComfyUI image-to-image workflow uses Google's rectified-flow inversion to edit an input image with prompts, including style transfer, inpainting, and face swaps; it takes an image, a standard prompt, and an optional unsampling prompt, and produces an edited output image.

Flow

The workflow first unsamples the input image into noise, optionally using a textual attention prompt, then sends that noise to a sampler with a standard prompt.
The added unsampling step lets you control the noise derived from the input image before sampling.

Inputs and outputs

The optional unsampling prompt can indicate which part of the image to mask for denoising, and the result is an edited image.

Setup

Use it in ComfyUI; additional instructions are provided within the workflow.
Its listed base model is Flux.1 D.

Dependencies

The workflow names these model files and models: clip_l.safetensors, Flux.1-dev-Controlnet-Depth.safetensors, Flux.1-dev-Controlnet-Upscaler.safetensors, flux1-dev-turbo8-fp8.safetensors, microsoft/Florence-2-base, and t5xxl_fp8_e4m3fn.safetensors.
It names these ComfyUI nodes: PromptBuilder //Inspire, RepeatLatentBatch, OverrideCLIPDevice, ImageResizeKJ, ControlNetApplyAdvanced, ConditioningZeroOut, VAELoader, Seed Everywhere, Power Lora Loader (rgthree), Image Comparer (rgthree), FluxForwardODESampler, DisableNoise,
It also names these ComfyUI nodes: ShowText|pysssss, iToolsPromptStylerExtra, Fast Groups Bypasser (rgthree), CR Simple Image Compare, PreviewImage, Florence2Run, OutFluxModelSamplingPred, InFluxModelSamplingPred, Fast Groups Muter (rgthree), DownloadAndLoadFlorence2Model, Image

Configuration

Use the optional unsampling prompt when you want to identify which part of the image should be masked for denoising.
Suggestion · not verified
Consider Hyper and Turbo merged models or LoRAs of Flux, which the description recommends.
Suggestion · not verified

Cautions

Before use, verify that PromptBuilder //Inspire, RepeatLatentBatch, OverrideCLIPDevice, and ImageResizeKJ are available in ComfyUI.
Suggestion · not verified
Before use, verify that the named model files and models are available: clip_l.safetensors, Flux.1-dev-Controlnet-Depth.safetensors, Flux.1-dev-Controlnet-Upscaler.safetensors, flux1-dev-turbo8-fp8.safetensors, microsoft/Florence-2-base, and t5xxl_fp8_e4m3fn.safetensors.
Suggestion · not verified

Does this need to be edited?

Sources

1 source

CivitaiPrimaryVersion 997919

civitai.com/models/886496

Verified Jul 20

Source excerpts

1 excerpt

NoteJul 20civitai.com/models/886496

Source context: 665 downloads · Type Workflows · Base model Flux.1 D