WorkflowPrésentéExterne

[Image] WAN Character (face/body) swap — v1.0

Long time since I posted something in the model section. And this time it is a workflow :)

Sourcecivitai.com/models/2073359

TâcheMultipleCapacitésImage en image · Mise à l'échelleFamilleWanVersionRelease 2346059 — malcolmrey-wan-vace-character-swap-workflow.jsonEstimé VRAMEstimation indisponibleUnratedClasse

Profil d'exécution

Accès: Flux de travail externe
Statut: Présenté
Version: malcolmrey-wan-vace-character-swap-workflow.json
Source: Civitai
Mis à jour: 3 nov. 2026

Description de la source

No release description captured.

Exigences

Modèles

4x_NMKD-Siax_200k.pth
Non résolu
Upscaler · Unknown
RealESRGAN_x2.pth
TrouvéOptionnel
Upscaler · 64.0 MB · PTH · Unknown
umt5_xxl_fp8_e4m3fn_scaled.safetensors
Dépôt identifié
Text encoder · Unknown
Unresolved model selection: Florence2ModelLoader
Non résolu
Unknown
Unresolved model selection: LoadDwposeTensorrtModels
Non résolu
Unknown
wan_2.1_vae.safetensors
Dépôt identifié
VAE · Unknown
wan_motokokusanagi_v1.safetensors
Non résolu
LoRA · Unknown
wan2.1_vace_14B_fp8_e4m3fn.safetensors
Trouvé
Diffusion model · 16.2 GB · SAFETENSORS · Unknown
wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors
Trouvé
Diffusion model · 13.3 GB · SAFETENSORS · FP16 + FP32 + FP8 (tensor header) · Unknown
Wan21_T2V_14B_lightx2v_cfg_step_distill_lora_rank32.safetensors
Fichier indiqué
LoRA · 302 MB · Civitai · 1585622

Packs de nœuds

comfyui_essentials
Trouvé
Pack de nœud · Registry
comfyui_ultimatesdupscale
TrouvéOptionnel
Pack de nœud · Registry
comfyui-custom-scripts
Trouvé
Pack de nœud · Registry
ComfyUI-Dwpose-Tensorrt
Trouvé
Pack de nœud · Registry
comfyui-easy-use
Trouvé
Pack de nœud · Registry
comfyui-florence2
Possible
Pack de nœud · Registry
comfyui-kjnodes
Trouvé
Pack de nœud · Registry
comfyui-rmbg
Trouvé
Pack de nœud · Registry
comfyui-various
Trouvé
Pack de nœud · Registry
Fast Groups Muter (rgthree)
Non résolu
Pack de nœud · Unknown
GetNode
Non résolu
Pack de nœud · Unknown
RES4LYF
Trouvé
Pack de nœud · Registry
rgthree-comfy
Trouvé
Pack de nœud · Registry
SetNode
Non résolu
Pack de nœud · Unknown
was-ns
Non résolu
Pack de nœud · Registry

Commentaire généré par l’IA

Explication générée par l’IA à partir des détails de la source et de la configuration. Les suggestions sont clairement signalées.

Objectif

This workflow is for replacing a character’s face or whole body in a source image. It takes a source image and character or person LoRAs, and can produce a face swap or whole-body swap with optional upscaling.

Flux

To use it, add a source image, choose your character or person LoRAs, and run the workflow.
Suggestion · non vérifié

Entrées et sorties

Input: a source image and character or person LoRAs.
Output: a face swap or whole-body swap, with optional upscaling.

Installation

The listing identifies Wan Video 14B t2v as the base model and Malcolmrey as the publisher.

Dépendances

The listing names these node packs: comfyui_essentials; comfyui_ultimatesdupscale; comfyui-custom-scripts; ComfyUI-Dwpose-Tensorrt; comfyui-easy-use; comfyui-kjnodes.
It also names these node packs: comfyui-florence2; comfyui-rmbg; comfyui-various; RES4LYF; rgthree-comfy; was-ns.
Model files listed include: 4x_NMKD-Siax_200k.pth; RealESRGAN_x2.pth; umt5_xxl_fp8_e4m3fn_scaled.safetensors; wan_2.1_vae.safetensors; wan2.1_vace_14B_fp8_e4m3fn.safetensors; wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors; Wan21_T2V_14B_lightx2v_cfg_step_distill_lora_rank32.sაფե

Configuration

The upscaler can be disabled. It supports LoRAs and can work as a “high-res-fix” kind of tool.

Précautions

Before using it, verify that RealESRGAN_x2.pth, wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors, comfyui-florence2, and was-ns are available in your setup.
Suggestion · non vérifié
No license is specified for this workflow.

Faut-il modifier cela ?

Connectez-vous pour envoyer une demande de modification.

Sources

1 source

CivitaiPrincipalVersion 2346059

civitai.com/models/2073359

Vérifié 3 nov.

Workflow instructions

Author notes from malcolmrey-wan-vace-character-swap-workflow.json · selected release

Drop here the image you want to use as a base (this is not the reference image, we do not use one)

Author notes from malcolmrey-wan-vace-character-swap-workflow.json · selected release

Put in 'b' the higher value of your max resolution that you can generate For example, if you can do 1280x720 / 720x1280 then you should put '1280'

Author notes from malcolmrey-wan-vace-character-swap-workflow.json · selected release

You can try masking prompt for many variants, I've checked those below but others should work too :) head head and body and hands head and hair

Author notes from malcolmrey-wan-vace-character-swap-workflow.json · selected release

You can hook up the Ultimate SD Upscaler with either VACE or regular WAN but the regular upscaler works better for me :)

Author notes from malcolmrey-wan-vace-character-swap-workflow.json · selected release

Add your lora(s) here.

Author notes from malcolmrey-wan-vace-character-swap-workflow.json · selected release

Hello! This is malcolmrey and this is WAN Vace (+WAN 2.2) workflow for replacing a full character or just some part of it (face, etc) This is v1.0.0 of this workflow, the date is 25.10.2025 -- You can find me on: Civitai: https://civitai.com/user/malcolmrey Some of my older models are still there (like perfect eyes and styles) but nowadays it is mainly for the articles - how to train and use models. -- Reddit: https://reddit.com/r/malcolmrey Here I share some various news concerning my models, etc. -- HuggingFace https://huggingface.co/malcolmrey Here I share most of my Loras nowadays. -- BuyMeACoffee https://buymeacoffee.com/malcolmrey Here is where you can support me, make some requests and also get access to the MEGA folder with all my loras. (I keep all my loras public, but it is better to be safe than sorry, so having a mirror is always nice) -- This is my first original workflow, made from scratch :) -- How To Use In Settings you: * hook your models, vae, clip * set if you want to use sage attention * maximum resolution your GPU can handle for a single image generation (this is important as you can use base source images that are bigger and it will make sure to automatically resize it so you do not get OOM) In Playground you: * load source image that will be manipulated to insert your character * load loras for character, styles, etc * enable/disable upscaler which also works as "high res fix" that adds some details (go there to set up character loras too) * set up positive prompt (and tweak negative if needed) * set up masking prompt (you want to change head or also full body, etc) In Upscaling you: * can load additional loras (character, styles, etc) * can change the upscale factor or pick your favorite upscaling model In Generation you: * can add the light/fast lora * can change steps and your favorite sampler In Masking you: * can enable the Invert node in case you want to change everything EXCEPT your character In Resizing you: * should not touch anything :-)

Author notes from malcolmrey-wan-vace-character-swap-workflow.json · selected release

<--- You can add character/style/etc loras here