UNIVERSAL AIO REGIONAL IMAGE DETAIL, EDIT & REPAIR STUDIO

Universal Mainstream Checkpoint Support • ALL IN ONE Finished-Image Repair • Face • Breast • Areola/Nipple • Hands • Feet • Pubic/Vagina • Piercings • Auto & Manual Mask Edit • Local LoRA • Makeup •Tattoos • Tattoo Copy/Paste Anywhere • FLUX Fill & Object Removal • Qwen Head/Face Swap • 2-Stage Upscale

This is a large, modular ComfyUI finished-image repair and editing workspace built around one simple idea:

Use the checkpoints and model ecosystem you already have instead of rebuilding an image from scratch just because one part of it needs fixing.

Load a finished image, select the model family you want to use, turn on only the stations you need, and make controlled local repairs or edits while protecting the rest of the image.

The primary regional-repair system supports the major mainstream checkpoint families through separate Standard and FLUX processing paths. The Standard side is intended for common SD1.5, SDXL, Pony and compatible checkpoints, while the FLUX side provides a dedicated FLUX.1 repair path.

“Universal checkpoint support” here means broad support for the mainstream model families people actually use. Extremely old, unusual, experimental or incompatible architectures are not guaranteed.

You are not required to download every model listed below. The specialized heavyweight models are only needed if you want to use the stations that depend on them.

I intentionally made the workflow as generic as possible and the settings are intentionally conservative to protect your image. Sampler settings may need to be adjusted if you do not get your desired results.

WHAT IT CAN DO...Edit, Repair &/or Detail

Face

Automatically locate and repair facial defects while preserving identity, expression, hair, makeup, jewelry, lighting and the surrounding image.

Breast / Chest

Dedicated regional repair for chest anatomy with automatic targeting, preservation controls and separate Standard/FLUX processing.

Areola / Nipple

Fine-detail segmentation and repair for small anatomical regions that are often difficult for general-purpose detailers.

Hand

Automatic hand detection and localized reconstruction of malformed fingers, anatomy, nails and other hand defects without regenerating the entire image.

Foot

Dedicated foot detection and repair with controls designed to preserve pose, footwear, reflections and occlusions.

Tattoo

Detect tattoos already present in the image and repair clarity, continuity and skin integration while attempting to preserve the original artwork, subject matter, text, colors and placement.

Pubic

Dedicated regional repair with automatic detection and separate Standard/FLUX processing.

Piercing & Jewelry Micro-Detail

Uses open-vocabulary YOLOE detection for visible jewelry and piercings, with a manual-mask option for extremely small, occluded, newly added or repositioned jewelry.

Custom Local Edit

Paint your own mask anywhere on the image and describe exactly what should change. This is the general-purpose station for anything that does not have its own automatic detector.

Local LoRA Editing

Apply compatible LoRAs to a selected local region instead of allowing the LoRA to influence the entire image.

Makeup / Cosmetic Editing

Add, change, enhance or repair makeup locally while protecting the rest of the face and image.

Tattoo Photo / Artwork Transfer

Load a tattoo from another photograph or clean tattoo artwork, extract it when necessary, project it onto a selected body region, integrate it into the skin and optionally apply a preservation-focused finishing pass. You can choose any image really and your target does not have to be on the body. You can use this to transfer an image to objects as well with varying degrees of success.

FLUX.1 Fill / Removal / Replacement

Dedicated masked FLUX Fill station for removing unwanted objects, replacing objects, rebuilding regions or describing what should remain after an object is removed.

Qwen Head / Face Identity Transfer

Dedicated Qwen Image Edit 2511 identity-transfer station. Use whole-head mode to transfer face + hair/head appearance or face-only mode to preserve the target hair.

Two-Stage Upscale & Detail Recovery

Stage 1 performs conventional neural upscaling. Optional Stage 2 performs low-denoise diffusion detail recovery using a family-appropriate processing path.

Final Output / Save

The completed image is routed to the final output and Save Image node.

Before / After Compare

The final station uses ComfyUI's native Image Compare node so the original and finished images can be inspected with a draggable before/after comparison without requiring another custom-node pack.

DESIGNED TO PROTECT THE IMAGE YOU ALREADY LIKE

This workflow is meant for situations where the image is already mostly correct.

Instead of regenerating an entire person or scene because of one malformed hand, damaged tattoo, missing piercing or other localized problem, each enabled station receives the current image, modifies its permitted region and passes the result to the next station.

Unused stations remain OFF.

The master switch also starts OFF, making the default state a safe source-image pass-through.

Every station includes two instructions directly inside the workspace:

WHAT / WHEN / WHY explains what the station does and when it should be used.

STEP-BY-STEP walks a beginner through operating that station.

INSTALLATION

The downloadable package contains the workflow and the companion helper nodes required specifically by this build.

Model weights are not bundled.

There are also no BAT installers. Third-party custom nodes should be installed through ComfyUI Manager whenever possible so users receive the original author's current version.

Update ComfyUI before installing the workflow.

REQUIRED CUSTOM NODES

These packs are used by the workflow itself.

ComfyUI Impact Pack — detailers, SAM support, pipes and lazy conditional routing. Impact Pack currently requires a modern ComfyUI version and recommends Manager installation.
Download / ComfyUI Impact Pack

ComfyUI Impact Subpack — provides UltralyticsDetectorProvider for the automatic BBOX/SEGM detector models. Impact Subpack is now installed separately from Impact Pack.
Download / ComfyUI Impact Subpack

KJNodes — workflow utilities, strings and Set/Get routing.
Download / ComfyUI-KJNodes

Multi LoRA Loader — powers the multi-LoRA controls used by the Local LoRA station.
Download / ComfyUI MultiLoRALoader

QwenEditUtils — provides the Qwen Image Edit custom encoder/configuration nodes used by the Head / Face Identity Transfer station. The repository itself notes that a newer EditUtils project exists, but the current workflow uses the node classes from this package.
Download / Comfyui-QwenEditUtils

Ultimate SD Upscale — used by the Standard-family Stage 2 detail-recovery path.
Download / ComfyUI Ultimate SD Upscale

Tiled Diffusion & VAE — used by the FLUX Stage 2 path and supports FLUX as well as SD-family models.
Download / ComfyUI Tiled Diffusion

YOLOE-26 — open-vocabulary segmentation used by the Piercing/Jewelry station. The pack can automatically download yoloe-26s-seg.pt on first use.
Download / ComfyUI-YOLOE26

Universal Regional Studio helper nodes — the MEGA mask/crop/stitch/geometry/tattoo projection helper nodes are included in this release package and do not need to be downloaded separately.

STANDARD CHECKPOINT SUPPORT

For the Standard repair family, select your own compatible installed checkpoint.

Typical supported mainstream families include:

SD1.5 • SDXL • Pony / Pony-derived checkpoints

There is no artistic checkpoint that everyone is required to use.

The workflow's saved example is currently CyberRealistic Pony V18, but you may replace it with your own compatible checkpoint.

Optional saved example — CyberRealistic Pony

Put Standard checkpoints in:

ComfyUI/models/checkpoints/

FLUX.1 MAIN REPAIR FAMILY

The FLUX repair side uses:

flux1-dev.safetensors
clip_l.safetensors
t5xxl_fp8_e4m3fn.safetensors
ae.safetensors

The official FLUX.1 Dev repository contains flux1-dev.safetensors and ae.safetensors.

FLUX.1 Dev — Black Forest Labs

The required CLIP-L and T5XXL files are available from the ComfyUI FLUX text-encoder repository.

FLUX text encoders — clip_l + T5XXL FP8

Install:

flux1-dev.safetensors → ComfyUI/models/diffusion_models/

clip_l.safetensors and t5xxl_fp8_e4m3fn.safetensors → ComfyUI/models/text_encoders/

ae.safetensors → ComfyUI/models/vae/

FLUX.1 FILL / OBJECT REMOVAL

The dedicated Fill station expects:

flux1-fill-dev-fp8.safetensors

The linked FP8 version is approximately 11.9 GB and is based on Black Forest Labs FLUX.1 Fill Dev.

Download — flux1-fill-dev-fp8.safetensors

Install in:

ComfyUI/models/diffusion_models/

It reuses the FLUX text encoders and VAE listed above.

QWEN IMAGE EDIT 2511 — HEAD / FACE IDENTITY TRANSFER

The Qwen station expects:

qwen_image_edit_2511_fp8mixed.safetensors
qwen_2.5_vl_7b_fp8_scaled.safetensors
qwen_image_vae.safetensors

Comfy-Org's Qwen Image Edit repository includes the exact 2511 FP8Mixed diffusion model used by this workflow.

Download — qwen_image_edit_2511_fp8mixed.safetensors

The Qwen Image repository contains both the FP8-scaled Qwen 2.5 VL encoder and Qwen Image VAE.

Download — qwen_2.5_vl_7b_fp8_scaled.safetensors

Download — qwen_image_vae.safetensors

Install:

Qwen diffusion model → ComfyUI/models/diffusion_models/

Qwen encoder → ComfyUI/models/text_encoders/

Qwen VAE → ComfyUI/models/vae/

TATTOO PHOTO / ARTWORK TRANSFER

IMPORTANT CREDIT — NocturnalWolfStudio

The Tattoo Transfer station exists because of NocturnalWolfStudio's FLUX2 Klein “Steal Tattoo / Body Writing” work.

Their original workflow demonstrated the tattoo-copy approach, and—most importantly—the author created the Place Tattoo and Extract Tattoo LoRAs that perform the specialized tattoo transfer operations used by this station.

The original release explicitly directs users to those two LoRAs.

Full credit for the original tattoo-transfer concept and the specialized tattoo LoRAs goes to NocturnalWolfStudio.

Please visit and support the original author:

Original: FLUX2 Klein — Steal Tattoo / Body Writing by NocturnalWolfStudio

Required LoRAs:

Place Tattoo LoRA — NocturnalWolfStudio

Extract Tattoo LoRA — NocturnalWolfStudio

The workflow expects them under:

ComfyUI/models/loras/flux/tatts/

with the filenames:

flux2klein_extract_tattoo.safetensors
flux2klein_place_tattoo.safetensors

FLUX.2 KLEIN RESOURCES FOR TATTOO TRANSFER

Tattoo Transfer also requires:

flux-2-klein-9b-fp8.safetensors
qwen_3_8b_fp8mixed.safetensors
flux2-vae.safetensors

The official Black Forest Labs FLUX.2 Klein 9B FP8 repository provides the exact flux-2-klein-9b-fp8.safetensors filename used by the release workflow.

Download — FLUX.2 Klein 9B FP8

Comfy-Org provides the Qwen3 8B FP8Mixed encoder and FLUX2 VAE required for the ComfyUI Klein setup.

Download — Qwen3 8B encoder + FLUX2 VAE for Klein 9B

Install:

flux-2-klein-9b-fp8.safetensors → ComfyUI/models/diffusion_models/

qwen_3_8b_fp8mixed.safetensors → ComfyUI/models/text_encoders/

flux2-vae.safetensors → ComfyUI/models/vae/

SAM MODEL

The workflow uses:

sam_vit_b_01ec64.pth

Impact Pack expects SAM models under ComfyUI/models/sams, and its documentation links the official Meta ViT-B checkpoint.

Download — SAM ViT-B / sam_vit_b_01ec64.pth

Install in:

ComfyUI/models/sams/

AUTOMATIC BBOX DETECTOR MODELS

Impact Subpack expects BBOX detector models under ComfyUI/models/ultralytics/bbox/.

Face

face_yolov9c.pt
Download — face_yolov9c.pt

Breast / Chest

female_breast-v4.2.pt
Download — female_breast-v4.2.pt

Hands

hand_yolov8s.pt
Download — hand_yolov8s.pt

Feet

foot-yolov8x-v20.pt
Download — foot-yolov8x-v20.pt

Lower Pelvis

vagina-v3.2.pt
Download — vagina-v3.2.pt

Install all five in:

ComfyUI/models/ultralytics/bbox/

AUTOMATIC SEGMENTATION DETECTOR MODELS

Impact Subpack expects segmentation models under ComfyUI/models/ultralytics/segm/.

Areola / Nipple

Nipple-yoro11x_seg.pt
Download — Nipple-yoro11x_seg.pt

Tattoo Detection

yolov8s_tattoo_manual_v3_best.pt
Download — yolov8s_tattoo_manual_v3_best.pt

Install both in:

ComfyUI/models/ultralytics/segm/

PIERCING / JEWELRY YOLOE MODEL

The saved Piercing/Jewelry station uses:

yoloe-26s-seg.pt

Manual download should normally not be necessary.

The YOLOE-26 custom node is designed to automatically download supported YOLOE weights, including yoloe-26s-seg.pt, on first use. Its first text-prompt inference may also download the MobileCLIP encoder, so internet access is recommended for the first run.

YOLOE-26 setup and model information

UPSCALE MODEL

The saved Stage 1 upscaler is:

4x-UltraSharp.pth

Download — 4x-UltraSharp.pth

The file is approximately 67 MB.

Install in:

ComfyUI/models/upscale_models/

You may substitute another compatible ComfyUI upscale model if desired.

YOU DO NOT NEED EVERYTHING

If you only use Standard checkpoint regional repair, you do not need to download FLUX, Qwen or FLUX.2 Klein.

If you want the FLUX regional-repair family, install the FLUX.1 Dev model, text encoders and VAE.

If you want FLUX Fill, add the Fill FP8 model.

If you want Qwen Head / Face Identity Transfer, add the three Qwen resources.

If you want Tattoo Photo Transfer, add FLUX.2 Klein, its Qwen3 encoder/VAE, and NocturnalWolfStudio's two tattoo LoRAs.

If you do not use the optional Stage 2 upscale, you do not need its additional processing beyond the basic upscaler.

This modular approach is intentional: download only the parts of the studio you actually intend to use.

MASK BASICS

For manual stations:

WHITE = allowed to change

BLACK or Unmasked = protected

GRAY = partial/blended influence

For most edits, paint the smallest practical white region around the defect and let the workflow's crop/context and feathering provide the surrounding information needed for a natural result.

RELEASE / USAGE PERMISSIONS

This workflow is distributed free of charge and should remain free.

You may use it to create images for personal or commercial projects.

You may modify it for your own use.

Free redistribution or free modified versions are welcome.

Do not sell this workflow, sell modified versions of this workflow, place it behind a paywall, or include it as a paid product/bundle.

Credit is appreciated, but not required.

These permissions apply to this workflow package only. All third-party checkpoints, diffusion models, LoRAs, detector models, custom nodes and other external resources retain their own licenses and terms. Users are responsible for complying with those licenses.

CREDITS

NocturnalWolfStudio

Special thanks and full credit to NocturnalWolfStudio for the original FLUX2 Klein — Steal Tattoo / Body Writing workflow and especially for creating the Extract Tattoo and Place Tattoo LoRAs that power the specialized Tattoo Transfer functionality incorporated into this studio. The original Civitai release identifies NocturnalWolfStudio as the creator and links both LoRAs as required components.

Visit NocturnalWolfStudio's original Tattoo Transfer release

Please support the original author and download their LoRAs from the original releases rather than redistributed copies.

Additional credit belongs to the authors of ComfyUI, Impact Pack, Impact Subpack, KJNodes, MultiLoRALoader, QwenEditUtils, Ultimate SD Upscale, Tiled Diffusion, YOLOE-26, the model creators, detector-model authors and everyone whose work makes a project like this possible.