The 7 Best ComfyUI NSFW Workflows in 2026

You installed ComfyUI, loaded an uncensored checkpoint, and now you’re staring at a node graph wondering which pipeline to actually run. This post is the answer. Seven ComfyUI NSFW workflows cover essentially everything people generate locally in 2026: photorealistic txt2img, anime, img2img transformations, inpainting and decensoring, LoRA stacking, upscaling, and Wan image-to-video.

Every one of these exists as a free JSON you can download and drop into ComfyUI. None of them require wiring nodes from scratch. If you haven’t set up ComfyUI yet, do that first with our ComfyUI NSFW guide, then come back here to pick your pipeline.

Quick comparison before the details:

#	Workflow	Best for	Models	VRAM	Difficulty
1	SDXL txt2img	Beginners, anime, speed	Pony V6 XL, Lustify, Illustrious	8GB	Easy
2	Flux txt2img	Photorealism ceiling	Flux-Uncensored-V2 (FP8/GGUF)	12GB	Easy
3	Img2img	Restyling existing images	Any SDXL or Flux checkpoint	8GB	Easy
4	Inpainting	Fixing regions, decensoring	SDXL inpaint checkpoints	8GB	Medium
5	LoRA stack	Characters, styles, poses	Base checkpoint + 2-4 LoRAs	8-12GB	Medium
6	Upscale + detailer	Print-quality finals	4x-UltraSharp + FaceDetailer	8GB	Medium
7	Wan image-to-video	Animating stills	Wan 2.x Spicy variants	16GB+	Hard

1. SDXL Txt2Img: The Starter Workflow

Start here. ComfyUI ships a built-in txt2img template (Workflow menu, then Browse Templates) that runs any SDXL-family checkpoint with zero extra nodes. Load Pony Diffusion V6 XL or Illustrious XL for anime, Lustify or BigASP for realistic output, type a prompt, queue.

Why it’s first: SDXL checkpoints are fast, run on 8GB cards, and have the deepest uncensored ecosystem. Nearly every explicit LoRA on Civitai targets Pony or SDXL. The one thing to remember is prompting style: Pony and Illustrious want booru tags (“score_9, 1girl, …”) while realistic fine-tunes take normal sentences. Our NSFW prompts guide covers both formats.

Settings that matter: 1024x1024 resolution, 25-30 steps, CFG 6-7. Going higher on resolution without workflow 6 below just produces mutated anatomy.

2. Flux Uncensored Txt2Img: The Quality Ceiling

When SDXL output starts looking plasticky, this is the upgrade. Flux txt2img needs a few extra pieces (two text encoders in models/clip/, a separate VAE), so grab a ready-made JSON instead of adapting the SDXL template. The Flux workflows on Civitai and the official ComfyUI examples both work; swap the checkpoint for Flux-Uncensored-V2 or another NSFW fine-tune.

Skin texture, lighting, and hands come out noticeably better than anything SDXL produces. The costs: slower generations, 12GB VRAM minimum for quantized versions (FP8 or GGUF), and a smaller LoRA pool than Pony. Also keep CFG around 3.5. Flux at SDXL-style CFG 7 produces deep-fried colors.

Full install detail, including exact encoder files and hardware tables, is in the Flux uncensored local guide.

3. Img2Img: Transform What You Already Have

Img2img feeds an existing image through the model instead of starting from noise. Load the built-in img2img template, connect a Load Image node, and the single setting that controls everything is denoise: at 0.3-0.4 the output keeps the original composition with a style shift, at 0.6-0.75 the model reinterprets the image freely.

For NSFW use this is the restyle pipeline: turn a render into photorealism, a photo into anime, a clothed character design into an explicit version of the same character. It’s also the base for pose-matching, where you generate a new subject in the exact pose of a reference.

One legal line worth stating plainly: transforming images of real people into explicit content without consent is illegal in a growing list of jurisdictions, local processing or not. Our AI nudify laws guide has the current state of play. If you’d rather skip local setup for this workflow, our NSFW img2img tools roundup compares the hosted options.

4. Inpainting: Fix Regions and Decensor

Inpainting regenerates only a masked region while leaving the rest of the image untouched. In ComfyUI you mask directly on the Load Image node (right-click, Open in MaskEditor), and a dedicated SDXL inpaint checkpoint fills the masked area from your prompt.

Three jobs this workflow handles:

• Anatomy repair. Bad hands and broken details on an otherwise good generation. Mask, prompt the correct detail, regenerate until it lands.
• Decensoring. Removing mosaics and censor bars from hentai is the classic use, and it’s popular enough that we wrote a dedicated hentai decensoring guide with model picks and step-by-step masking.
• Outfit and detail swaps. Change clothing, add or remove elements, adjust expressions.

The skill is in the mask: feather the edges and mask slightly beyond the region you want changed, or you get visible seams. Expect a few attempts per fix. That’s normal, not a broken setup.

5. LoRA Stacking: Characters, Styles, and Poses

This is where local generation pulls away from every hosted tool. A LoRA stack chains multiple small adapter models onto one checkpoint: a character LoRA, a style LoRA, and a pose or concept LoRA at once. Hosted generators let you pick one style preset; a local stack gives you a specific character, in a specific art style, in a specific explicit scenario, reproducibly.

The workflow is just workflow 1 or 2 with LoRA Loader nodes chained between the checkpoint and the prompt encoders. The craft is in the weights. Start every LoRA at 0.7-0.8 strength and lower whichever one starts distorting anatomy. More than four stacked LoRAs usually degrades output, and mixing LoRAs from different base families (a Pony LoRA on a Flux checkpoint) produces garbage.

Civitai is where the LoRAs come from. Filter by your base model plus the NSFW tag, sort by downloads, and check sample images for the trigger words each LoRA expects. Our Civitai review covers how the site’s filters and on-site generator work.

6. Upscale and Face Detailer: The Finishing Pass

Raw 1024x1024 generations look fine as thumbnails and fall apart on a big screen. The finishing workflow fixes that in two stages: an upscale-with-model pass (the community standard 4x-UltraSharp, or a latent hi-res fix at 0.3-0.4 denoise for extra detail), then a FaceDetailer node from the Impact Pack that automatically finds faces and regenerates them at higher resolution.

Faces are where AI images fail inspection, and the detailer pass is the difference between output that reads as AI at a glance and output that doesn’t. The Impact Pack installs in one click through ComfyUI Manager, and searching “FaceDetailer workflow” on Civitai turns up complete JSONs that chain generation, upscale, and face repair in a single queue.

This workflow also chains onto any of the others. A common full pipeline is workflow 5 into workflow 6: LoRA-stacked generation straight into upscale and detail.

7. Wan Image-to-Video: Animate Your Stills

The heavyweight. Wan’s open-weight models turn a still image into a short video clip, and community “Spicy” fine-tunes handle explicit motion that the base model won’t. ComfyUI is effectively the only maintained local UI that runs the full pipeline, which makes this the workflow people ultimately build the whole stack for.

Requirements are steep: 16GB VRAM realistically, more for higher resolutions, and generation takes minutes per clip rather than seconds. The usual pattern is generating a strong still with workflows 2 and 6 first, then feeding it to Wan. The complete pipeline, model variants, and settings live in our Wan 2.7 NSFW video guide, and hosted options are compared in the NSFW image-to-video roundup.

Start here only after workflows 1 through 6 feel routine. Video multiplies every failure mode images have.

Where to Download These ComfyUI NSFW Workflows

Three sources cover everything above:

1. Civitai’s workflow section. Thousands of shared JSONs including explicit pipelines, sortable by rating. The fastest route to a working NSFW setup.
2. Embedded PNGs. Every ComfyUI-generated image carries its workflow inside the file. Drag any sample image from a Civitai model page into ComfyUI and the exact pipeline that made it loads instantly. Honestly the best feature in the app.
3. Built-in templates. The Workflow menu ships txt2img and img2img starting points that work with any SDXL checkpoint.

When a downloaded workflow throws red missing-node errors, open ComfyUI Manager and hit “Install Missing Custom Nodes.” That fixes nearly all of them.

If This All Sounds Like Too Much Work

Fair. The local stack pays off if you generate regularly and care about privacy or zero content restrictions. If you just want good explicit images tonight, hosted tools run these same pipeline types behind a prompt box: Seduced AI covers the photorealistic txt2img and img2img lanes with an extensions system that works like pre-built LoRA stacks, and Promptchan is the budget pick with a huge remixable gallery. The full field is ranked in our best NSFW AI image generators comparison.

FAQ

Are ComfyUI NSFW workflows free?

Yes. The workflows are JSON files shared free on Civitai, GitHub, and Reddit, and ComfyUI itself is open-source. Your only costs are hardware and the time to set things up.

Which workflow should a beginner start with?

Workflow 1, the built-in SDXL txt2img template with a Pony or Lustify checkpoint. It needs zero custom nodes and teaches the settings every other workflow reuses. Add one LoRA when that feels comfortable, then move to img2img.

Do these workflows have content filters?

No. ComfyUI applies no filtering, and uncensored checkpoints generate whatever you prompt. The legal responsibility sits with you: fictional subjects are fine in most places, real people without consent are not, per the laws covered in our nudify laws guide.

How much VRAM do I need?

8GB runs workflows 1, 3, 4, and 6 comfortably. Quantized Flux wants 12GB. Wan video wants 16GB and up. Under 8GB, stick to SD 1.5 checkpoints or use a hosted generator.

Can I combine these workflows?

Yes, that’s the point of the node system. LoRA stack into upscale and detailer is the standard image pipeline, and its output feeds the Wan video workflow. Chained JSONs for full pipelines are common on Civitai.

---

Setup first: ComfyUI NSFW Guide

Pick your checkpoint: Uncensored AI Models Ranked

Prompting reference: NSFW AI Prompts Guide

SDXL model detail: Stable Diffusion NSFW Guide

Video pipeline: Wan 2.7 NSFW Video Guide