How to Make Cel Shaded NSFW AI Art in 2026

To make cel shaded NSFW AI art, use an anime base (Illustrious, Pony, or WAI) with cel shading, flat shading, hard shadows, toon shading, thick outline, flat color, then keep refinement denoise low (0.2 to 0.3) so SDXL cannot sneak gradients back in. The trick is sharp shadow boundaries and flat fills, killing gradient bleed.

What the cel shaded look actually is

Cel shading is a rendering technique, not a genre. It reproduces the look of traditional cel animation and modern toon-shaders: color is applied in flat, unblended fills, and shadows are hard-edged shapes with a sharp boundary rather than a soft gradient falloff. A cel-shaded arm has a lit region of one flat tone and a shadow region of a second flat tone, with a crisp line where they meet, plus usually a clean dark outline around the whole form. Sometimes there is a third tone for a highlight or a deep core shadow, but the defining trait is discrete steps, not continuous blending.

This is different from broad cartoon art, which is a whole visual genre with its own proportions and could be shaded any number of ways. Cel shading is specifically about the hard-edge shading method: you can apply it to a realistic figure, an anime figure, or a stylized one, and the common thread is the flat fills and sharp shadow boundaries. It is closely related to the 3d toon-render look, where a 3D model is shaded with a cel shader to look hand-drawn, and to modern anime shading, though true cel is flatter than most current anime which sneaks in soft gradients.

Here is the core problem with AI and cel shading. SDXL and its anime derivatives are trained on enormous amounts of soft-shaded, gradient-rich digital art, so their strong instinct is to blend. Ask for cel shading and the model will often give you an anime image that is 80 percent flat but with soft gradient bleed creeping into every shadow edge, especially on skin and hair, softening the crisp boundaries that define the style. Winning at cel shading is mostly a fight against that gradient bleed: forcing flatness in the prompt, and then refusing to let refinement passes reintroduce softness.

It helps to be precise about what a clean cel shadow looks like, because that is your quality target. Picture a thigh under a single light. In true cel shading, the lit front of the thigh is one flat tone, then at some line the surface turns away from the light and the color jumps in a single step to a darker flat tone. There is no smooth ramp between them; the transition is a shape with a defined edge, almost like a coloring-book region. Highlights work the same way, a discrete bright shape rather than a soft glow. When you internalize that shadows are shapes with edges rather than gradients with falloff, you know exactly what to look for when you cull a batch, and you stop accepting the almost-flat results that quietly ruin the style.

Best checkpoints and LoRAs for cel shading

Anime bases already lean toward flatter shading than photoreal ones, so they are the right foundation. A dedicated flat-shading or cel-shading LoRA sharpens the boundaries further and is worth having.

Pick	Base	Why it works for cel shading
WAI NSFW Illustrious	Illustrious	Clean anime shading, sharp boundaries with a cel LoRA
Illustrious flat-art merge	Illustrious	Flexible, responds well to flat-color tags
Pony Diffusion art merge	Pony	Best NSFW pose control, add a cel-shade LoRA
NoobAI merge	Illustrious	Strong lineart, good outline response
Anime SDXL merge	SDXL	Solid default anime flatness

Add a cel-shading or flat-shading LoRA at 0.5 to 0.8 to harden the shadow edges and flatten the fills. A separate lineart or bold-outline LoRA at 0.4 strengthens the clean dark outline if your base leaves it thin. Keep any NSFW body LoRA at 0.4 so it does not soften the shading. For driving these bases, see how to use Illustrious models, the WAI NSFW Illustrious guide, and the Pony Diffusion guide. The best Illustrious checkpoints list has current merges, and best anime NSFW generators covers hosted options.

A caution on stacking: a cel LoRA and a strong outline LoRA can both push contrast, and together at high weight they sometimes crush the flat fills into harsh banding or blow the outline into a thick sludge. Start the cel LoRA around 0.6 and the outline LoRA around 0.4, generate, and only raise them if the specific quality is still missing. The base anime checkpoint should already be carrying most of the flatness, so treat the LoRAs as sharpeners rather than as the whole engine. If your fills start banding or the color posterizes unpleasantly, that is a signal your LoRA weights are too high, not too low.

The prompt: copy-paste positive tags

Lead with the shading method and repeat it in a couple of forms, because redundancy on the flatness helps overcome the model’s blending instinct.

cel shading, cel shaded, flat shading, toon shading, anime cel style,
hard shadows, sharp shadow edges, hard shadow boundary, two tone shading,
flat color fills, flat colors, minimal gradient, no gradient,
thick clean outline, bold lineart, crisp linework,
(portrait of an adult woman:1.2), curvy adult figure,
clean vector-like shading, high contrast shadows, poster style, clean

The load-bearing tags: cel shading, cel shaded, flat shading, toon shading stated together push hard on the method. hard shadows, sharp shadow edges, hard shadow boundary, two tone shading are the specific instruction that shadows must have a crisp edge, which is the single most important cel trait. flat color fills, minimal gradient, no gradient directly attack the gradient bleed. thick clean outline, bold lineart bring the defining dark outline. Keep the subject an explicit adult. For building the prompt structure, see the NSFW AI prompt formula. Naming a two tone or three tone shading count is a strong, underused tag because it tells the model exactly how many flat steps you want.

The number of tones is a stylistic decision worth making consciously. Two-tone shading, one light and one shadow, is the flattest and most graphic, closest to a screen-printed poster or classic game sprite. Three-tone adds a deep core shadow or a highlight and reads a little richer while staying clearly cel. Go past three or four and you are drifting toward the soft-shaded look you are trying to avoid, because many small steps start to approximate a gradient. Pick two for a bold graphic result and three when you want a touch more form, and put that count in the prompt so the model does not default to its usual over-blended many-step rendering.

Negative prompt for cel shading

The negative is where you kill softness and blending, which is the entire battle.

soft shading, gradient shading, smooth shading, soft light, blurry shadows,
airbrushed, blended, painterly, realistic shading, subsurface scattering,
soft gradient, ambient occlusion gradient, 3d render photoreal, muddy,
photorealistic, sketch, rough lineart, lowres, blurry,
extra fingers, deformed hands, bad anatomy, watermark, signature, text

soft shading, gradient shading, smooth shading, soft gradient are the primary weapons against gradient bleed, telling the model in several ways not to blend the shadow edges. airbrushed, blended, painterly reinforce that. soft light, blurry shadows keep the shadow boundaries crisp. realistic shading, subsurface scattering stop the model applying photoreal skin light that always softens edges. sketch, rough lineart keep the outline clean rather than scribbly. The negative prompts master list covers which tokens actually pull.

Settings: sampler, CFG, steps, resolution

Cel shading benefits from slightly higher CFG than a painterly style because you want firm, decisive edges, but not so high it fries the flat fills into oversharpened noise.

Setting	Value	Note
Sampler	DPM++ 2M Karras	Clean, decisive edges
Alt sampler	Euler a	Slightly softer if edges get harsh
CFG	5.5 to 7	Firm edges without frying flat fills
Steps	26 to 32	Enough to resolve the outline cleanly
Resolution	832×1216 or 896×1152	Portrait suits character art
Clip skip	2	Standard for Illustrious and Pony
Hires fix	1.5x, denoise 0.25	Low denoise so gradients do not creep back

The critical setting is the hires and img2img denoise: keep it at 0.2 to 0.3. Every point higher gives SDXL a chance to repaint the shadow edges with soft gradients. The CFG and sampler settings guide explains the CFG tradeoff. Steps also matter less than you might expect here: because the target is flat color rather than fine texture, piling on steps mostly wastes time and can over-refine the edges, so the modest 26 to 32 range is plenty for a clean outline and crisp boundaries.

Step-by-step workflow

1. Set the base. Load WAI or an Illustrious merge plus a cel-shading LoRA at 0.7, clip skip 2. Confirm the flat-shading tags are front-loaded.
2. First txt2img pass. Paste both blocks, DPM++ 2M Karras, CFG 6, 30 steps, 832×1216. Batch 6 to 8 and select a seed where the shadow edges are visibly crisp and the fills are flat, not the ones with soft gradient creep.
3. Judge the edges. Zoom into a shadow boundary on the arm or thigh. If it fades gradually, that seed lost; if it steps sharply, keep it. This edge check is the whole quality bar for cel shading.
4. Hires pass at low denoise. Enable hires fix at 1.5x with denoise 0.25. Going higher here is the number one cause of gradient bleed sneaking back in, so resist the urge to raise it for more detail.
5. img2img only if needed. If you refine in img2img, stay at 0.2 to 0.3 denoise with the full cel prompt and negative. The img2img guide covers keeping denoise low enough to preserve flatness.
6. Fix face and hands carefully. Run ADetailer on the face at 0.2 to 0.25 with the same cel prompt, because ADetailer at higher denoise loves to add soft realistic shading to the face and break the style. Inpaint hands via the inpainting guide at matched low denoise.
7. Control the pose. For a specific pose, drive it with ControlNet OpenPose or a lineart controlnet on the first pass so the flat pass only handles shading.

Where cel shading breaks, and the fix

Failure	Cause	Fix
Soft gradient shadows	Model blending instinct	Add soft shading, gradient shading, smooth shading to negative, push hard shadows
Gradient creeps back on hires	Denoise too high	Drop hires denoise to 0.25 or lower
Outline too thin or missing	Weak outline tags	Add thick clean outline, bold lineart, add a lineart LoRA at 0.4
Face goes soft-shaded	ADetailer denoise too high	Run ADetailer at 0.2, keep cel prompt in its settings
Too many tone steps, looks painterly	No step count named	Add two tone shading or three tone shading
Edges harsh and noisy	CFG too high	Drop CFG to 5.5, switch to Euler a
Looks fully realistic	Realism token leak	Add realistic shading, subsurface scattering, photorealistic to negative

The one failure that matters most is gradient bleed, and it has two cures: aggressive soft-shading negatives, and low denoise on every pass after the first. Get those right and the rest falls into place. For a broader diagnostic order, how to get better NSFW AI results is the reference.

Preserving flatness through refinement

The hardest part of cel shading is not the first generation, it is keeping the flatness alive through the refinement steps that normally improve an image. Upscaling, img2img, and face-detailing all give the model another chance to blend, and by default it will. The mental model to hold is that after your first clean cel render, every subsequent pass should be as low-denoise as you can get away with, because you are refining, not repainting. If you need more resolution, prefer a shading-preserving upscaler over a heavy img2img enlarge. If a shadow edge softens during upscale, a quick fix is to run a lineart or flat-shading controlnet on the refinement pass so the model is anchored to the hard edges you already have. For the deepest fidelity work, how to add detail to NSFW AI images walks the upscale stack, and controlnet for NSFW AI covers using a lineart guide to lock structure. One more trick: if a finished image has a few soft edges you cannot fix in-model, a posterize or threshold adjustment in any editor snaps near-flat regions into truly flat ones, which is a fast manual way to enforce the cel look on a stubborn area.

When to level up

Once single cel frames come out crisp and flat, push toward consistency and control. If you want the same adult woman across a cel-shaded set, character consistency techniques keeps her design locked while you change poses and scenes. To capture your exact edge-hardness and outline weight in one trigger word, train a style LoRA on a curated set of flat cel references so every generation starts hard-edged. And because cel shading is uniquely sensitive to the refinement stack, mastering the low-denoise upscale workflow in how to add detail to NSFW AI images pays off more here than in almost any other style. Force the flatness, negate the softness, keep every pass low-denoise, and the clean toon-shaded look holds from first render to final export. Treat the shadow edge as the one thing you never compromise on, and everything else about the style is comfortably within reach.

Frequently asked questions

How do I stop SDXL from adding soft gradients to my cel shading?

Attack it from two sides. Put soft shading, gradient shading, smooth shading, soft gradient, blended in the negative, and push hard shadows, sharp shadow edges, flat color fills in the positive. Then keep every refinement pass at low denoise, 0.2 to 0.3, because each higher-denoise pass lets the model repaint the edges with gradients.

What denoise should I use for cel shading refinement?

Keep hires and img2img denoise at 0.2 to 0.3, and run ADetailer at 0.2 to 0.25. Denoise is the single biggest cause of gradient bleed returning after a clean first render. Higher denoise gives SDXL room to soften your hard shadow boundaries, so refine conservatively and prefer a shading-preserving upscaler over a heavy enlarge.

Which checkpoint is best for cel shaded NSFW art?

Anime bases like WAI NSFW Illustrious, Illustrious flat-art merges, or a Pony art merge, since they already lean flatter than photoreal models. Add a cel-shading or flat-shading LoRA at 0.5 to 0.8 to harden the shadow edges, and a bold-outline LoRA at 0.4 if the clean dark outline comes out thin.

How is cel shading different from cartoon or 3d toon art?

Cartoon is a whole genre with its own proportions and could be shaded any way. Cel shading is specifically the hard-edge shading method: flat unblended fills and sharp shadow boundaries. The 3d toon look applies a cel shader to a 3D model. Cel shading is the technique, and you can apply it over realistic, anime, or stylized figures.

Why does my cel shaded face come out soft while the body stays flat?

ADetailer is the usual culprit. At higher denoise it repaints the face with soft realistic shading and breaks the style. Run ADetailer on the face at 0.2 to 0.25 and put your full cel prompt and soft-shading negative into the ADetailer settings so the refined face inherits the flat, hard-edged shading instead of turning painterly.

How do I get a strong clean outline around the figure?

Add thick clean outline, bold lineart, crisp linework to the positive, and if the outline still comes out thin add a dedicated lineart or bold-outline LoRA at 0.4. A lineart controlnet on the refinement pass can also enforce a consistent outline. Avoid sketch and rough lineart in the positive, since those give a scribbly rather than clean line.

What CFG and sampler suit cel shading?

Use CFG 5.5 to 7 with DPM++ 2M Karras for firm, decisive edges. Cel shading wants crisp boundaries, so slightly higher CFG than a painterly style helps, but above 7 the flat fills can fry into oversharpened noise. If edges get harsh and noisy, drop to 5.5 and try Euler a. Clip skip 2 is standard on anime bases.

Can I fix soft edges after generating without regenerating?

Yes. If a finished image has a few stubborn soft edges, a posterize or threshold adjustment in any editor snaps near-flat regions into truly flat color, manually enforcing the cel look. You can also run a low-denoise img2img at 0.25 with a lineart controlnet anchoring the hard edges. Both salvage a near-miss faster than hunting a new seed.