Something Unlimited Style - Anima v3.0 for Anima

Something Unlimited Style - Anima - v3.0

NSFW

Something Unlimited Style (Anima)

An anime style LoRA for the clean, vibrant look of Something Unlimited art. Choose a version from the tabs at the top. Each version's About this version box has its full details.

Versions

v3.0 (recommended). Trigger @sustyle. Retrained on the cleaned and re-captioned v1 dataset, so it keeps v1's SU look and v2's flexibility. New in v3.0:
- SU lips on any character: add lips right after @sustyle, for example esdeath, akame ga kill, @sustyle, lips,. Anime characters that usually get a plain line for a mouth get SU-style lips, and the rest of the image stays the same. For a stronger look, use red lips or lipstick.
- Darker men: (dark-skinned male:1.3), black man, very dark skin, interracial. Leave out male skin tags to get the usual tan SU man.
- Dark-skinned women: dark-skinned female, plus very dark skin for deeper tones.
- Colored skin: colored skin plus a color, such as orange skin, green skin, grey skin, blue skin or purple skin. This works for characters like Starfire, Gamora, Raven, Mystique and Ahsoka.
v2.0 (experimental). Trigger @sustyle. Claude Code (an AI assistant) picked the images, wrote the tags and chose the epoch.
v1.0 (original). Trigger sustyle, which also works without the trigger. If 1boy brings in Lobo, add (sulooboo:1.2) to the negative.

Settings (all versions)

LoRA strength: 0.85. Use 0.7–0.75 for furry characters, and up to 1.0 for the strongest style.
Euler a, Normal, 28 steps, CFG 4, Shift 3, 832×1216 or 1216×832.
Tested in Forge Neo and ComfyUI. In ComfyUI, use euler_ancestral + normal and load the LoRA in a LoRA loader.

Prompt layout (one idea per line, top to bottom; works for v2.0 and v3.0, and for v1.0 if you use sustyle):

masterpiece, best quality, score_9, score_8, safe,
<people count>
<character>, <series>, @sustyle, <hair, eyes, body>,
<pose>,
<outfit>,
<expression>,
<scene>,
<camera>

Change safe to sensitive, nsfw or explicit for spicier pictures.

Negative prompt

worst quality, low quality, lowres, score_1, score_2, score_3,
blurry, jpeg artifacts, watermark, signature, text, logo, (japanese text:1.2),
realistic, extra arms, extra legs, cropped feet, (emphasis lines:1.2), speech bubble

Guide and workflow: Something Unlimited Style (Anima) Guide covers prompts, expressions, head angles and inpainting. The ComfyUI workflow is in the guide's Attachments.

Share your creations: post your images on this page and leave a review. Feedback helps the next update.

Description

v3.0

Trained from my original v1.0 Something Unlimited dataset, cleaned up and re-captioned, to keep v1's SU look with v2's flexibility.

Trigger word: @sustyle

Tested in: ComfyUI (Anima_SU workflow, embedded in every preview image) and Forge Neo.

Recommended settings

Checkpoint: Anima (tested on dustVisionAnima v1.0)
LoRA strength: 0.85. Use 1.0 for the strongest SU look; poses stay flexible at both.
Sampler: Euler a (euler_ancestral), Schedule: Normal
Steps: 28, CFG: 4, Shift: 3
Size: 832×1216 or 1216×832

New in v3.0

SU lips on any character. Anime characters often come out with a plain line for a mouth. Add lips right after @sustyle to get SU-style drawn lips without changing anything else. Use red lips or lipstick for a stronger look.
Darker men. Ask for a darker man with (dark-skinned male:1.3), black man, very dark skin, interracial. Leave out male skin words to get the usual tan SU man.
Dark-skinned women. Use dark-skinned female, and add very dark skin for deeper tones.
Colored skin. Use colored skin plus a color: orange skin, green skin, grey skin, blue skin or purple skin. This works for characters like Starfire, Raven, Gamora, Mystique and Ahsoka.
3D game characters. Characters from 3D games (Ada Wong, Lara Croft, Tifa...): add 3d to the negative.

Prompt layout (one idea per line, top to bottom):

masterpiece, best quality, score_9, score_8, safe,
1girl, solo,
<character>, <series>, @sustyle, lips, <hair, eyes, body>,
<pose>,
<outfit>,
<expression>,
<scene>,
<camera>

Change safe to sensitive, nsfw or explicit for spicier pictures.

Training: 315 images, 4,000 steps, 8 epochs (epoch 8 released), dim 32 / alpha 16, LR 1e-4, cosine.