NOT a furry-exclusive model.

NSFW + FurryPlus general (big) LoRA for MOSS Base. Works decently well on SFT for the non-Turbo LoRA.

Credit for the DMD2 Turbo (Lodestone) - https://huggingface.co/lodestones/Kromoss/blob/main/8-steps-dmd2-lora.safetensors.
NOT tested with Kromoss.

I HIGHLY SUGGEST USING THE TURBO SETUP, IT'S USUALLY WAY BETTER QUALITY.

Turbo (DMD2) Setup:

• 8, 12, or 16 steps, CFG 1 (slightly higher for more contrast/negative following)

• 1-1.5MP usually shows good one-shot results

• sgm_uniform/euler is fine but probably not optimal

• High-res refine at 0.7-0.9 for >2MP, up to 3MP tested, normal gen can work sometimes (can just copy the workflow in the images for something basic). CFG can be a little higher on it to prevent desaturation during refine.

Normal Setup:

• 40-50 steps, CFG 3 (not really thoroughly tested)

Strengths

• General NSFW, usually

• Better adherence with specific positions

• Realistic scenes and generally unique photography with depth

• Darkness

Prompting (not required, but helps coherency a LOT and unlocks better features)

1. Prose, 2 to 5 paragraphs, in plain, specific sentences. It names the species, the fur and skin colours, the pose, where each body part is, and exactly what is inside what. Anatomy is named directly: penis, shaft, knot, sheath, testicles, vulva, anus, nipples.

2. Tint line (optional): a short comma list for lighting, color and rendering. E.g. "high contrast", "warm yellow tint", "sharp lines", "bold lines", "soft lines", "extreme detail", "muted colors", any mix and match.

3. Tag line (optional): "by <artist>.", then pose, act, species and setting tags. Anatomy and penetration tags were removed from the tag line on purpose; that information lives only in the prose.

4. If you don't specify something in the prompt, it may not appear, even if assumed.

Prompting Specifics/Keywords

- Every entity you describe gets drawn. A back-reference such as "the one on the left" can produce a second copy of that character. Describe each character once, fully, and give positions in absolute terms.
- Be explicit about anatomy and placement.
- Open with a style. For anime: "Digital anime art style with clean line art and cel shading." For photos: "hyperrealistic, realistic."
End with a tint line, using the words the model was trained on, any combo of these:
- dim lighting, brightly lit, very dark scene, deep shadows, crisp highlights, high contrast, low contrast, pitch black
- vivid colors, muted colors, near monochrome, color wash, oversaturated colors, warm orange tint, golden tint, cool blue tint, purple tint, red tint, teal tint, olive tint
- flat cell shading, painterly shading, smooth gradient shading, bold lines, smooth clean lines, sketchy lines, lineless, textured, heavy texture, very smooth, very detailed
- clean composition, minimal composition, cluttered

Artists

• Not too many artists trained in depth. Here are the ones that MIGHT show an effect, from most examples trained on to least:

• by hakya11, by snowskau, by johnwulffe, by furflection, by braeburned, by mawmain, by dagasi, by cold-blooded-twilight, by r-mk, by drako1997, by feretta, by zackary911, by claweddrip, by zourik, by codyblue-731, by oro97, by truegrave9, by loimu, by hooves-art, by sincastermon, by keadanger, by the-minuscule-task, by tom fischbach, by pikajota, by dimwitdog, by miso souperstar, by magnetus, by jishinu, by dragonplayer, by securipun, by twang, by sijimmy456, by riska (artist), by nepentz, by felino, by backsash, by taran fiddler, by photonoko, by inu-sama, by cooliehigh, by angellsview3, by zaush, by jay naylor, by desertkaiju

Issues

• Extra tail/fingers/characters sometimes (mostly just a base model issue)

• Complex scenes or interactions can result in character parts or attributes blending

• Prompting an anthro female shark usually gives Nika Sharkeh unless you specify colors (whoops)

Credit in merges plz so I know if anyone has a use for it at all