模型库 /
LoRA /
Podkin
触发词
p0dk1n
文件清单 线路:原始(huggingface.co) · 切换
下载后放进对应的 ComfyUI 目录。models/ 指 ComfyUI 的模型根目录。
说明
Podkin
作者:privatewolf
Podkin: Gloomy Little Folk Small sculpted figures with oversized heads and tiny bodies, in gloomy, whimsical scenes: rainy streets, candlelit rooms, foggy piers, snowy forests at night. Faces have heavy-lidded eyes, long noses and strong expressions. Settings are lit by lamps, moonlight or dusk. Trigger word: p0dk1n Base model: Trained on Krea 2 Base. Works on Krea 2 Turbo. Recommended settings
- LoRA strength: 1.0. Drop to 0.8 if faces come out too young or too stylised.
- Sizes: 1024 x 1536 (2:3), 1216 x 1536 (4:5), 1536 x 1024 (3:2), 1536 x 1536 (1:1) How to prompt Start with p0dk1n, then describe the content in plain sentences: who is in the picture, their hair, eyes, brows and mouth, what each hand is holding, what they're wearing, and the setting and light. Leave out style words like "3D", "clay", "toy", "render" or "stop-motion". The training captions never used them, so the trigger already carries the look. Adding them pulls the result toward generic versions of those styles. For adult characters, write "young woman in her twenties with slim adult proportions" or "old man". Big heads on small bodies tend to read as children unless you state the age. Give each face its own expression ("one brow raised, crooked smirk", "eyes shut, cheeks puffed"). Without one, faces drift toward a blank stare. Dataset 28 images, 1536px on the long edge, each with a hand-written caption that describes content only. About half are round creatures and half are human characters, so the LoRA does both.
Known limits
- Lettering comes out as made-up glyphs. Keep signs and labels out of prompts.
- Creatures sometimes appear unprompted in scenes with people. Name everyone who should be in the picture.
- Hands holding several objects can merge. Give each hand one job.
