ComfyPilotComfyPilot
模型库 / LoRA / Spinner Rack
Spinner Rack

Spinner Rack

3061126

LoRA 19 次下载 2 喜欢 更新于 2026-11-22

底模:Krea 2

style

下载模型 在源站打开

触发词

sp1nr4ck

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。

权重 sp1nr4ck_v1.safetensors 218 MB 下载

说明

Spinner Rack

作者:privatewolf

Spinner Rack | Krea 2

Mid-century crime paperback covers in loose, heavy oils. Women in satin gowns and trench coats, men in fedoras and double-breasted overcoats, and a lot of trouble between them: a slap in a restaurant booth, a revolver pointed at the reader, a match held to a cigarette, a whisper behind a hand. Nights are wet and foggy, lit by street lamps, neon and green-shaded card-table lamps, and the walls glow orange, mustard and red against teal shadow. Faces get the most attention, with heavy lids, sharp brows, red mouths and a sideways look that says she already knows how this ends. Trigger word: sp1nr4ck

How to use it

Start every prompt with sp1nr4ck, and write the rest the way the training captions read. Each caption is one plain paragraph in this order:

  1. Who is there and what they're doing, with a job for every hand
  2. The face: shape, brows, eyes, nose, lips and expression
  3. The hair
  4. The clothes, piece by piece
  5. The props and setting, with where things sit in the picture ("a lamp post glows at the left") None of the captions use style, medium or lighting-mood words. Leave out "oil painting", "pulp", "noir" and "vintage cover". Krea has its own idea of each, and it pulls the picture away from this set. The trigger carries the look. Template:
sp1nr4ck, a woman [DOING WHAT, WHERE], seen from the waist up, [WHAT EACH HAND HOLDS, NEAR HER FACE OR CHEST]. She [ONE CLEAR EXPRESSION]. She has [FACE SHAPE], arched [BROW COLOUR] brows, [EYE COLOUR] eyes, [NOSE] and [LIP COLOUR] lips. Her [HAIR COLOUR] hair [STYLE]. She wears [CLOTHES, PIECE BY PIECE]. [ONE THING BEHIND HER], [THE WALL OR SKY COLOUR].

Example:

sp1nr4ck, a woman stands on the landing of an iron fire escape at night, seen from the waist up, clutching a pair of silver high-heeled shoes to her chest with both hands. She looks up over her shoulder with wide eyes and her mouth open in a round gasp. She has a round face, arched light brown brows, blue eyes, a small upturned nose with freckles across it and coral lips. Her platinum blonde hair falls to her chin in a soft wave. She wears a silver beaded evening dress and a white fur stole slipping off one shoulder. Above her at the top right a man in shirtsleeves leans out of a yellow lit window, frowning down at her. Red brick behind her and teal fog.

Settings

Krea 2 Turbo: weight 1.0, 8 steps, guidance 1. If the brushwork comes out smooth, raise it to 1.2. If faces start to look alike, drop to 0.9. Portrait 832 × 1216, landscape 1216 × 832. Two thirds of the training images are portraits (2:3 or 4:5), the rest landscape, plus one square.

Tips

  • Frame waist-up or head-and-shoulders. Most of the training images are cropped that tight, with a big face and one or two props. Knees-up scenes with a lot going on behind the figure shrink the faces and lose the look.
  • Give every face its own expression, and make it a big one: a gasp, a laugh behind a glove, eyes closed, a grin, an eye-roll, a sulk. The set has all of these. If every prompt asks for a cool, scheming stare, every face comes out with the same scowl.
  • Write "arched brows". Words like "heavy", "straight", "drawn low" or "drawn together" stack into thick V-shaped brows on every woman. Save heavy brows for the men.
  • Keep the action in the hands, near the face or chest: holding something up, clutching, covering a mouth, whispering. Whole-body motions like climbing down a ladder or crouching tend to come out as a standing pin-up pose.
  • Keep the background to one or two things, such as a lamp, a lit window or a wall colour. Asking for a different story at the top, middle and bottom of the picture crowds the frame.
  • Name the props. Cigarettes, revolvers, dice, whisky glasses, telephones and banknotes all appear in the training set, so the LoRA knows them, and one well-placed prop tells the story faster than a pose.
  • Count the hands before you render. Cigarette holder, gun and drink in one prompt asks for three hands. Give each hand one job and say which hand does it.
  • Two-figure scenes work because a quarter of the set is couples. Put the second person close, at her shoulder or beside her, rather than far off in the background. Describe each figure in its own sentence, with what they're doing to each other.
  • Men are 7 of about 30 figures in the set. They come out rough, square-jawed and brooding. Give them their own face description, or they'll borrow the woman's.
  • Leave signs blank or out. Only one training image has readable lettering, so asked-for words will mostly come out garbled.
  • Krea ignores negative prompts at guidance 1. To keep something out, describe what should be there instead.

Training

21 hand-picked images: 11 single women, 5 couples or groups, 1 pair of women, 3 single men and 1 still life of a card table. The artist signatures were painted out before training. Each image was captioned by hand with content only: the action, a full face description, hair, clothes, props and setting, with no style or lighting words. Trained on Krea 2 with Civitai's trainer.

示例图