ComfyPilotComfyPilot
模型库 / LoRA / Sour Greens
Sour Greens

Sour Greens

3052875

LoRA 0 次下载 更新于 2026-11-15

底模:Krea 2

style

下载模型 在源站打开

触发词

v3rd1gr1s

文件清单 线路:镜像(hf-mirror.com) · 切换

下载后放进对应的 ComfyUI 目录。models/ 指 ComfyUI 的模型根目录。

权重 sourGreens_v1.safetensors 218 MB 下载

说明

Sour Greens

作者:privatewolf

Sour Greens | Krea 2

Scratchy ink and gouache caricatures on worn, chipped board. The grounds are sea-green, teal, mint and olive, with hot pink in the cheeks, noses, ears and the odd dress or sofa. The faces do the work: long hooked noses, hollow cheeks, bulging or hooded eyes, necks like stalks, and hair that explodes off the head. Most of the cast are women, young punks and old matrons alike, and they get the full caricature treatment rather than a pretty face with a scratchy finish. It handles groups well. Eleven of the 18 training images have two to five people in them, each pulling a different face, so you can cram a tram, a theatre box or a witches' cauldron and expect every face to stay distinct. Trigger word: v3rd1gr1s

How to use it

Start every prompt with v3rd1gr1s, and write the rest the way the training captions read. Each caption is plain description: the framing, then each face in turn, then the clothes and the background colour. None of them has style or medium words. The trigger carries the ink and gouache, and adding "caricature" or "gouache painting" pulls Krea toward its own idea of those. Describe each expression by what the brows, eyes and mouth are doing. "Furious" on its own gives a mild frown. "Brows slammed together, teeth clenched and bared, nose wrinkled" gives fury. Template for one figure:

v3rd1gr1s, [head-and-shoulders / half-length] view of [WHO SHE IS] with [ONE EXAGGERATED FEATURE: a long hooked nose / a very long thin neck / hollow cheeks], [HAIR]. [BROWS], [EYES], [MOUTH]. She wears [CLOTHES, mostly black, one hot pink item]. [Sea-green / teal / mint / olive] background.

Template for a group:

v3rd1gr1s, [NUMBER] [WHO] [WHERE / ARRANGEMENT], cropped at the chest. The left one [FACE]. The middle one [FACE]. The right one [FACE]. [SHARED CLOTHES OR HAIR]. [COLOUR] background.

Example:

v3rd1gr1s, two young women at a small cafe table. The left one leans in and whispers behind her raised hand, her eyes slid sideways and glinting. The right one has her brows shot up, her eyes popping and her mouth dropped open in a round O, a coffee cup frozen halfway to her lips. Black and hot pink dresses, spiky black and white hair. Teal background.

Settings

Krea 2 Turbo: weight 1.0, 8 steps, guidance 1. If the board texture and scratchy line come out weak, go up to 1.2. If faces slide toward the training cast, drop to 0.9. Portrait 896 × 1344 for one figure, landscape 1344 × 896 for two or three, 1344 × 768 for a crowd of four or five.

Tips

  • Give every woman one exaggerated feature. Without one, Krea smooths her into a pretty face and only the men and old women look like caricatures.
  • In a group, name each face left to right and give each a different expression. Four faces with no instructions drift toward the same blank stare.
  • The set leans on vacant, bug-eyed and deadpan looks. Those come easily. Fury with bared teeth, worried brows, a wink and disgust are thin in the training data, so spell them out feature by feature and expect to try a few seeds.
  • Keep clothes mostly black with one hot pink item. A whole outfit in bright colours fights the green ground.
  • Krea ignores negative prompts at guidance 1. To keep something out, describe what should be there instead.
  • Leave out photo words like "photograph", "realistic" or lens terms.

Training

18 images made with a Midjourney moodboard, picked for clear, readable expressions: 7 single figures and 11 groups of two to five. Signatures were cloned out before training. Each image was captioned by hand with content only: framing, each face's brows, eyes and mouth, hair, clothes and background colour, with no style words. Trained on Krea 2 with Civitai's trainer.

示例图