( Not my Model )

https://civitai.com/models/3047149/paprika-dusk?modelVersionId=3464347

Chalky dry-brush paint over scratchy charcoal outlines, inside a torn black border. Faces are long and angular, with big noses, heavy eyelids and the look of someone who has stopped being impressed by anything. Warm orange lamplight cuts into deep teal shadow. It works on harbours, kitchens, night trains, market stalls, snowy streets, still lifes and animals as well as on people.

Trigger word: chk0dry

How to use it

Start every prompt with chk0dry, . After that, describe the picture in plain sentences, in this order:

1. The framing ("half-length view of", "close head-and-shoulders view of")

2. The face: age, skin, face shape, nose, eyes, brows, hair, expression

3. What she's doing, and her clothes

4. The setting, and where the light comes from

Leave out style words like "painting", "sketch", "gritty" or "illustration". The trigger carries the look. Style words pull Krea back toward its own finish.

Name the light source and its colour ("a brass lamp on the left lights her cheek orange") and let everything else fall into teal shadow. That split gives the strongest results.

Settings

Krea 2 Turbo: weight 1.2, 8 steps, guidance 1.Portrait 832 × 1216, landscape 1216 × 832. Most of the training images are landscape.

Tips

• Heavy lids, a long nose and a flat or skeptical expression get the most from this LoRA. Smiles work, but the faces lose some of their character.

• Give every face its own features: a bumped nose, a mole, big ears, freckles. Otherwise the faces drift toward the same pale woman with a black bob.

• Don't ask for signs, labels or posters. The LoRA tends to fill them with scribbled fake text.

• Scenes with no people in them work well. About a quarter of the training images are streets, rooms, still lifes or animals.

Training

58 images, each captioned by hand with content only: framing, faces, clothes, setting and light, with no style words. Trained on Krea 2 Base with Civitai's trainer, released at epoch 9.