Krea 2 Thara — v1
A South Indian woman in her mid-thirties, built as a character, not a look-alike. Fictional: she does not exist and no real person's images were used. Trained for continuous storytelling — the same face at every framing: temple silks, gold borders, paddy fields, market mornings.
Companion to Krea 2 Revathi from the same studio; the two are different women and different builds.
Use
Trigger: zxthara. Always add her age — the model reads it:
zxthara woman, 35 years old, <framing>, <expression>, <hair>, <outfit>, <light>, <place>
Base
Krea 2 Turbo (trained on Krea 2 Raw, run on Turbo)
LoRA strength
1.0
Steps / CFG
8 steps, CFG 1, Euler / Simple
Size
1024x1344 or 1152x1440 native
Say "photograph" in the prompt, not "photorealistic". Candid lines ("raw phone photo, … natural unpolished skin texture, slight grain") give the best skin.
Measured, not claimed
Identity is scored with InsightFace (buffalo_l) against five reference pictures verified by hash to be outside the training set — a mean cosine of 1.0 would be the same face.
18-prompt eval on Krea 2 Turbo, 768x1024, 8 steps, CFG 1, strength 1.0:
slot
score
close-up
0.9
waist-up
0.84
head & shoulders
0.9
full body
0.75
wet hair
0.9
newspaper
0.75
profile
0.74
selfie
0.88
mean (16 trigger prompts)
0.832, 16/16 above 0.60
30 new scenes at 1152x1440, one seed each, no cherry-picking: mean 0.867 over the 26 rendered, 26 of 26 above 0.75.
Training
58 images, one source, fictional; every visible body detail captioned. Rank 32 / alpha 32, 768/1024 buckets, 2,000 steps on Krea 2 Raw, text encoder untrained.
Known limitations
HOW TO PROMPT HER
Describe the scene, not her. Krea 2 is sensitive to descriptors: eye, face or body words change her proportions. Write the trigger, then clothes, pose, place, light, camera. Her face, dark brown eyes, skin, hair and nose stud come with the trigger; do not restate them.
Her build. On Krea 2 Turbo the trigger carries her figure. On other Krea 2 finetunes (Sick Ollie, CyberRealistic) v1.0 can render her slimmer, and her build can vary between scenes. Body words ("curvy figure, full bust, full hips") steady one shot but do not make her consistent; treat that as a defect of this version, not a setting. The next version is being built to carry her build in the trigger on every base, and its card will carry a measured consistency figure.
Makeup. She is built with little makeup, so lipstick words are weak at strength 1.0; use 0.8–0.9 for makeup looks.
Nose stud. Part of her, on purpose. "no nose stud" removes it about half the time.
Base. Trained on Krea 2 Raw, made for Krea 2 Turbo. Reviewer-tested on Sick Ollie INT8 (Forge Neo, Euler, 9 steps, CFG 1); results on finetunes will be stated here as numbers.
She leaks without the trigger. A plain "a 35-year-old South Indian woman, …" still renders her (0.90 / 0.81 against the references). For a different woman, don't load this LoRA.
The trigger word can print as text on signs, boards and newspapers — keep readable print out of the scene.
She is a full-figured woman by design; prompting her slim fights the model.
Her face is designed to read 35.
Licence and credit
This is a Derivative of Krea 2. NOTICE.txt and the Krea 2 Community License Agreement ship with the file and travel with any copy — including its Acceptable Use Policy, content-filtering duty and revenue threshold. Not an official Krea product; not endorsed by Krea.
She is written as a 35-year-old woman. Do not use her to depict, imply or impersonate a real person.
Made for dearai.live, where she lives as a character. Credit appreciated if you post her work.
Description
First release. Trigger zxthara, strength 1.0, 8 steps, CFG 1. See the model description for settings and measured scores.
Comments (4)
Hi, there's no model file for download (except the doc)
Thanks for flagging — the model file didn't attach on upload. It's up now (Krea_2_Thara_v1.safetensors, 218 MB).
@dearai_studio alright thanks - already downloaded your other model, looks like a very unique way (natural thicc Indian MILF type). Gonna have to review them ASAP!
(I'll also delete this post once you've read)
@LurkingAI @LurkingAI Thank you — that's exactly the idea: consistent South Indian characters with real builds, the same face at every framing. Would love to see what you make with them; if you post a review with pictures I'll add the best ones to the showcase.



















