// Dataset Library
Training datasets,
ready to use.
Every LoRA on neuraldrift was trained on one of these datasets. Curated, captioned, and structured for Kohya SS. Download and start training in minutes.
01
Download dataset
Get the image + caption ZIP
02
Check captions
Verify trigger word is first in every .txt
03
Configure Kohya
Use the linked training guide
04
Train your LoRA
Follow the guide for optimal settings
Fat Bigfoot — Character Dataset v4
42 curated images of Fat Bigfoot in varied settings, poses, and lighting. WD14-captioned with trigger word. Used to train the Fat Bigfoot Character LoRA v4.
fatbigfootTraining Notes
Repeat count: 10. Best results at 8–10 epochs with SDXL + Kohya SS.
Slacker Alien — Character Dataset v2
38 images of the Slacker Alien character in suburban environments. Diverse angles and expressions. Captioned with WD14 + manual edits.
slackeralienTraining Notes
Repeat count: 10. 8 epochs optimal. Strong character lock at 0.75 LoRA weight.
Highway Ghost — Style Dataset v3
85 cinematic night highway reference images. Rain-soaked asphalt, neon light reflections, motion blur, and dramatic street photography. Used for the Highway Ghost style LoRA.
highwayghostTraining Notes
FLUX natural language captions. Rank 32, 10 epochs. Style transfers well to LTX Video.
Desert Pursuit — Style Dataset v1
64 golden hour desert chase reference images. Dust trails, dramatic shadows, wide open landscapes, and action cinematography aesthetics.
desertpursuitTraining Notes
Still in beta — works well for still images, video transfer WIP.
ComfyUI Node Graph — Concept Dataset
30 dark tech node graph images. Glowing connections, dark backgrounds, cyan accents. Used to train the Node Graph concept LoRA for tech thumbnails.
nodegraphTraining Notes
Small dataset — concept LoRAs need fewer images. Rank 8 works best.
GoPro POV Fisheye — Style Dataset v2
120 handheld GoPro POV images with fisheye distortion. Action sports, forest runs, vehicle mounts. Covers wide variety of POV angles.
gopropovTraining Notes
Larger dataset = subtle style. Use 0.55–0.7 weight for best fisheye effect.
// Coming Soon
Community dataset uploads
Submit your own curated datasets. Get them verified, listed here, and linked to your trained LoRAs.
Start with the Training Guide →