Flux LoRA: AI Slop - 3D Animation Style (v2) πŸš€

This repository is now fully open source. We've included the v2 model, the full training dataset, and the configuration files so you can reproduce or build upon this style.

A style LoRA trained for the "AI Slop" aesthetic - vibrant 3D animation (Pixar/CGI style) with expressive characters and candy-like colors. Perfect for YouTube Shorts and kids content.

πŸ†• What's New in v2

  • Fixed subject bleeding - v1 had issues where subjects would sometimes take on pug-like features.
  • Diverse training set - Trained on 70 images with various animals (cats, dogs, pandas, raccoons, etc.).
  • Pure style learning - The LoRA now captures only the visual style, not specific animal features.

πŸ”“ Open Source Contents

This repository contains:

  • /dataset: The full set of 70 images used for training (diversified subjects).
  • /config: The training configuration file (config.yaml).
  • flxdev_lora_v2.safetensors: The latest stable weights.
  • optimizer.pt: Included to allow continuing the training exactly where we left off.

πŸš€ Quick Start

Parameter Value
Trigger word flxdev_custom
Base Model Flux Dev
Recommended Strength 0.5 - 0.6
Guidance Scale 3.5
Best for Vertical images (9:16 for Shorts)

⚑ Strength Guide

Same prompt with different strengths (seed: 42):

Strength 0.4 Strength 0.6 Strength 0.8
0.4 0.6 0.8
Subtle Recommended Strong

πŸ“ Example Prompts

Panda

flxdev_custom, a cute panda with huge sparkling glossy eyes eating bamboo, 3d cgi animation style, vibrant forest background

Cat

flxdev_custom, a cute fluffy orange cat with huge sparkling eyes playing with colorful yarn ball, 3d cgi pixar style, cozy room

βš™οΈ Training Hardware

Trained on an RTX 3090 via Vast.ai using the Flux Dev training pipeline.

πŸ“„ License

This project is open-source under the MIT License.

Downloads last month
2
Inference Providers NEW

Model tree for Eiby777/flxdev-lora-aislop

Adapter
(36488)
this model