Anima TrainFlow simplifies LoRA training with 3-click dataset prep
Smart cropping and auto-captioning in one tab—no more tool hopping for LoRA training.
Get AI news that actually matters
One email a day. Zero fluff. Join 10,000+ professionals.
Anima TrainFlow, created by Reddit user ThetaCursed, is a zero-tab LoRA trainer designed for anime-style image generation models like Anima. The latest update completes the entire pipeline by integrating dataset preparation directly into the single-page UI. Users now go from raw images to training in just three clicks: dump 20–100 images into a folder, click two buttons to prep the dataset, and hit start. This eliminates the typical friction of jumping between separate cropping, tagging, and training tools.
The tool adds two key AI-powered features. First, smart object-aware cropping powered by U^2-Net detects the main subject in each image and crops it intelligently to fit optimal training aspect ratio buckets—no heads or details get chopped off. Second, built-in auto-captioning uses the wd-eva02-large-tagger-v3 model running locally via ONNX to generate precise Danbooru-style .txt captions with adjustable tag thresholds. Anima TrainFlow also offers a pre-configured portable environment, live preview gallery, Prodigy learning rate handling, and is optimized for 6GB+ NVIDIA GPUs. It's an all-in-one solution that cuts prep time from hours to minutes for both beginners and experienced LoRA trainers.
- Smart object-aware cropping powered by U^2-Net dynamically resizes and rescales images to fit optimal training buckets, preventing head or detail cut-off.
- Built-in auto-captioning uses wd-eva02-large-tagger-v3 via ONNX for fast, accurate Danbooru-style tag generation with adjustable thresholds.
- Zero-tab UI combines dataset prep, tagging, and training controls in one portable, low-VRAM (6GB+) package—no complex Python setup required.
Why It Matters
Saves hours of dataset prep for LoRA training, making professional-grade model customization accessible to all skill levels.