Anima TrainFlow: Windows setup and LoRA training

This tutorial uses Anima TrainFlow to train a character LoRA for Anima with four images. The speaker describes Anima as an anime illustration model with a model file of about 4.2GB, and says it appears less suited to live-action images. The model, text encoder and VAE require separate files and paths.

Anima TrainFlow wraps sd-scripts in a simpler interface that hides many training parameters. Its default Prodigy optimizer sets the learning rate automatically. The Windows setup covers cloning the repository, installing the Python environment through a batch file, downloading captioning and cropping models, and launching the WebUI. A portable package is also described.

The training walkthrough sets the model paths, image and caption folder, and the trigger word "Z-chan". It uses 2,400 steps with a preview every 300 steps. The speaker combines tags with natural-language descriptions and recommends checking captions first when results do not match expectations; detailed caption creation is left for another video.

The demonstrated run uses an RTX 4070 Ti with 12GB of VRAM and finishes in 1 hour 41 minutes. The speaker reports about 8GB of VRAM usage while a video editing environment is also open. TrainFlow reportedly supports 6GB, but this run does not test that minimum.