Flimmer — open source video LoRA trainer for WAN 2.1 + 2.2

Full pipeline toolkit from raw footage to trained LoRA checkpoint. Early release, building in the open.

• WAN 2.1 + 2.2 (T2V and I2V)

• Scene detection, captioning, validation, pre-encoding, training — all in one place

• Phased training for multi-stage runs and WAN 2.2 MoE expert specialization

• Data prep tools work standalone with any trainer (kohya, ai-toolkit, etc.)

• RunPod + local GPU support

• LTX coming next

http://github.com/alvdansen/flimmer-trainer

Built by Alvdansen Labs. Early release — actively fixing bugs, feedback welcome.