LoRA Training

Train a Flux LoRA

Fine-tune a LoRA adapter on Flux.1 from a small image set — for a style, subject, or product look you want Flux to reproduce consistently.

Estimated $3–$12 per training run · 20–60 min

What inputs you need

  • 10–30 training images
  • Captions (auto or manual)
  • Training config (rank, steps, learning rate)

What Badgr returns

  • Trained LoRA safetensors file
  • Training loss curve / checkpoint log
  • Sample generations at checkpoints

Recommended GPU routes

Estimated cost: $3–$12 per training run · Estimated runtime: 20–60 min

Example command

badgr run "python train_lora.py --base flux1 --config flux_lora.yaml" \
  --gpu A100_80GB \
  --max-cost 12 \
  --max-runtime 90

Common setup failures Badgr avoids

Out-of-memory during gradient accumulation on large Flux weights — default config uses 80GB routes and gradient checkpointing.

Training exceeds budget before convergence — Badgr checkpoints periodically so a capped run still yields a usable LoRA.

Ready to run this?

Launch from the dashboard, CLI, or Compute API. Max-cost protection included.