Skip to content
FlixMLFlixML Docs

Training a LoRA

Training config templates, caption files, and dataset folders for character LoRA fine-tuning.

Source of truth for hyperparameters: RunComfy FLUX.2 LoRA training guide.

What’s in this directory

Section titled “What’s in this directory”

Templates (provided):

  • flux2_character.yaml — FLUX.2-dev character LoRA template
  • wan22_i2v_character_template.yaml — Wan 2.2 image-to-video character template (untested)

You create these before training:

  • datasets/{character_id}/ — caption files (.txt) per training image

Generated at runtime (do not edit):

  • config/ — job configs generated from templates per training run
  • output/ — LoRA checkpoints and sample previews

Workflow

Section titled “Workflow”
  1. Mark 30–60 training images per character in the FlixML Studio gallery (toggle “Include in training dataset”).
  2. Write one caption file per image in datasets/{character_id}/{image_stem}.txt.
  3. Start training from the UI or via POST /api/lora-training/start.
  4. Monitor samples and download checkpoints when done.

Trigger word

Section titled “Trigger word”

Short, unique, non-dictionary token.

  • Good: ch4rtrig, xy_char01, midnight_tarot
  • Bad: sam, ana, alex, or the character’s real name

Common words fight the base model and dilute identity learning.

Captions

Section titled “Captions”

Format:

<trigger>, a person, <scene description>

Example:

ch4rtrig, a person, sitting at a wooden desk in a navy blazer, soft side lighting, medium shot

Rules:

  • Start with <trigger>, <class word>, where the class word is the generic category of your subject (e.g. a person, a robot, a creature). For FlixML character LoRAs this is usually a person.
  • Describe visible scene details (pose, clothing, setting, lighting, framing).
  • Do not describe identity-specific features (face shape, eye color, hair color).
  • No quality tags (masterpiece, 8k, photorealistic).
  • Keep it to 15–30 words, neutral tone.

The class word lets the base model handle the generic concept while the trigger absorbs identity.

Starting training

Section titled “Starting training”
POST /api/lora-training/start
{
  "job_name": "character_v1",
  "trigger_word": "ch4rtrig",
  "character_id": "<character uuid>"
}

The backend reads captions, verifies each marked image has a caption, uploads everything to ai-toolkit, generates the job YAML from flux2_character.yaml, and starts the run.

Monitoring

Section titled “Monitoring”
  • GET /api/lora-training/jobs — list jobs and status
  • GET /api/lora-training/jobs/{name}/samples — training previews
  • GET /api/lora-training/jobs/{name}/checkpoints — downloadable LoRA weights

Templates

Section titled “Templates”

flux2_character.yaml and wan22_i2v_character_template.yaml follow the defaults in the RunComfy guide. Most settings should not be changed without reading the guide first.

References

Section titled “References”