train

Manage the custom model training lifecycle. Actions cover dataset upload, training-image updates, pair mapping, cost estimation, and start/stop controls. action="start" is the ONLY action that launches training. It is paid and irreversible: the CU are charged the moment the job is created, not as epochs complete, and cancelling the run (action="stop") refunds nothing: a run cancelled at 0% still costs the full quote. Get the user's explicit go-ahead on the quoted CU before calling it. action="configure" prices a configuration and launches nothing. It always sends ?dryRun=true, so it never creates a job and never charges. To launch what you priced, repeat the call as action="start" with the same config. Args: - action: required — upload_images, update_image, delete_image, set_pairs, configure, start, or stop - model_id: required for every action - images: required for upload_images - training_image_id, image_data: required for the relevant update or pair actions - config: training hyper-parameters, required for action=configure, optional for action=start. epochs (the primary duration/cost lever; cost scales linearly), nb_repeats, batch_size, learning_rate. Per-family ranges and defaults are documented on each config field in the input schema. - dry_run: optional, only meaningful for action=start. true returns a cost estimate without starting the training job (sends ?dryRun=true). configure is always a dry run; dry_run=false there is rejected. - team_id, project_id: required for OAuth callers Returns: training-image updates, cost estimates, or training state changes depending on the action you call. Estimates (configure, or start with dry_run=true) carry creativeUnitsCost plus training_started=false and no job. Only a real action="start" returns a job; track it with jobs_wait or job_get. Examples: - "Upload training images" -> action="upload_images", model_id="model_xxx", images=["asset_xxx", "asset_yyy"] (upload local files with upload_asset first) - "Estimate cost of 20 epochs" -> action="configure", model_id="model_xxx", config={"epochs": 20} - "How much would this training cost?" -> action="configure", model_id="model_xxx", config={"epochs": 12} - "Start training with 12 epochs" (after the user approved the quote) -> action="start", model_id="model_xxx", config={"epochs": 12} - "Start training with defaults" -> action="start", model_id="model_xxx" Don't use when: You want to run inference on an existing model. Prefer model_run instead.

{
  "action": "upload_images",
  "model_id": "model_custom_abc",
  "images": [
    "https://cdn.example.com/ref1.png",
    "https://cdn.example.com/ref2.png",
    "https://cdn.example.com/ref3.png"
  ],
  "team_id": "team_abc123",
  "project_id": "proj_xyz789"
}

{
  "status": "uploaded",
  "count": 3
}