LoRAs

A LoRA is a trained look — a fine-tuned style or character you can apply to image generation. Unlike skills and agents, LoRAs are personal: each is scoped to your account, so there isn't a public catalog. Train one, then attach it to a project and it's auto-applied to every render.

Your LoRAs

Connect your Daydream API key in Settings to see and manage your trained LoRAs.

Train a LoRA

# in the app:  /lora train  (upload 5–20 reference images)
# CLI:        livepeer lora train --images ./refs --trigger mystyle

Use a LoRA

# attach to a project — every create_media render uses it
attach_lora_to_project { "lora_id": "lora_…", "project_id": "proj_…" }

Agents can also bundle LoRAs (an agent's lora_refs apply on every render) — see the Agents tab.