Developer TrackLesson 745 min

Training: LoRA, AV-LoRA, and IC-LoRA

By the end of this lesson you can
  • Name the four real training modes and pick the right one for a stated goal
  • Explain what IC-LoRA actually is and reject the identity-conditioning myth
  • Prepare a dataset that satisfies the resolution and frame-count constraints before a run starts
  • Size and budget a training run against the documented hardware tiers

Training is where the fabrication problem in the LTX ecosystem is worst, because training advice is expensive to check. A wrong endpoint fails in two seconds; a wrong training plan fails after you have rented a GPU for eight hours. Two myths in particular have propagated widely enough that they show up in AI coding assistants' output. Both are corrected here before anything else.

ltx-trainer is not on PyPI. Unlike ltx-core and ltx-pipelines, you use it from inside a clone of the Lightricks/LTX-2 monorepo, driven by uv:

Developer Track — locked

Unlock “Training: LoRA, AV-LoRA, and IC-LoRA”

The 5-lesson Foundation module is free and always will be — you have already read the part most guides get wrong. The paid lessons go deeper, with labs, knowledge checks, and every technical claim verified against live documentation.

LTX-230 — Developer Track
You write the integration
$149one-time
Save $49
LTX-230 — Complete Course
Both tracks, everything unlocked
$249one-time

Already bought it? Sign in with the email you used at checkout.