Draw Things LTX-2: Mac and iPad video generation tutorial

This tutorial introduces the distilled LTX-2 model in Draw Things for audio-video generation. The presenter describes the app as free on Mac, iPad and iPhone, with the model available to download inside it. The walkthrough covers the app's first release with audio, so its settings and limitations reflect that release.

For local AI hardware context, the presenter reports generating clips of 3 to 4 seconds at almost 720p on an iPad Pro with 16 GB of memory. On an M2 Ultra Mac Studio, they report making 20-second videos, with rendering times ranging from about five minutes to over 30 minutes depending on clip length. These are personal results rather than hardware guarantees.

Image-to-video generation starts with a reference image covering the canvas; text-to-video starts with a cleared canvas. Both use the same model. The presenter recommends trying the Compression Artifacts setting when movement starts too slowly, saying it adds compression noise to the reference image. Audio-to-video and video-to-audio workflows were not implemented in the demonstrated release.

Prompt advice focuses on a scene-setting sentence followed by visible actions in present tense. Explicit stage directions, timing words and simultaneous actions help describe how characters should move. For dialogue, the presenter uses "they say" followed by quoted speech, but reports overlapping voices and swapped speaking roles without a reliable way to control speaker timing.