LTX Director 2.0: ComfyUI video-to-video on 8GB VRAM

This tutorial explains how the speaker built a roughly one-minute drama from six clips using LTX Director 2.0 inside ComfyUI. The local AI workflow starts with image-to-video generation, then uses video-to-video extensions and a timeline to connect shots. To preserve the character's voice, the speaker includes source audio containing dialogue and enables the audio input switch.

On a system with 8GB VRAM and 32GB RAM, the speaker reports limits of eight seconds at 1080p, 12 seconds at 900p and 19 seconds at 720p. The reported generation time for a 19-second, 25fps clip at 720p is about 10 to 12 minutes. These are results from the speaker's setup, rather than guaranteed hardware limits.

The walkthrough separates motion prompts from scene layout: text directs behavior, while guide frames establish locations and camera angles. A failed attempt to reveal a cottage through a continuous camera move explains why the speaker chose a cut and a new reference frame.

Setup guidance covers updating ComfyUI and its nodes, global and segment prompts, and an FP4 text encoder. The speaker says the GGUF encoder did not support the node during testing. Fast motion, hands and wide shots still produce errors; IC LoRA support is mentioned but not tested here. The speaker recommends short sections and simpler camera moves, with cloud generation reserved for difficult shots.