H3 Lip Sync on RTX 5090: 30-Second Video Takes 12 Minutes

R34vspec · reddit · 2026-08-14

The author shares benchmark results for generating a 30-second lip-sync video using the H3 model. On a single RTX 5090 GPU, the generation process took approximately 12 minutes.

Additionally, a practical prompting tip is provided: when inputting audio for lip-syncing, pasting the actual English lyrics into the prompt using the dialogue syntax <d> [English] Lyrics </d> significantly improves the final lip-sync accuracy and overall quality.

Original post →

More from Multimodal

Multimodal channel →