H3 Video Prompting Tip: Replace 'Subject' with Custom Tags

Hdfjds · reddit · 2026-08-14

A user shared a prompting tip for the H3 video generation model: when referring to elements in images, text, or videos, you don't need to rigidly use the <Subject N> tag.

Creators can use more semantic words like <Girl 1> or <Dialogue 1>. The post includes a complete example prompt demonstrating how to use custom tags combined with multimodal descriptions to precisely control multiple characters' actions, dialogues, and soundscapes to generate a coherent party scene.

Original post →

More from Multimodal

Multimodal channel →