6G VRAM can actually run two-person dialogue videos? Just tried this LTX LoRA.

The Forge / 工坊 Workflows / ComfyUI / 工作流
1

My laptop only has 6GB VRAM, so I never thought I’d be able to mess with video generation. Then I saw someone using a new LoRA on LTX 2.3 to make two characters talk at the same time—just one image, a prompt, and a custom audio file, and it generates. So I gave it a shot.

This LoRA is specifically trained to improve consistency and lip-sync quality in dual-character scenes, which are exactly the hardest parts in the standard workflow. The tutorial covered a few points: how to generate the starting image first, use Prompt Relay to boost accuracy, make the prompt more obedient, and that the LoRA itself works at 1240×720 but can be upscaled to full HD output.

The tips for dual-character lip-sync were pretty solid. Being able to run on 6GB is really student-friendly, though don’t expect blazing speed.

2

6GB VRAM can even run video now, thank god.

3

Yeah, the standard workflow for two-person lip-sync is definitely tricky.

4

“Prompt relay” is basically passing your prompt through multiple models or tools in sequence to refine or transform the image step by step. Like, you start with a rough idea in one model, then feed the output into another to add details or change the style. It’s kind of like a chain reaction for generating images. Some people use it to get more complex results without doing everything manually. Hope that clears it up!

5

Slow but at least it runs, that’s a win.

6

Nice, even laptop GPUs can run this.

7

Will it look blurry if I max out the resolution to full HD?

8

Saved the post.