← sources

How do you guys seriously run local video models?

source: reddit (reddit:StableDiffusion)published 2026-09-23 12:31ingested 2026-09-24 02:38

retained text

I have a 7900XTX, 24gb vram. I cant seriously run any video model.

I tried both Minimax H3 and LTX 2.5, LTX was supposed to be easier to run, and it still takes me 2 min to generate 2 seconds of video at 480p, I tried to generate 10 seconds and the estimated time was around 20 min. This was on a rocm sd server.

I also tried comfy ui. I got the default workflow from Comfy Desktop, downloaded everything, ran the default prompt + image. Over 10 minutes the job still said 8% done, for a 5 second 720p generation.

Is this a NVIDIA vs AMD thing? I have seen people on reddit post that their 5070ti could get LTX to generate videos in under a minute, for longer lengths.

Any help is appreciated.

document fetched 2026-09-24 00:35 · canonical https://reddit.com/r/StableDiffusion/comments/1wo4pgf/how_do_you_guys_seriously_run_local_video_models