# Performance preservation plus video extension with minimax H3. Is it possible yet? — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/34f8c4ac66af8ad2

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-10-23 17:03 ingested 2026-10-24 02:38

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1xdjr6q/performance_preservation_plus_video_extension/)

## retained text



```
Quick background. For me, being able to use human actors’ performances while replacing their environments and maybe even their outfits is the holy Grail of AI filmmaking.

I have managed to figure out how to do this, more or less, which for me ends up looking much better than compositing from green screen, as the lighting, etc. really sells the actor being in the new environment so much better.

However… On my 3090, the longest clip I have been able to create is nine seconds long before OOM. 

I am seeing all these amazing new video continuation workflows for H3, and wondering if anyone could suggest a method for doing reference to video on longer clips and seamlessly joining them together to allow for one long uninterrupted take of an actor’s real performance.

Is this possible? Or would I be better off using H3 for all of the shorter clips and then using something like Seedance when I simply have to have a long uninterrupted take? Any suggestions would be greatly appreciated!
```



document fetched 2026-10-24 10:34 · canonical https://reddit.com/r/StableDiffusion/comments/1xdjr6q/performance_preservation_plus_video_extension
