I built a site to create free AI videos using LTX 2.3 running on my own GPUs
Lately I’ve been working on my project loremotion.com **.**The goal was simply to let anyone create AI videos without credits, subscriptions, or limits. To actually make that possible, I had to skip the APIs and build my own infrastructure.
I’m mostly using open-source models like LTX 2.3 and Wan 2.1. I’ve personally found LTX 2.3 (specifically the 1.1 distilled version) to give the best results for the speed I’m aiming for. Right now, I’ve capped it at 720p/10-second clips for both Text-to-Video and Image-to-Video.
The Hardware Setup: I’m running this on my own cluster. I’ve got four of my own GPUs (30 and 40 series) and I rent the rest on-the-spot (A100s and RTX Pros). It actually keeps my costs incredibly low—around $8 a day—which is why I might be able to keep the generations free. all wired to Wan2GP
Performance: Depending on which GPU grabs your task, a 720p 10-second render usually takes between 50 and 110 seconds(if there's any way i can get much lower generation time, please do let me know)
Features:
- Dashboard: Your clips stay there for 48 hours before they’re cleared.
- Discover: You can choose to push your best renders to a public gallery.
- Email Alerts: If the queue gets backed up, you can drop your email and I’ll ping you when it's done.
The Catch: To keep the lights on and break even, I had to put ads on the site. I know they’re annoying, but it’s the only way I can offer unlimited generations without a paywall.
Next on the list is getting Video-to-Video working, so if you have ideas on how to improve the generation speed, better models to check out, or features you actually want, please let me know.
Check it out here:loremotion.com