Hunyuan or LTX on Serverless Endpoints?
Hi there! I'm working on building a web app where the user uploads an image for the first frame of a video to a react front end, and types a text prompt of what they want to happen in the video, and then it'll send an api request to an h100 serverless endpoint, it'll spin up a gpu with ltx to start with for the mvp and eventually I'd like to mess with hunyuan as well (probably in batches or charge more in app credits for immediate processing), and it'll save that video to a storage volume on runpod and then it'll serve the generated video back to the front end when it's ready.
I've already deposited some money and created a storage volume because I'm assuming I'll want to load the model weights off a persistent storage volume, but I'm not sure how to get started. I see there are lots of templates for pods but I don't need comfyui or anything, more likely something like fastapi but idk. Can anyone help me out with a list of what I need to learn in order to build this? Perhaps some links to recommended resources and guides that you've found helpful? I have the react front end on firebase is all so far. Thanks!
7 Replies
Also, from what I've read, it takes quite a bit of time to set up on the storage volume the first time and then after that each time will only take a few minutes to maybe 5 minutes to load the models into the vram and spin up and everything? Is that correct?
Unknown User•4mo ago
Message Not Public
Sign In & Join Server To View
How long wpuld ltx take to spin up from cold start on a h100 on average you think? If downloaded each time?
Unknown User•4mo ago
Message Not Public
Sign In & Join Server To View
I don't know that, I'm new to all of this and have no idea how it works which is why I asked for a list of things to learn and recommended resources for places to learn from
Unknown User•4mo ago
Message Not Public
Sign In & Join Server To View
OK tyvm