I would share more details but there is a chance that I am mis-using the ToS of a large tech company by running it in this way so going to hold off until I've built + deployed it on a GPU cloud host. If anyone has AWS/GCP gpu host credits that you want to gift me I'd def love that - would even have a credits page or something when I put a site up.

I would like to use an affordable GPU cloud hosting solution but I had some difficulties with Vast.ai even though they are pretty affordable. It seems like because the instance they give you is just a running Docker container, I can't easily run Docker inside those instances, so I will probably just have to rewrite this in a more scalable way and have it deploy a container that I have designed to operate in an equally simple way (handle webhook request from Twilio/parse message/run message prompt through txt2img and then upload to s3 and send back the S3 link to the original sender).

I plan to support img2img via photos taken on a phone camera and ingested as an MMS message after I have rewritten this to make it actually productionized and scalable. This was a 1.5 hr proof of concept after I tried to write a better version yesterday actually running on a GPU host machine but was facing the above infrastructure problems.

Basically just FastAPI for the API/hooks, Twilio for messaging and [REDACTED] for running the model.