Show HN: SMS Interface for Stable Diffusion | Hacker News Reader

Show HN: SMS Interface for Stable Diffusion

50 pointsarthurcolle36 comments

If you text 8145594701, it will send back an image with the prompt you specified. Currently only US numbers can send/receive texts because Twilio. Sorry to the rest of the planet!

I think this will likely fall over but I figured this would be a cool little thing to demo. I removed the NSFW filter so be mindful of your prompts!

I don't persist numbers and there is no linkages being saved between the original text message and the generated images.

View on HN

Similarly, if anyone wants to try it via a Telegram bot, try mine! https://t.me/piratediffusion

Why does this comment keep getting bumped up to the top?

Maybe people like my bot? Sheesh. Have you tried `/enhance`?

No worries, I wish you success. Just not understanding why this post keeps getting bumped up to the top.

Was it difficult integrating with Telegram?

I quite like the Telegram API, but it took some backflips (websockets style ones) to hook up a single instance of the bot code (Elixir) to multiple rendering GPUs

Took 3 min for me, but it's neat. Great interface idea, mate! As these things get faster with more compute, it will be a neat thing to be able to use in text messages instead of giphy or whatever. Love the idea.

Yeah sorry I had to change the SMS parsing encoding type from ascii to utf-8 after someone sent in a funny symbol so the queue was growing.

Neat idea! What are you running it off of?

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.

Do you have any sort of website or social media I can follow? img2img sounds awesome.

arthurcolle

reddit.com/r/sms2img/comments/x7nnlv/how_to_use_sms2img/

Subreddit created! Feel free to follow. Website/API coming sometime this week :)

> I would like to use an affordable GPU cloud hosting solution

Tristan from Genesis Cloud here. Please feel free to reach out to me (email is on my profile). Happy to provide some resources for this project.

Hello Tristan, thank you kindly! I will send you an email when I'm back in my dungeon. Cheers

I hope you have a cap on your twilio billing

damn, whalesalad.. I was really inspired by your blog and designs when I was a little kid. I remember how satisfying that Tasty WP theme was. Miss those web 2.0 days. Cheers Michael!

Heyyyy thanks! I’m working on a new blog now and going to revive the OG theme from those days. I’ve also got a post in the works discussing that entire process (the tasty theme). Glad to hear I could be an inspiration to someone <3

Awesome! Looking forward to that post

I don't right now. Do you have any recommendations for what should I do?

Turn off Twilio's auto-renewal / auto-refill functionality as a first step.

Cool I already had it set to that. Thanks! Any other suggestions?

Yeah wow 30 bucks all said and done, crazy. Is there a cheaper alternative to twilio

Did you set up your own infrastructure? I’ve been looking for a Heroku-like service to deploy some of these GPU intensive models

You can use https://replicate.ai for heroku like experience. It's very expensive though but they charge compute per second based on requests.

Otherwise, I recommend using runpod, lambda, etc for an order of magnitude of savings.

Crowd source services like vast.ai aren't reliable and barely less expensive than dedicated low cost providers.

I could use an assist/pointers on running this better. Would love to chat since you seem to have some knowledge here. If you're game, feel free to contact me at my HN username at Google's email service. Cheers!

As mentioned in a sibling comment those services are relatively expensive.

If you don't mind ~30 minute one time setup you can get stable diffusion up and running incl. a nice webinterface on basically any cloud provider offering GPUs. I used https://rentry.org/GUItard as a guide and adopted it slightly for my needs.

As my desktop only has a RTX3060 I am renting a GPU instance on Genesis Cloud (billed by the minute and no cost while instance is stopped) using a RTX3090 with 24 GB of vram.

How long does 50 step inference take with the rtx3090?

512x512 with k_lms is around 5s

Nice, that's pretty fast

Super cool. I’ve been using it for the past 20 minutes and have been sending everyone I know a link to this post, but most people don’t seem to realize what it is!

It’s so rad to have a project like this so accessible through text.

I want this to be a thing as long as I can support it but I would definitely like to get some help with scaling it. Right now if I told you how this worked you would laugh for real. Look at my recent comments if you want to find a way to contact me, I'd love some other people to help me make this better and build out img2img support.

Really happy to hear you are enjoying it :) Please forgive me if it falls over, I am operating in a hostile environment haha.

This is a joy to play with! It's the perfect UX for making this tech more accessible.

Everyone in my house has been thinking up silly prompts to see what comes back. Thanks for setting this up.

Thank you for your kind words. I'm so pleased to hear my little bit of work lead to some fun experiences for you and your family! Any fun prompts you particularly enjoyed and would like to share?

I hope it will start working again! Even if not, thank you!!

This is fantastic. Awesome project!

Reply on news.ycombinator.com