Show HN: Generate Images with Stable Diffusion | Hacker News Reader

Show HN: Generate Images with Stable Diffusion

diffusion.labml.ai

5 pointsvpj6 comments

Thanks for posting!

Curious — What’s the point of this? Specifically it’s lacking controls for parameters, user prompts do not appear to show in feed, unable to filter feed, etc.

Did you make the interface? What did you learn so far? What do you hope to accomplish? How much is this costing? How many prompts hour are you running? Where are prompts coming from? Anything specific feedback you’re looking for? Any changes you’re planning to make? Have you shared this elsewhere publicly and what are URLs?

Also checked your profile, which links to a blog, but nothing on it either.

blog.varunajayasiri.com

_______________________________

Edit: What if any affiliation do you have with LabML?

- https://labml.ai/ - https://github.com/labmlai/ - https://twitter.com/labmlai - https://papers.labml.ai/papers/weekly

Hi, I made this at labml.ai, based on our previous playground [1]. It is for people to play around with the Stable Diffusion model [2]. The feed shows the inferences in random order. All the items in the feed are images generated by users.

We will add features based on feedback. We are working on adding feed filters and inference parameters.

We have already shared this on our Twitter account [3]. Planning to do a write-up about this and will be posting it on our Twitter account.

[1] https://neox.labml.ai/playground [2] https://github.com/CompVis/stable-diffusion [3] https://twitter.com/labmlai

Thanks for replying!

Guess as is, while I can see this being of use to people, in part because it’s appears to not require login and be fast, there number of services already online that currently offer vastly better experiences in my opinion. If you’re unable to create a unique value offering that’s defensible, unclear to me what the goal is.

(Please don’t take me being direct as negative, great you’re sharing and if I didn’t know of other systems this would be a super useful resource as is. If you have any questions or feedback for me, let me know.)

- Curious, how’re you sourcing the papers in your feed, have you posted about this process?

This was fun while it lasted! Lightly loaded, it worked great, and provided an introduction to the technology for which I'm grateful.

Sadly, now if you put in a prompt, you are told that the queue has 20-something items in it. OK, that should surely clear in the space of ten minutes. Couple hours later, no update, no images; the ETA countdown has hit zero and reset umpteen times.

I obtained numerous images and kept over 70 fascinating, aptly generated ones.

Here is something funny/interesting. It looks like some of the training data has taught the AI to generate gibberish stock photo watermarks.

https://imgur.com/qgW6il5

I can hardly imagine those artifacts to be anything else.

I think as the system is getting more crowded, people are giving up waiting.

Is the system removing the queued jobs when users close the page?

Reply on news.ycombinator.com