ArtBot for Stable Diffusion | Hacker News Reader

ArtBot for Stable Diffusion

tinybots.net

70 pointswebmaven19 comments

Hello! ArtBot dev here. Happy to answer any questions. Been hacking away on this in my spare time over the last year. It’s been a lot of fun!

I’ve taken a bit of a break over the last month or so but am slowly picking things up again. Hoping to improve the site much more in 2024.

Just going reiterate that ArtBot is awesome, thanks for developing it.

The AI Horde in general is awesome too. Its such a happy alternative to generation via expensive APIs, especially with nice web front ends like this.

Set the model to "I can't believe it's not photography" for better results.

An alternative free site, stable-diffusion.site isn't bad. It's locked to a reduced model, but it's a reasonably good reduced model.

Hopefully the 5m wait is due to hitting HN front page! Does this thing scale? How do you donate compute? How do you skip the queue?

Given the name chances are they are using consumer GPU volunteered to this clustering system https://stablehorde.net/

Yup! It's since been renamed to the AI Horde, but same backend service.

https://aihorde.net/

I think if you want to skip the queue you use a different service, where that is part of the setup.

Or donate your own GPU time for kudos, which lets you skip the queue!

This is what I do. I sometimes run llm inference on my 3090/2060 for the horde, and use the kudos to impulsively generate images on mobile when I need them.

Yeah, that’s a long time to wait. I recently started paying for a generation service, and even the +/- 10 second wait for any generation there almost feels like too much.

You can build up Kudos with a Horde worker (aka a cloud or local GPU instance) and build up all the generation time you could ever need very quickly.

I love this scheme. It feels like I'm cutting out the middle man, and just paying for the pure inference cost (as well as enjoying the fruits of the pipeline maintainers for free). I also like the idea of "donating" GPU time for those without a inference capable machine.

I believe that simply registering for an API key on the AI Horde website and ensuring you use that jumps you up past a ton of the queue work.

ArtBot visitors generate about 1,100 images an hour and it has only accounted for about 10% of the total traffic the AI Horde is using at any given time (I believe most people are interacting with the AI Horde through Discord chat bots or Telegram).

As a long time participant in thr hoard you have three options of you want to jump ahead: get some kudos by rating images, help the hoard by running a server in the cluster, or donate $ and get some kudos as a thank you.

Source: https://github.com/Haidra-Org/AI-Horde-Worker#ai-horde-worke...

You can also run koboldcpp on a Mac (or pretty much any PC with a small dGPU) to generate kudos running LLMs.

The diffusion/llm kudos is interchangeable.

Its literally a 1 click download + GUI on Windows, and a nearly zero dependency compile on OSX/Linux, and it takes the same Horde key artbot takes that you can register here: https://aihorde.net/register

Certain types of requests (like loras or less common models) can make the queue much longer.

“Brisket on a table” yields brisket on a plate. I hope I never have to ask for “brisket on a table”

I've been playing around with DALL-E 3 a lot recently. One of the things they do is to expand a user's prompt in order to add a lot of detail. Via their API, you can see the expanded prompt (whereas, you can't through the ChatGPT interface).

They obviously have the power of their LLM behind them and can generate some really interesting prompts. There is an open source implementation that the creator of Fooocus made which attempts to expand on prompts using some commonly used keywords[1] with some sort of basic context.

e.g., I typed in "Brisket on a table" and got: "Brisket on a table, product photography, Michelin star, award winning photo, 8k, trending, HD. High quality image, highly detailed, stunning lighting, flawless render, masterpiece, still from the movie directed by Denis Villeneuve with art direction"

You get a much better image with that prompt vs just the basic: "Brisket on a table"

[1] https://huggingface.co/spaces/daspartho/prompt-extend

You can see the expanded prompts on chatgpt too! U just have to tell it to print it out

Hey, why was this flagged?

Reply on news.ycombinator.com