Show HN: Pictero – a playground for text-2-image generation on HF
pictero.com
It's a playground for all those text-2-image models that Huggingface is offering. I am playing around with them for a while now and sometimes, for other sideprojects, I am using them to create small illustrations. In the beginning I find it quite confusing to read through all the docs and connect the dots, because the docs sometimes does not really follow a straight path, and most demos does not always explain all those parameters. That's why I was building this interface: To make the setup easier and hopefully a little easier to understand.
I added a couple of well known HF models as a starter, but it's quite easy to add more because the interface is "generic".
It also supports adapters (textual inversions or multiple LoRAs) to "infuse" the process with fine-tuned models. And of course there are schedulers/solvers to select from.
Well, there are lot of other parameters and last but not least it has a sharing feature to keep track of your "settings" (via a GET-parameter in the URL) as well as a code area, that shows you the code for your current setup so you can easily paste and run it anywhere else.
A little tech background: It's running on free space right now, so it's probably quite slow. I recommend to clone it locally. I am using the Gradio library which is a perfect companion if you want to create prototypes at HF. I had to implement a little workaround to keep settings persistens in the local cookie storage. Gradio does not really supports that.
(don't check the history of the domain, it's my "garbage domain" that I am using for all kind of private projects - well, I bet now you gonna check the history, I warned you)