49
[dev] New text model is still quite bad sorry - working on it (lemmy.world)
submitted 1 year ago by to c/perchance@lemmy.world
94 comments fedilink hide all child comments
 

I didn't expect it to take this long for me to get the new text model under control. It's much more intelligent in general, but the repetition and random fixations can get surprisingly bad.

Sorry about that. This is just a heads up that I'm still working on it, and it'll likely remain a bumpy ride for the next couple of months.

Thanks to those who have complained here and on reddit, especially the posts that give concrete examples of issues, and attempted workarounds, etc.

Please do keep posting any issues/discoveries, ideally with share links to give me specific examples that I can test.

Other news:

  • I'm still working on the next image gen update, and am hoping it will improve fine details, text rendering, and overall image quality. It should also be about 2x faster. If all goes well (which, from experience, it won't) I should be able to get that done in the next month or so.

  • For a long time I've been very close to shipping a multiplayer plugin and database plugin. I've actually made multiple 'ready to ship' draft plugins of both, but have been disatisfied with the design and ended up not publishing each time. Using these plugins won't trigger ads on the generator, so I need to be really careful about the overall architecture to ensure the the servers for them are dirt-cheap to run. I also want to make them fun/simple to use of course. Once I get text and images stable, I'm going to focus on these two plugins and finally get them done.

you are viewing a single comment's thread
view the rest of the comments
[–] 2 points 9 months ago*

After the recent update, I’ve been trying to understand what changed.

body: 
  uniform jet-black coloration across entire body
  build: androgynous slender body
  fur: black fur
  fur texture: sweaty
  skin: black skin beneath fur, barely visible

I carefully built them with detailed layers like this.

However, after the recent update, even prompts under 300 tokens are no longer being processed at all. All of the seed values and prompts I had saved in the past have completely stopped functioning as expected.

Is this because something was temporarily changed due to the recent DDoS attack? Or was there a specification change introduced by an update?

As a precaution, when I use a very simple, flat prompt without hierarchical structure, I can still get some kind of result. However, the illustration style appears to have changed significantly.

Any clarification would be appreciated.

  • permalink
  • fedilink
  • source
    •