Timeline
Post
Remote status
Context
8
@graf @pixel8edanime whew, that's pretty good
@coolboymew @pixel8edanime I'm working on a new gen of AI. Here's a young Zelda it generated maybe 20 mins ago with volumetric lighting . Will be easy to adapt to 16/32-bit pixel
🇵🇸
@graf @coolboymew @pixel8edanime Pretty nice, it does backgrounds well. Novelai makes every background an acid trip, or I just suck at prompting them.
@Shadowbroker2135 @coolboymew @pixel8edanime >it does backgrounds well
here's pippa floating a dinghy in a flood
here's pippa floating a dinghy in a flood
@graf @pixel8edanime @Shadowbroker2135 Wow
Yeah NovelAi sucks at backgrounds. It excels in the characters themselves, but the background, oof. The minute you want a scene that is more than exceedingly simple, it just can't
Yeah NovelAi sucks at backgrounds. It excels in the characters themselves, but the background, oof. The minute you want a scene that is more than exceedingly simple, it just can't
@graf @Shadowbroker2135 @pixel8edanime Maybe around gen 5. However they're busy on the (about goddamn time) text gen
@coolboymew @pixel8edanime @graf I need to try out the text gen, maybe it'll help me get back into writing.
Replies
1
@Shadowbroker2135 @pixel8edanime @graf YMMV
The first few textgen models they had were made for creative writing help and they're absolutely spectacular for that. If you actually bother with lorebooks and entering stuff properly and giving proper responses, the language and variety it shat out was pretty impressive, but you had to give an equal amount of effort, and it wasn't made for the long term (too low context token), basically, you had to tard wrangle it a lot
The new model is much smarter and has a massive context size, however it tends to fall a lot into repetitive AI slop. However, if you use their own interface instead of Sillytavern, you can now add scripts to do stuff for you. One of which can cut down on the slop. Much better for long term roleplaying IMO, still doesn't surpass the old model for creative writing, but much less tard wrangling and having to fill lorebooks and such, which is why it's better for roleplaying. However, it's still a "beta" and still not fully tuned I believe?
The first few textgen models they had were made for creative writing help and they're absolutely spectacular for that. If you actually bother with lorebooks and entering stuff properly and giving proper responses, the language and variety it shat out was pretty impressive, but you had to give an equal amount of effort, and it wasn't made for the long term (too low context token), basically, you had to tard wrangle it a lot
The new model is much smarter and has a massive context size, however it tends to fall a lot into repetitive AI slop. However, if you use their own interface instead of Sillytavern, you can now add scripts to do stuff for you. One of which can cut down on the slop. Much better for long term roleplaying IMO, still doesn't surpass the old model for creative writing, but much less tard wrangling and having to fill lorebooks and such, which is why it's better for roleplaying. However, it's still a "beta" and still not fully tuned I believe?