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.
Sorry to hear it works badly for others. I just used the ai-story-generator and had the best performance I have ever gotten! By a mile! I would like to know what LLM it is so I can find it and download it for my own use. Given the comments here about the model identifying itself differently, I’m not sure what to believe!
Thanks for all the hard work!
Did you also change the model? I asked the AI which model it was, and it said Claude 3, yesterday DeepSeek-R1, sometimes GPT-4. I’m kinda confused. Anyway, thanks for the hard work!
LOL It just told me that it is Mistral-Chat. Seems like it has a personality disorder!
I mean, that’s a huge context window
I’m just happy to hear you’re still working on it and giving it attention. What is new with the text gen exactly? I have yet to notice any difference, but maybe I’m not looking in the right places…
Thank you so much for the update and for all your hard work! I’m grateful to know that you are reading user feedback.
By and large, I’ve been really happy with the new text gen. When it works well, it works really, really well. There have been a few times that it was working amazingly and I was really impressed by the quality – times where it took a really complex scenario and performed beautifully, picked up on foreshadowing, did a great job at portraying nuanced characters and writing realistic in-character dialogue, etc. with minimal corrections needed.
I also already find it a huge improvement from the old model in almost every single way, outside of times the model starts severely and repeatedly hallucinating. More than anything else, I’m so, so glad that it isn’t afraid of dark themes like the old model was. I love that if I give a character a dark backstory, it will sometimes add to and embellish it, and will also bring those details up when relevant of its own accord (without me having to repeatedly instruct and nag it to do so). So I’ve been really happy with the update!
Yeah, this has definitely been my experience, too. As a story generator it works really well for me.
Thanks for making a wonderful facility for us. FYI, T2I is really awesome on Perchance. Keep up the great work!
Thanks for the update and for your work, Dev! Also, can you enlighten us about the context text window? Should we expect any developments, or is it a long shot?
Lo, the Savior hath spoken… Thank you savior. Can’t for the improved text to image gen.
glad to know! need more Dev posts been a while
Salute o7
Thanks dev! There are actually two questions many of us have,
- The already mentioned context window (this was the one thing we hoped most for) and
- Which model did you choose (so we can look for workarounds easier)?
Thanks for all your hard work! I really noticed the text gen getting better little by little. Besides the hyperfixations and repetitions you mentioned there is also the on-off problem of following the prompt and writing instructions, other than this I’ve noticed only minor quirks (ignorable for me as the other problems are more important).
Hi Dev!! Personally I think the new text model is a HUGE improvement over the old one. The story maker has gotten worlds better! If you think it can be improved that’s AWESOME! I look forward to that. Don’t beat yourself up or take anyone’s negativity too seriously, you’re making a really cool resource for free and that’s really cool.
What is multiplayer plugin and database plugin actually?
This is just a guess but multiplayer would allow multiple users in the AI Character chat or other generators, the database plug in would allow the generator database to be stored elsewhere, rather than your browser local storage.
Nice
i choose elf sorcerer. you choose orc ninja. we enter room and there r 10 badguys. i shoot lazerbeam. what do you do?
hence multiplayer. It is the thing currently lacking with ai games. personally, i tried making games and asked ai to build some cool stuff from management games to rpgs, strategy, and racing games. but eventually i stopped because there’s only so much fun they can be since they all were singleplayer. They become just simulations one person can do until bored.
Multiplayer allows competition, socialness, and infinite playability. I think it will be difficult to incorporate multiplayer plugins in a nonhackable way, since sufficiently knowledgeable ppl can often just open up teh console and run custom scripts. Thats the problem i had when i tried to make a leaderboard plugin functional on perchance. But, if Perchance has a good solution, Multiplayer functionality essentially ‘opens up’ the next era of Perchance Games; especially if the ai understands how to incorporate the multiplayer plugin as it currently does with ai image plugin. Tbh tho im not sure unhackability is possible since the solution i know of is to have ‘the brains’ of the game be on a server that players can query and interact with but cant directly manipulate. So we may or may not have to settle for hackable games. i do not know.
I have not checked teh ai coder ai recently, but if Perchance makes a legit multiplayer plugin, ai code assist understands it, and ai code assist is top tier updated, imo Perchance will be one of the coolest creative spaces on teh internet for people, including those with no code experience at all, to make browser games.
Even just a game where each player can have a car, choose a crew, and then the page simulates a race, would instantly become one of the most fun pages on perchance. Especially with database storage so progress is keepable instead of wiped. Idk exactly how it would work yet because each page would need the option of basically being it’s own little server. It’s definitely possible tho, since each page is part of a bigger server ecosystem already.
anyway, those just some thoughts. im curious to see where it goes as much as teh rest of u
Arigatou sensei
Multiple players - realtime database, like a game server, can counts score, render map, etc
Database - just like gallery
They are just like comment plugin or gallery plugin but more versatile, you can use them to vote, play game, create forum, IDK all the above is my guess.
Thank you ive hope AI chat bots become normal because current chat bot are…werid. Speaking of image gen, can you improve list of charachters can be generated? Like Alex, Clover and Sam from Totally Spies, or Will, Irma, Taranee, Cornelia and Hay Lin from W.I.T.C.H? Because auto random prompt aware of these charachters but cannot generate thier basic likeness like Lola Bunny or Ariel from Little Mermaid. Im suprised in old models gen it possile to generate basic likeness of Sophitia from Soulcalibur games, but on new model gen it require write better prompt.
So anyway good luck,
Do take your time with it, but i personally preferred the old ai somehow…? It just seemed…much better at following instructions fully without trying to hallucinate things? But the more intelligent part is definitely better, still would’ve been cool if there was an option to use the old ai (〒﹏〒)
Okay this is starting to be miserable, when my bots that worked perfectly months ago now can’t write more than 2 sentences or follow instructions it’s…not worth the smarter ai imo, especially if even if the old ai was “more dumb” it could actually write a lot more and also follow instructions (༎ຶ ෴ ༎ຶ) (not really saying it’s bad or anything just needed to vent\rant)
We perchance users are actually too much grateful to you. at first lot of thanks. But still I have 2 questions to ask you man.
- why you are not in reddit
- why are not you talking about context window of the text plug in. seems like even after the update still its 6k.
reddit can go to hell. what?
That’s fantastic news. There’s moments perch really shines and you can see some fantastic quality and to have that all the time would be heaven.
