I don't mind the slowness but the actual AI Chat model is such a giant downgrade
This model NON STOP fights what you tell it to do. Before you could start a sentence and it would finish the chat for you. Now you have to finish 90% of the chat yourself or it fights you aggressively.
Example "Robert cries"-continue
"Robert cries, but not tears, he's actually fine and the family laughs together.
I really understand perchance is free and it's an expense and appreciate it. I don't mind having to wait a few extra seconds for the chats to work, but this model is truly worse than any before and it's got me considering other websites. Can we please get a fix for this?
5 replies
I agree, I've tried but I have yet to find a way to make it work well. Before the change happened to Text, I was really enjoying the results for the most part.
I got so tired of it, I switched away. I installed SillyTavern, and set up an account with OpenRouter. In three weeks of pretty heavy use, I have spent less than $5.00 on chat AI. DeepSeek V4 Flash is very cheap, has a 1M token context allowance, seems to have zero NSFW filter, and goes about 3x faster than the perchance AI character chat. If you're ready to graduate, and can afford to spend a few bucks, and have a machine to run it on, you might give it a try. There are free models on OpenRouter too, though I have not tried them for quality.
If you don't know how to do it (and it can be intimidating) just ask ChatGPT or Claude for help setting it up and running it. I am about 600 responses into a roleplay right now which has been very good, with only a few "dumb AI" moments. And I'm pretty sure if I wanted to spend more, I could use a better model and get even better results.
I have used ST in two ways... at first, I was doing exactly what perchance does, one character, one user, chat. Then, I wanted a more RPG type experience, so I have developed a prompt that makes the AI be a game master, and that makes having third parties and world events more doable. Then, the main character(s) that the GM will play end up as lorebooks, and the 'setting' goes in the GM's scenario box.
It still seems to need an occasional kick to keep from "going with the flow" too much, but as I said, 600 responses in, and I'm still pretty happy with the story. It's a reverse Isekai slow-burn gay romance tale where an elf appears in Central Park, and my character befriends and protects him.
Hell nah bro I'm not paying to talk to AI people
I agree, I've never paid to chat either other than Gemini pro for coding. I think given some time the Dev will improve the performance of the Text gen. As he looks at feedback he slowly irons out a lot of issues.
plot twist