Fact #1:
The Perchance text models, or at least AI Character Chat, uses the DeepSeek model api. A reminder that DeepSeek is free and open source!
Fact #2:
Its been said that the new AI art model Perchance AI art image generators uses Flux. My preferred AI art image generator strongly implies this is true. Interestingly, the AI generated images still mentions Perchance uses Stable Diffusion despite being claimed the AI model is currently Flux. I tried looking into the text-to-image plugin page for any mention of Flux, couldn't find any mention.
Edit #1: New Image Model Notes seems to officially confirm the new AI art model is indeed Flux, specifically FLUX.1-schnell (well, based off of FLUX.1-schnell). I don't know when that page was created, though.
Edit #2: Page was last modified on Wednesday, May 22, 2024. So, its unknown if its still FLUX.1-schnell. More likely another newer and more improved version of Flux.
Edit #3: This post from the forum,
spoiler
(TRIGGER WARNING: NSFW)
But I do know this: I have used different Stable Diffusion models and the current Perchance AI art model enough to know the current AI art model isn't Stable Diffusion. The current AI art model produces images much differently. The quality and rendering looks far better than some vanilla and custom/mixed Stable Diffusion models.
~~I won't say how I even found out this information but just wanted so share some facts I discovered LOL~~
Jk, it was through Google DevTools LOL
Cloudflare Worker is a service where Cloudflare provides llm models api. How do I know they are using it? Some time ago, there was an issue on the Cloudflare Worker server, and perchance's AI stopped working. When asked about Perchance, they said they use cloudflare worker and don't host llm model themself. And Cloudflare was having an issue with their server, and that will be fixed when Cloudflare fixes it. They (perchance) only host the image model themself. And for LLMs, they pay Cloudflare to get the service. And yes, but again, if some model is trained on Google's dataset, it won't be DeepSeek. And I never said it is Gemma; I said it is likely Gemma because it says things like knowledge cutoff and other stuff. And let's be real here: there is no other way to know which model it is until perchance themself says what model it is or we have access to the backend servers. So until then, we can guess using those unreliable methods.