Why isn't RuinedFooocus uncensored?

[ghost]

I was using RuinedFooocus but I noticed that some parts of it are highly censored. Example when I asked story teller to create for me a specific prompt it denied. Also when I created a prompt myself and used it a picture of a troll appeared. I think all this is due to the LLMs used. Why not use an Uncensored Model? There are many. For example Dolphin? Is there a way that I can uncensore RuinedFooocus on my PC? I really like it, I think that it's the best Fooocus fork. Thank you very much for your help.

[yownas]

There are no specific reason why we choose a "censored" model. We just started experimenting a little with the original llama-model.

You can choose other models. Documentation is lagging behind quite a bit... it is more fun to make things than to write documentation. :) You can change model by adding these settings to RuinedFooocus\settings\settings.json.

  "llama_repo": "DavidAU/L3.2-Rogue-Creative-Instruct-Uncensored-Abliterated-7B-GGUF",
  "llama_file": "*Q4_k_m.gguf",

(This will probably be improved later so you can download models and point directly to the file.)

When it comes to making your own chatbots, the Troll is the default avatar if it can't find one. But it isn't difficult to add one. Just make a folder for your assistant, create a file called info.json that should look something like this:

{
  "name": "Cat",
  "greeting": "Meow?",
  "system": "You are a cat. You can only reply with meows. You want treats, the food bowl is half empty and the water bowl has probably not been changed in more than an hour. User! Help!"
}

And then put a avatar.png to the folder. (You can also put a "avatar": "some-file.jpg" in the json file, but that feature was mostly used during testing.)

...or just make a copy of any of the folders in RuinedFooocus\chatbots and give it a new name and start changing things.

[ghost]

Thank you very much. I will try your suggestion.

[yownas]

In the latest update we made it a bit easier to manage the llm models. Just download them to models/llm/ and add it to settings/settings.json.

(The old format with llama_repo and llama_file will still work.)

  "llama_localfile": "DeepSeek-R1-Distill-Llama-8B-Q4_K_M.gguf",

RF knows about some repos, so you can actually just add the above to settings.json and it will be automatically downloaded.