Instructions to use DeepNeuralNerd/Gemma-4-12B-it-uncensored-heretic-DeepNeuralNerd-LTX_2.5_ComfyUI with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use DeepNeuralNerd/Gemma-4-12B-it-uncensored-heretic-DeepNeuralNerd-LTX_2.5_ComfyUI with Transformers:
# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("DeepNeuralNerd/Gemma-4-12B-it-uncensored-heretic-DeepNeuralNerd-LTX_2.5_ComfyUI", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Heretic snake oil
Author of Heretic - using it as text encoder - it does not work:
https://www.reddit.com/r/StableDiffusion/comments/1vmdxzk/psa_im_the_creator_of_heretic_and_i_advise_you_to/
Such a stupid comment. They do work in the case of LTX where Gemma is heavily censored. You have to use LoRA's in conjunction with the text encoder. It comes down to how well the encoder was trained against failure rates. Some heretic's can be snake oil and others are quite good. I had a 95% success rate with heretic encoders in 2.3. He just doesn't want people to use it in videos where characters are 'shucking their corn'.
It's true that H3 is mostly uncensored, but it can censor certain requests.
Everything worked perfectly for me, and the effect was excellent. This model is designed for LTX 2.5 and is not suitable for other uses. Thanks to the author for sharing.