CMMMD releases bilingual uncensored diffusion model NSFW-gen-v2

An uncensored text-to-image model supporting bilingual prompts appeared on HuggingFace this week, targeting practitioners who need unrestricted generation in English and Portuguese.

NSFW-gen-v2 is a diffusion model from CMMMD that ships as Safetensors weights under the not-for-all-audiences tag. The model card lists 3d and unfilteredai among its labels, signaling the checkpoint is tuned for stylized rendering without content filtering. The diffusers pipeline tag indicates compatibility with the Diffusers library, making it straightforward to load locally via Python and integrate into existing text-to-image workflows.

The bilingual capability sets NSFW-gen-v2 apart in a field dominated by English-only models. Portuguese language support remains rare among open-weight image generators, despite Brazil's large AI practitioner community. CMMMD has not disclosed whether the model was trained from scratch on multilingual caption data or fine-tuned from an existing English checkpoint with Portuguese prompt pairs. The card does not specify parameter count, base architecture, or training resolution.

The 3d tag hints at a stylistic focus—whether 3D-rendered aesthetics, volumetric lighting styles, or compatibility with 3D asset pipelines remains unclear without sample outputs. The unfilteredai label explicitly signals the absence of safety classifiers or prompt filtering, consistent with the not-for-all-audiences designation.

CMMMD has not published benchmark scores, sample grids, or training details. The Safetensors format ensures the weights can be inspected for embedded code before loading, a standard safety practice for open-weight models from new sources. Practitioners interested in bilingual uncensored generation can pull the weights directly from the HuggingFace repository. The model's compatibility with the Diffusers pipeline means it should work with standard ComfyUI and Automatic1111 loaders that support HuggingFace checkpoints, though users will need to verify prompt syntax and any custom conditioning tokens CMMMD may have introduced.