Z-Image Turbo LoRA strips safety filters for unrestricted generation

rachel-luxangel-ai released zit-nsfw-2, a Diffusers-compatible LoRA adapter that fine-tunes Tongyi Mai's Z-Image Turbo text-to-image model for unrestricted generation. The adapter is tagged not-for-all-audiences and hosted on HuggingFace as of May 26, 2026. Z-Image Turbo is an open-weight base model optimized for fast inference; the LoRA retargets it by stripping safety classifiers and retraining on unrestricted datasets—a workflow enabled by the model's local-runnable nature. The model card lists zero downloads and zero likes at upload, with no sample outputs or training dataset details provided.

Z-Image Turbo remains a minor player in the open-weight text-to-image space, overshadowed by FLUX, SDXL, and Illustrious variants that dominate community workflows. The zit-nsfw-2 naming implies at least one prior version, though no zit-nsfw-1 repository exists under the same creator. Civitai and HuggingFace host thousands of uncensored LoRAs, many reaching tens of thousands of downloads within days of release.