Bl4ckSpaces releases three int4 NSFW diffusion checkpoints

Bl4ckSpaces released three int4-quantized diffusion models on HuggingFace on May 18, each carrying Apache-2.0 licenses and explicit NSFW tagging. The trio — z-image-turbo-nsfw-int4, z-image-turbo-nsfw-v2-int4, and flux-schnell-nsfw-int4 — all use the diffusers pipeline and ship as safetensors checkpoints. The first checkpoint drew 12 downloads within hours of posting; the other two remain at zero.

The z-image-turbo variants cite three recent arXiv preprints (2511.22699, 2511.22677, 2511.13649) in their model-card metadata, suggesting they build on turbo-sampling research. The flux-schnell-nsfw-int4 checkpoint derives from FLUX.1-schnell and carries the "endpoints_compatible" tag. Int4 quantization typically cuts memory footprint to roughly a quarter of fp16, making these checkpoints viable on consumer GPUs with 8–12 GB VRAM.

The Apache-2.0 license permits commercial use, modification, and redistribution without royalty. The explicit NSFW labeling signals the creator removed or bypassed any safety filters present in upstream models. Practitioners running local ComfyUI or Automatic1111 setups can drop these weights directly into their pipelines without additional fine-tuning. Over the coming weeks, download patterns and community feedback should clarify which checkpoint delivers the best trade-off between speed, fidelity, and VRAM overhead—and whether the v2 iteration of z-image-turbo-nsfw meaningfully improves output quality or prompt adherence over the initial release.