ReleasesNSFW

My-NSFW-Wan 1.3B fine-tune debuts on HuggingFace under Apache 2.0

A new 1.3-billion-parameter fine-tune of nsfw_wan_1.3b appeared on HuggingFace this week under an Apache 2.0 license, joining the growing roster of uncensored image models.

ByAlex Sokoloff·June 9, 2026

My-NSFW-Wan 1.3B fine-tune debuts on HuggingFace under Apache 2.0

My-NSFW-Wan-1.3B is a 1.3-billion-parameter fine-tune built on nsfw-api's nsfw_wan_1.3b base model, released this week on HuggingFace under an Apache 2.0 license. The weights are freely available for commercial and non-commercial use without restriction. At 1.3 billion parameters, the model fits comfortably on consumer GPUs with 8–12 GB VRAM, making it accessible to practitioners running local inference without enterprise hardware.

The Wan architecture is a newer diffusion branch optimized for efficiency at smaller parameter counts. Fine-tunes like this one typically adjust the base model's attention layers, timestep schedules, or output distributions to shift aesthetic style or improve prompt adherence. The Apache 2.0 license permits modification, redistribution, and commercial deployment—the same terms that have made models like Stable Diffusion and FLUX popular in the open-weight community. As of June 7, the checkpoint shows zero downloads and zero likes, typical for a fresh upload that hasn't yet circulated through community channels.

More in Releases

ReleasesNSFW

Huihui abliterates Gemma 4 12B, strips safety filters from Google's multimodal model

An uncensored version of Google's Gemma 4 12B instruction-tuned model is now available, with built-in safety filters and restrictions removed.

Jun 9, 2026

Community

Developer releases Claude auto-research loop to enforce consistent task execution

AlexWortega open-sourced claude-autoresearch-skill, a Python wrapper for Claude Opus 4.8 that automates multi-step research workflows and enforces output quality across repeated turns.

Jun 9, 2026

ResearchPlatform

Mind Lab scales LoRA to millions of users on trillion-parameter models

Researchers propose infrastructure to cache millions of personal adapters on a single trillion-parameter base model, stabilize ultra-compact LoRA training under reinforcement learning, and demonstrate emergent collective intelligence across adapter populations.

Jun 8, 2026

Research

Elmes* automates LLM teaching evaluation across 330 educational scenarios

Researchers released Elmes*, an automated system that constructs fine-grained rubrics to evaluate how large language models teach across 330 educational scenarios spanning 11 subjects and three grade levels.

Jun 8, 2026

Research

Causal graphs reveal how LLMs organize concepts during inference

Researchers map LLM reasoning with causal graphs and counterfactual chains, revealing class-discriminative concept dependencies across diagnosis, sentiment, and judge tasks.

Jun 8, 2026