Meta is back to open weights with Muse Glimmer, a 30B model under Apache 2.0 that runs on one consumer GPU. It was built for tool use, not conversation, and it ships safety-tuned. Whether it ever matters for adult roleplay depends entirely on the fine-tuning community, so here is what they have to work with.
Short answer: Muse Glimmer is significant for what it signals, Meta publishing weights again under a permissive licence, and not yet for what it does. It is a 30B agent model: function calling, local coding, long tool sessions. It ships with Meta’s safety tuning, has no roleplay training, and its conversational register is functional rather than warm. For adult roleplay it is a candidate base, nothing more, until tuners take it on. This page tracks whether they do.
What shipped
Muse Glimmer | |
|---|---|
Released | 10 August 2026, Meta Superintelligence Labs |
Size | 30B: 2B perception encoder plus 28B text decoder, distilled from the larger internal Muse |
Licence | Apache 2.0 |
Input | Interleaved text and images; text output |
Context | 131,072 tokens or more |
Languages | 100+ |
Hardware | 55 GB at full precision; under 20 GB at 4-bit, one consumer GPU or a Mac |
Built for | Always-on local agents, function calling, coding, LLM-as-judge |
Knowledge cutoff | January 2026 |
Weights are on Hugging Face with llama.cpp, MLX and ExecuTorch integrations landing after launch.
Why it is interesting anyway
Two things make it worth a page on a site about adult AI models:
Licence and size. Apache 2.0 at 30B is exactly the shape the roleplay tuning community likes: big enough to be good, small enough to serve and to fine-tune on rented hardware. The Gemma 4 26B tunes and the Qwen 3.5 27B writer tunes live in the same bracket.
Meta’s conversational history. Llama-family models were the roleplay base of choice for two years because of their natural dialogue. If the Muse lineage inherits any of that, tunes could be strong.
What is against it
Agent tuning. The instruction layer is optimised for tool calls and structured output. Persona chat needs the opposite: emotional register, pacing, staying in character. Tuners would be retraining against the grain.
Safety tuning. Like Gemma and Qwen, it refuses adult content out of the box; an abliteration or fine-tune is required before any evaluation is meaningful.
No track record. Gemma 4 and Qwen 3.5 have months of community tuning behind them. Muse Glimmer has days.
What we did
We ran the base model through the SFW portion of our persona evaluation with a heavy system prompt. Coherence and instruction following were excellent; the register was assistant-like and it dropped character faster than either Gemma or Qwen base models. That is expected for an agent model and says little about what a tune would achieve.
What would change our view
An abliterated or de-refused build (likely within days of this writing).
A roleplay fine-tune from one of the established tuners.
Evaluation results on our 40-scenario set that beat the current Gemma 4 tune on staying in character, or the Qwen 3.5 tune on long-scene consistency.
Until then, nothing changes in the MyBabes chat stack. We will update this page when a tune exists.
Related in the Lab
Key takeaways
Muse Glimmer is Meta’s first Apache 2.0 open model in a while: 30B, multimodal, one-GPU.
It is built for agents and ships safety-tuned; it is not a roleplay model today.
Its licence and size make it a plausible base for future adult roleplay tunes.
Gemma 4 and Qwen 3.5 tunes remain the working choices.
FAQ
Is Muse Glimmer uncensored?
No; it is safety-tuned. Community de-refused builds are likely but were not available at time of writing.
Can I run Muse Glimmer locally?
Yes; the 4-bit build fits under 20 GB on one consumer GPU or a Mac.
Is Muse Glimmer good for NSFW roleplay?
Not as released. It may become a base for fine-tunes; Gemma 4 and Qwen 3.5 tunes are the current options.
Published: August 2026.
Models covered
Skip the filter
Run these models with adult content enabled
MyBabes runs Wan 2.7, Seedance 2.0 and Krea inside its own generation stack, so the prompts refused on official apps generate as written. No GPU, no setup.
Open MyBabes
Editorial team
The MyBabes Lab team tests AI image, video and chat models for adult use cases: what they allow, what they block, and how to get the best output. Every note here is based on hands-on runs, not vendor marketing.
Follow on X
More from the Lab
Latest articles
All articles
Analysis
AI Porn and the Law in 2026: What's Actually Illegal, What's Regulated, and Where the Lines Sit
2025 and 2026 turned AI adult content from a legal grey zone into a regulated space with real enforcement: a US federal crime for non-consensual imagery, deepfake laws in 48 states, UK fines for missing age checks, EU labeling duties. Here is the map — and the one principle that organises all of it.
·
5 min read
How-To
Why AI Porn Looks Off: Every Artifact Explained, What Causes It, and How to Fix Each One
Everyone can tell when an AI image is off, few can say why. Each artifact has a specific technical cause and a specific fix — hands, limbs, faces, skin, teeth, text, seams, smear, flicker. This is the field guide, for images and video.
·
5 min read
Analysis
ChatGPT and NSFW in 2026: The Adult Mode That Never Shipped, What's Actually Allowed, and Where People Went Instead
"Treating adults like adults" was the promise in October 2025. Ten months later the adult mode is paused with no timeline, ChatGPT still refuses explicit content, and the search interest hasn't gone anywhere. Here is the full story, the current policy line in practice, and what the alternatives actually are.
·
4 min read