Meta Opens Early Access Program for New Muse Features

Meta opens an invite-based early access program for new AI creator tool Muse features.
Meta has launched an Early Access Program for new Muse features, using an application-based gradual rollout model rather than a broad release. This approach filters for engaged early testers, helping Meta collect feedback and validate stability before a full launch. Official details on the specific features, coverage scope, and review criteria have not yet been disclosed — interested users can apply to Muse to join the waitlist.
Meta Launches an Early Access Channel for Muse
Meta has opened an Early Access Program for new Muse features to interested users. To participate, users must proactively reach out to Muse and request to be added to the waitlist — only then will they have the opportunity to try out features that haven't officially launched yet.

Based on publicly available information, this is a classic invite/application-based gradual rollout approach. Rather than pushing new features to all users at once, an early access program allows Meta to gather real-world feedback, monitor feature stability, and iterate before deciding whether to roll out more broadly.
Muse is Meta's AI tool product aimed at creators, positioned to assist with content creation and creative generation. As part of Meta's broader AI ecosystem strategy, Muse is closely tied to the company's ongoing investments in large language models, image generation, and related areas. Opening up early access is a common cadence for Meta to accelerate iteration and accumulate user data in the competitive AI creative tools market — limited access in exchange for real-world usage signals to inform future product decisions.
How to Apply and What to Expect
The barrier to entry for this early access program is relatively low, but it does require users to take the initiative. According to official information, anyone interested must actively request to be added to the list. In other words, the feature won't be automatically unlocked — it follows an "apply first, review later" model.
The advantages of this approach are clear: it filters for users who are genuinely interested and willing to provide feedback, while also giving Meta control over the rollout pace and preventing unvalidated features from impacting the experience of all users.
Why an Early Access Model?
Early access programs are quite common in tech products, and have become especially mainstream as AI-related features undergo rapid iteration. For companies, early adopters are a valuable testing resource — they can surface potential issues and validate product assumptions before a full launch.
For users, joining such a program means getting first access to new features and staying ahead of the curve. That said, features in the early access stage are often still being refined and may be unstable or subject to change — that's the trade-off that comes with being an early tester.
Gradual rollout is a standard risk management practice in internet product engineering. The core idea is to release a new feature to a small subset of users first, validating system stability and user acceptance with real traffic before progressively expanding coverage. For AI features in particular, gradual rollouts are especially meaningful: model inference costs are high and outputs are difficult to exhaustively test in a sandbox — limited real-world usage often surfaces edge cases that controlled environments simply can't catch. Early access adds an extra filtering layer on top of gradual rollout by requiring users to proactively apply, ensuring early testers have higher engagement intent and stronger feedback capabilities, which in turn improves the quality of testing signals.
Summary
For now, Meta has not disclosed further details about the specific new Muse features, the scope of access, or the review criteria. Users who are interested can try submitting a request to Muse to join the waitlist and wait for further updates. As the rollout gradually expands, more information is expected to become available.
Related articles

Random Pruning of the DPO Computation Graph: 1.4x Training Speedup and Lower Peak Memory
DPO training too slow and memory-hungry? This paper proposes a random pruning framework for the DPO computation graph, sampling independently across position, sequence, and branch dimensions; branch pruning achieves a 1.426x step-time speedup and cuts training time by about 30%, with in-depth attribution of memory and time gains.

Why Does JEPA Underperform MLM in Longitudinal EHR Representation Learning?
New study compares JEPA and MLM for longitudinal EHR representation learning on MIMIC-IV: three JEPA variants train stably without collapse but underperform MLM; target size, not heterogeneity, drives prediction error.

Transfer Learning + Visible Light Communication: A New Approach to Energy-Efficient Indoor Positioning
A study proposes a transfer learning-based visible light communication indoor positioning method, achieving 47% accuracy improvement on the BOSCH factory dataset, 67.5% model efficiency under high noise, and introducing the ME composite metric for energy-efficient positioning in energy-constrained industrial IoT scenarios.