The rest of the field

Meta Muse Video

This roundup gathers the video models that each earn a mention without needing a full page of their own, and the most notable of them is Meta Muse Video. Previewed on 7 July 2026, Meta Muse Video shares a pre-training base with Muse Image, the company’s image model, adds native audio, and entered at third on the Artificial Analysis text-to-video board. A debut in third place on a public leaderboard is a strong showing for a first preview, and it puts Meta Muse Video straight into the conversation with the established names.

The shared pre-training base is the interesting technical detail. Building a video model on the same foundation as an image model lets the two systems draw on common visual understanding, and it hints at Meta’s intention to run a single family across still and moving pictures rather than treating them as separate projects. Native audio, generated with the footage rather than added afterwards, brings the model into line with the other frontier systems that now expect to deliver sound and picture together.

Two caveats

Two caveats belong with the praise. First, there is no public API, so for now Meta Muse Video is a preview rather than a tool a developer can build on. Second, and worth reporting plainly, public Instagram accounts are opted into at-mention remixing by default, meaning content can be drawn into the remixing feature without an explicit choice by the account holder. That default has drawn criticism, and it is the kind of consent question that tends to follow generative tools once they are attached to a large social platform.

The other three

The rest of the field covers three more names, each with a clear niche. Pika 2.5 remains the fast, stylised social option: a tool built for quick, characterful clips aimed at feeds rather than for photoreal output, which keeps it useful for creators who value speed and a distinct look over strict realism. It holds much the same social-first territory as the other creator-focused models, and it has kept that position through several releases. For a lot of everyday social output, that combination of speed and a recognisable house style is exactly what a creator wants, and it explains why the tool has kept a loyal following even as flashier rivals have arrived.

Grok Imagine is xAI’s entry, covering both image and video generation under one banner. Bundling still and moving generation together fits xAI’s pattern of folding new capabilities into a single product rather than shipping them separately, and it places Grok Imagine as the video arm of a broader generative push (see our image models hub for the still-picture side of the same trend). Marey, from Moonvalley, sits at the opposite end of the spectrum from the fast social tools: it is the licensed-data option built for film production, aimed at studios that need to know their footage came from a model trained on properly cleared material. For professional film work, where a rights question can halt a project, that clean data footing is the feature that overrides everything else.

What to watch

Taken together, these four sketch the breadth of the video field beyond the headline models. Meta Muse Video shows a platform giant entering strongly and immediately running into the consent questions that come with scale. Pika 2.5 holds the fast, stylised social corner. Grok Imagine ties video into xAI’s wider ambitions. Marey stakes out the licensed, film-grade ground. What to watch across all of them is the same pattern visible everywhere in generated video: native audio becoming standard, licensing and consent moving from a footnote to a central concern, and the gap between a preview and a usable product shrinking with each release. For the fuller picture, the AI video hub tracks the models covered in depth elsewhere in this section.