Meta Spins Vibes Into a Standalone App, and the Feed Is Entirely AI

Meta turned Vibes into a separate app on 27 September, and the pitch is unusual enough to deserve a plain description: it is a short-video feed where nothing was filmed. Users type a prompt, get a vertical clip, and scroll a stream built only from generated video. The company shipped it to North America, Australia, New Zealand, the UK and parts of Europe under the label "early preview," which is corporate language for shipping something before it is finished on purpose.

Vibes started life in September 2025 as a button inside the Meta AI app. TechCrunch reported in February that Meta was already testing a standalone version in Brazil and Mexico after what it called strong early traction. The September release is that test going wide. Separating the feed from the main app lets Meta iterate on it without dragging the assistant along, and it lets the company treat AI video as a product line rather than a feature.

What the app actually does

Four things happen inside Vibes, and only the first is about generation.

Text to video is the entry point. A user writes something like a cat kneading dough or an astronaut dancing in a cyberpunk station, and the app returns a short clip with image and sound. Generation from existing assets is the second door: upload a photo or a clip, and the app animates it, restyles it, or rebuilds the scene. Meta has said most of this processing happens on the device, which it frames as a speed and privacy advantage over a cloud-only pipeline.

Remixing is the feature Meta actually promotes. Instead of the duet format most platforms use, Vibes lets a user pull a clip out of the feed, including some content from Instagram, and have the model reinterpret it. Characters change, environments shift, the color palette moves. The structure of the original survives.

Then comes distribution. A finished clip can go to the Vibes feed, into a direct message, or out to Instagram and Facebook Stories and Reels. Instagram already has a path back in: a Meta AI video seen on Instagram can be tapped through to be remixed in the app. That loop is the part worth watching, because it makes an AI-only feed a feeder for the surface where hundreds of millions of people already post filmed video.

The editing tools have grown with the app. The current version adds reusable AI characters, lip-sync for speaking figures, and a timeline editor for longer projects. There is also music, including generated music, plus overlays and filters.

The models behind it

At launch, Meta's chief AI officer Alexandr Wang said the early version leaned on partnerships with Midjourney and Black Forest Labs while Meta developed its own models. Social Media Today reported the Midjourney partnership in the same news cycle. Vibes as a product therefore sits on a mix of outside image models and Meta's own video engine, with Llama models handling the language layer. That arrangement is worth remembering when the quality of the feed is discussed, because it is not a single-vendor stack.

Freemium, and the AI slop question

Meta is running Vibes as freemium. Basic access stays free, with paid tiers that raise monthly creation limits. The subscriptions are being explored across Facebook, Instagram and WhatsApp with Vibes playing a central role, which is a familiar shape: give away the discovery, charge the heavy creators.

The obvious risk is what critics call AI slop. A feed made entirely of generated video has no floor of real footage to anchor it, and the volume can rise without the quality moving with it. Every platform that has opened a firehose of generative content has hit this. Meta's answer so far is labeling. Its disclosure stack, built on C2PA and IPTC metadata plus self-disclosure, is what carries into Reels, and the question policy people are asking is whether an origin tag from Vibes counts as the visible label European rules require at first exposure when a clip resembles a real person or event. Remixing someone else's generation does not exempt the result from that duty.

There is a rights problem folded into the fun part. A remix is downstream of an upstream prompt, an upstream likeness and an upstream music bed. Treating that output as original brand work is a mess, and the music layered inside Meta, inside Meta, does not travel with a file that leaves. For anything with a client or a real face in it, starting from assets you hold is the only clean route.

Where it sits against the other tools

Vibes is not entering an empty field. The comparison that keeps coming up is with Sora, which offers higher-quality generation and longer clips of up to 60 seconds, at the cost of a subscription and slower generation. Runway sits at the professional end with consistent characters and scene coherence across shots for around 16 seconds a clip. Pika covers the style-driven end. Vibes is positioned at none of those price points, because it is not really selling generation. It is selling discovery, with creation attached.

That distinction matters for what the app is good at. A creator who needs one polished 30-second spot will not find Vibes the right tool. Someone who wants to publish three clips a day into a feed where the algorithm decides who sees them is squarely in its territory. The 15-second ceiling on clips reinforces the point. Vibes is built for volume and speed, not for a piece of work anyone will revisit.

The early user reports back that up. One first-hand write-up described feeding three product images and a slide deck into the app, getting a 10-second product scene generated in about two minutes, then asking for a 360-degree rotation that came back looking more realistic than a manual 3D render, and finally having the two clips stitched into a single commercial. The same user noted that downloads carry no AI watermark, which they counted as a point in the app's favor and which others will count as a point against. Meta's disclosure stack is not the same thing as a visible mark on the file.

The European rollout is the harder test

Meta has taken Vibes into parts of Europe, and that is a more demanding environment than the US launch. The EU's rules on AI transparency are stricter, and the practical requirement is that content resembling a real person, place or event has to carry a visible label at first exposure rather than only metadata inside the file. A C2PA provenance tag can travel with a clip and still leave a viewer unaware that what they are watching was generated. The label on screen is the part that does the work, and it is the part that is easy to leave to the platform.

The remix feature makes the labeling question sharper. When a clip is generated, then remixed, then cross-posted to Reels, the chain of provenance runs through several generations, and the person posting may not know the origin of the original. A disclosure regime that assumes one clean source per upload has to handle a stack of them. Meta has not published how it resolves that, and it is the kind of detail that tends to get worked out after a complaint rather than before one.

What the launch implies

Vibes is not trying to replace Reels, and Meta has not promised that. What it is doing is giving AI-native video a dedicated room and a door into the main platforms. If the format sticks, the interesting lesson will not be that Meta built another model. It will be that the distribution mattered more than the generation, because the generation was partly bought in.

The counterpoint is that a closed feed of synthetic video is a narrow place to build a habit. TikTok's pull comes from real people doing things the algorithm could not have predicted. A feed that can only contain what a model imagined has to work harder for the same attention. Whether Vibes clears that bar is the open question, and the early preview label suggests Meta is not certain either.