ElevenLabs will now build the whole ad — you just describe it

Describe a thirty-second product ad in one sentence, and ElevenLabs will build the pipeline for it: pick the image model, pick the video model, wire in the voice, sync the mouth, score it, and run the whole chain on a single canvas. That is not a launch-video promise. It is what the Flows Agent does inside ElevenCreative, and according to ElevenLabs’ Flows documentation it is available on every plan — the free one included.

For a couple of years, ElevenLabs was the tab you opened when you needed a voiceover. In 2026 it became the room where the whole video gets made.

One canvas, fifty models, one login

Flows launched on 11 March 2026 with more than 35 image and video models sitting next to ElevenLabs’ own speech, lip-sync, music and sound-effects models. The current documentation now puts that number above 50. Veo, Sora, Kling, Flux and Seedance are among the names ElevenLabs lists on its Flows page — the models you were paying four separate subscriptions to reach.

You reach them from one login, on one canvas, drawing a line from one box to the next.

The part that changes how a workday feels is not the model count, though. It is what happens when the client comes back with notes. In a normal chain, a new voiceover means re-exporting the video, re-uploading it somewhere else, and rebuilding the timing. In Flows, ElevenLabs documents non-destructive iteration: re-run the speech node, and the generated video above it stays exactly where it was.

A quick orientation

ElevenLabs presented itself as three platforms in its February 2026 Series D announcement: ElevenCreative for making, editing and localizing content, ElevenAgents for interactive voice and chat agents, and ElevenAPI for developers embedding the models in their own products.

If you make videos, you want ElevenCreative. That is where Flows, Studio, dubbing and the downloadable, licensable version of the music live. There is also a separate consumer app called ElevenMusic for creating and remixing tracks socially — ElevenLabs’ own explainer is clear that the professional downloading, management and licensing sit in ElevenCreative instead. Start in the right place and you save yourself an afternoon.

Watch a real job go through it

Say you are making six versions of a product short for a client — three visual treatments, two music directions, one voice.

Here is how that maps onto the canvas. Drop in the packshot. Send it to three image nodes with different treatments. Pick the one that works, and feed that frame into a video model to animate it. Attach a speech node for the voiceover, run it through lip-sync if there is a presenter on screen, and hang two music nodes off the end. Preview the whole thing in a composition node, then export the assets or push the project into Studio for timeline work.

The client picks treatment two and asks for a warmer read. You re-run one node.

That is the pitch, and it is a real one. Two things are worth knowing before you build your month around it. Every successful re-run spends credits — a node that fails does not, per the Flows documentation. And programmatic execution, the kind where a script fires a thousand jobs, is still on the roadmap rather than in the product. Flows was labelled Alpha at the time of writing.

The audio is still the reason it works

The generation layer is new. The audio stack is the part ElevenLabs has been building for years, and it is why the pipeline holds together at all.

Eleven v3 left Alpha on 2 February 2026. The models documentation credits it with expressive direction tags, multi-speaker dialogue and more than 70 languages, within a 5,000-character request. It is the one to reach for when a line needs acting. For long, even narration, the same documentation points at Multilingual v2 and its larger 10,000-character ceiling.

Scribe v2 handles transcription and captions in more than 90 languages, with word-level timestamps and speaker separation — which is what makes automatic captions usable rather than a cleanup job.

Dubbing v2 covers more than 90 languages too. It is still Alpha, and the free tier adds a visible watermark to the dub. Paid tiers do not.

Sound Effects v2 generates effects up to 30 seconds from a prompt, and will loop them.

Music v2 writes full tracks with vocals. English, Spanish, German and Japanese are the languages ElevenLabs names, with more implied. Read the licence section below before you put one under client work — this is the single place where ElevenLabs’ rules are stricter than the interface suggests.

Studio: where the rough cut comes together

Flows is the nonlinear side. Studio 3.0 is the linear one, and it is where a narrated video actually gets assembled: tracks for video, captions, narration, music and effects, with sentence-level timing, paragraph-level regeneration, voice changing, comments and video export, per the Studio documentation.

Picture a ten-minute explainer. Upload the edit, transcribe it, fix the script, regenerate the narration for the two paragraphs you rewrote, drop in a music bed, and burn in captions — on one timeline, without a single export-and-reimport round trip.

What Studio does not document is colour, compositing or finishing. In our reading, that puts it before or beside your main editor rather than instead of it: build the narrated first cut or the localized version here, then finish in Resolve or Premiere when the job needs more than timing and tracks. For a large share of YouTube, social and explainer work, it never does.

Which plan you actually need

Credits are shared across every product, so the number on the plan is not a fixed quantity of finished video. Speech bills by characters, transcription and dubbing by audio time, and generation nodes by whichever model you picked.

Prices are monthly self-serve rates before tax, from the ElevenLabs pricing page, checked 5 August 2026.

Here is the short version. Free is a real trial, not a demo — Flows runs on it, and you can build a full pipeline before spending anything. Starter, at $6, is the line that matters, because it is where the commercial licence begins: below it, nothing you make can be monetized. Creator, at $22 with a first month at $11, is where most working solo creators land, because it adds Professional Voice Cloning and removes the watermark from Studio video exports. Pro is for when audio delivery quality is the point.

Unused paid credits roll over for up to two months, capped at twice your monthly quota. Downgrade or cancel and the balance goes at the end of the cycle.

That recommendation is our analysis, not ElevenLabs’ own.

Three things to know before you start

Music rights depend on where the track ends up. Under the Eleven Music Model-Specific Terms, Starter through Business covers online and offline commercial work, but film, television, radio and multi-platform commercial studio games are excluded; only the full Enterprise Music plan lifts those. Distributing tracks to music streaming services needs Creator or above. Reselling output as a sample pack or music library is off the table on every self-serve tier. ElevenLabs’ ElevenCreative FAQ also flags marketing campaigns and advertising as needing an additional licence, so check the terms against your specific use before you score a client campaign. For a monetized YouTube video, a paid self-serve licence fits comfortably. For a short film going to a broadcaster, it does not.

Iteration costs money. Every successful re-run spends credits. On expensive third-party video nodes, an afternoon of heavy iteration is a real line item — worth knowing when you quote a job.

The image and video models are not ElevenLabs’ own. They are partly powered by outside providers, and ElevenLabs excludes its usual IP indemnity for those outputs in its Image & Video Terms. One invoice does not make third-party models first-party ones. In practice this matters for high-stakes brand work, not for a YouTube upload.

Who this is for

If you make narrated, captioned or localized video on a repeating schedule — explainers, product shorts, course modules, social cutdowns — this is the strongest fit, and the reason is not any single model. It is that the pipeline is reusable. Build the chain once, then run it with new inputs next week.

If you need deep colour work, heavy compositing or broadcast finishing, keep your editor. ElevenCreative is the generation room, and it is very good at being one.

For the full breakdown — scoring, export limits, data handling and how ElevenLabs compares with everything else we researched — see our ElevenLabs review and the full tool directory.