Introduction

Chimera is a new workflow from ElevenLabs that makes a character talk on camera. ElevenCreative, the company's creative workspace,

: you create or upload a character, write their line, pick a voice or design one, and Chimera returns a video of that character speaking it. The voice comes from Eleven v4, ElevenLabs' newest text-to-speech model, and the picture comes from ByteDance's Seedance 2.5 video model.

It's closed and hosted, with no weights to download, and you run it inside ElevenCreative in the browser. What makes it interesting for anyone who works with voice is where the control sits. You direct the performance in the script itself, with audio tags like [whispers], [laughs] and [yells], and the face follows the read. In ElevenCreative's words, "Eleven v4 puts the emotion in the voice. Seedance 2.5 carries it onto the face."

What Shipped

Chimera isn't a new model. It's a ready-made chain of two models that creators were already running one after the other, now joined into a single step.

  • Voice: Eleven v4, which ElevenLabs launched on September 28. It reads inline tags for timing and emotion, so one line can shift from a whisper to a laugh without splitting it into separate takes. We covered it in Eleven v4 brings emotive TTS.
  • Picture: Seedance 2.5, which animates the character and syncs the lips and expression to the audio.
  • Input: one character image (yours, or one you make in ElevenCreative), a line of dialogue, and a voice from the library or one you design.
  • Output: a video of the character delivering the line.

The ambassador clips that flooded X over the last two days show the range people are going for: whispered ASMR, horror monologues, court gossip, and one face cycling through five emotions where, as one creator put it, "all I changed was the brackets."

The Specs Behind It

ElevenLabs hasn't published a Chimera page, spec sheet or per-video price yet. The limits you'll run into come from the video side, which ElevenLabs does document on its Image & Video capabilities page:

  • Clip length: Seedance 2.5 generates from 4 to 30 seconds per clip.
  • Resolution: 480p or 720p, in 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16.
  • References: up to 30 images, plus up to 10 video and 10 audio references of 30 seconds combined each.
  • No seed control, so you can't lock a take and re-run it exactly.
  • Cost: listed as varying with settings and duration, paid in ElevenLabs credits.

Who Can Use It

There's one big catch. ElevenLabs' docs say plainly that Seedance 2.5 is not available in the United States, because it comes from a China-based provider. Chimera runs on Seedance 2.5 and ElevenLabs hasn't announced an exception, so expect the template to be missing for US accounts. Creators elsewhere can use it today.

Try It Yourself

Keep the line short, since every second counts toward a 30-second clip, and put the tags right where you want the shift to happen. A few to start from:

[whispers] Don't move. [pause] Did you hear that? [laughs nervously] Okay. It was just the wind.

I told you the kingdom keeps no secrets. [leans in, whispers] But this one... [laughs] this one is mine.

[tired] Three hundred years of guarding this door. [sighs] And nobody ever knocks.

For the cleanest lip sync, use a front-facing character image with a clear, unobstructed mouth, and design the voice to fit the face before you start tagging.

Conclusion

Voice is usually the weakest part of an AI character clip, and Chimera tackles that head on by letting the voice lead and the face follow. The tag-based control is the real draw: you direct the read like a voice actor's script instead of adjusting sliders. The open questions are price per clip, how well a character holds up across many lines, and US access, so watch ElevenLabs' docs for a proper Chimera page.

If you'd rather build Seedance-style shots yourself and add your own voice track later, try this on Gen.

—Anabel ♡