Skip to content

A motion design studio on your own machine: Claude makes the video step by step, you approve every stage. Bring your own Claude Code or API key.

6.1 KB
Raw Blame

Narration, sound effects and music

Mortiflix can narrate your videos three ways. Pick one in Settings › Narration (web) or with mortiflix voice. ElevenLabs is the default and the only voice the studio generates: there is no local voice (Qwen3-TTS and ComfyUI were removed on 2026-11-01). Without an ElevenLabs key, Settings and the Voice stage say so plainly; nothing falls back to a local model.

ElevenLabs Your own voice No narration
Sound the most natural voices; 90+ languages on Eleven v4 you on-screen text and music carry it
Cost per character, on your ElevenLabs plan free (your time) free
Privacy text goes to ElevenLabs nothing leaves your machine nothing to send
Needs an API key from elevenlabs.io a microphone (and ffmpeg for the terminal booth) nothing
Voices thousands (yours, the defaults, the Voice Library, your own clones) yours
Also sound effects and music beds

Generated lines are checked by speech to text after they're made: a take with missing or extra words is made again, and the timings of every word go to the animation. Your own recordings are timed line by line.

Your own voice: the recording booth

Choose Your own voice in Settings › Narration (or mortiflix voice own). When a video reaches its narration, the studio writes the script as short lines and pauses the project with a note asking you to record. Then any of:

  • In the web studio: open the project and choose Record narration. Check your microphone and the room, then read each line: hold Space (or the big button) while you speak, let go to stop. It plays the take back and flags a take that clipped, is very quiet, or seems too short or long. Record as many takes as you like, Enter keeps the selected one and moves on, ← → move between lines. Takes save as you go, so you can stop and come back. When every line has a kept take, Done hands it to the studio. (Browsers allow the microphone on the computer the studio runs on, or over HTTPS.)
  • In a terminal: mortiflix record <project>. Enter starts and stops a take, p plays it, k keeps it, n/b move between lines, q stops for now. It records with ffmpeg from the default microphone; pick another with --device (mortiflix record --list-devices).
  • From files you recorded elsewhere (a DAW, a phone): name each file after its line (b01-1.wav, b01-2.m4a, …; the booth and voice/lines.json show the ids) and run mortiflix record <project> --import <folder>.

If the studio rewrites a line after you recorded it, that line asks for a new take.

ElevenLabs

  1. Get an API key: elevenlabs.io › Developers › API keys.
  2. mortiflix keys (or Settings › Narration › ElevenLabs), paste it, Connect. You'll see your plan, credits left, when they reset, and whether you may use the audio commercially.
  3. Choose a voice: your voices and the default voices, or Browse the Voice Library (thousands of community voices; "Add & use" copies one into your account). ▶ plays each voice's own preview, free.
  4. Pick the model and settings, then Try it with a line of your own.

Everything is read from your account: the models list comes from the API, so new models appear by themselves.

The options, and what they do

Option What it does
Model eleven_v4 (default and recommended: highest quality, best cloning, 90+ languages), eleven_v4_turbo (fast), eleven_v3, eleven_multilingual_v2 (29 languages, has style and speed settings), eleven_flash_v2_5 / eleven_flash_v2 (fastest)
Stability lower is more expressive and varied, higher is steadier (all models)
Similarity how closely it sticks to the original voice (all models)
Style, speed, speaker boost only on models that have them; Eleven v4 doesn't (it's directed with audio tags instead)
Language an ISO 639-1 code to force the language and how numbers are read; blank detects it
Text normalization auto, on (spell out numbers, dates, abbreviations), off (read as written)
Pronunciation dictionaries up to 3 of your dictionaries, applied in order. Phoneme rules work on v4, v3 and Flash v2; other models use alias rules only
Audio format MP3 128 kbps (default) up to WAV 48 kHz. 192 kbps MP3 needs the Creator plan or above, 44.1 kHz WAV/PCM needs Pro
Server for accounts on a data-residency server (US, EU, India, Singapore)
Checks Scribe v2 speech to text on every take (recommended), or off
Sound effects lets sessions make effects with eleven_text_to_sound_v2: 0.5–30 s, loopable
Music beds lets sessions compose with Eleven Music (music_v2_5): instrumental, 3 s to 10 min. Usage terms depend on your plan: elevenlabs.io/music-terms

How lines are made: each one is sent with the text before and after it (so the delivery flows), a seed (so a good take can be reproduced), and only the settings the chosen model accepts. Lines run in parallel up to your plan's limit minus one (Free 2, Starter 3, Creator 5, Pro 10, Scale and Business 15).

Directing Eleven v4

Sessions write delivery into the script itself: audio tags like [warm], [whispers], [excited]; CAPITALS for emphasis; ellipses for pauses; and IPA between slashes for names ("/ˈkoʊmæl/"). Eleven v4 doesn't support SSML.

Good to know

  • Free plan: audio may be used non-commercially only, with attribution to ElevenLabs. Paid plans include commercial rights. Settings shows which you're on, and sessions say so in their notes.
  • Zero-retention mode (no logs kept at ElevenLabs) is for Enterprise accounts only.
  • Voice clones: only of your own voice, or with the speaker's clear consent.
  • Your key stays in the studio folder (secrets.json, readable only by you). Sessions get it only while ElevenLabs is the narration engine.

No narration

Videos are made with on-screen text, music and sound. If a brief asks for a voice anyway, the session asks you at the next review whether to continue without one.