Can AI make music and sound effects for a video?

Yes. FLORA has music models such as Lyria 3.5 and Suno v6 and a sound effects model from ElevenLabs, all called from the same canvas or agent chat as your video. In our run, Lyria 3.5 scored a still from the video and returned a 2 minute 55 second track in 65 seconds for $0.105. A 5-second ElevenLabs sound effect came back in about 9 seconds for $0.144.

At a glance

Key features

• Music from an image: Lyria 3.5 accepts an image, so the track can be briefed with a frame from the video.
• Sound effects by length: The sound effects model takes a duration from 0.5 to 30 seconds.
• Voiceover: Text-to-speech models in FLORA read a script for narration.
• Timeline editor: FLORA's timeline lays music, effects and clips on separate tracks and renders one MP4.

Best for

• Social clips that need a music bed in minutes
• Product videos where a few sound effects sell the action
• Animatics that need scratch music for timing
• Creators without a music library license

Step by step

1. Step 1: Brief the music like a composer

   Give mood, instruments, tempo and what to leave out. Attach a frame from the video if the model takes an image.

   Make a music track with Lyria 3.5 through FLORA, using this frame as the image input. Quiet instrumental score for a short documentary clip of a fisherman coiling rope on a foggy pier at dawn. Solo acoustic guitar and soft cello, slow tempo around 70 bpm, melancholy but warm, sparse, no vocals, no drums. Fades out at the end.

2. Step 2: Ask for each sound effect by name and length

   One effect per request, with the length you need. Ask for a quote first if you want the cost before it runs.

   Make a 5-second sound effect with ElevenLabs Sound Effects through FLORA: wet rope dragged over wooden pier planks, waves lapping against pilings, distant gulls.

3. Step 3: Lay it under the clip

   Ask the agent to put the clip, the music and the effects on a FLORA timeline, trim the music to the clip and render.

What happened when we made music and a sound effect in FLORA?

On October 16, 2026, we scored the 5-second fisherman clip from our Claude video guide. The agent called flora_list_models for audio, then flora_create_generations with type audio, model t2a-lyria-3-5-t2a, the music brief from step 1, and the clip's first frame in params.image_url.

Lyria 3.5 finished in 65 seconds and was charged $0.105. We downloaded the file and checked it with ffprobe and ffmpeg: a 192 kbps stereo MP3 at 44.1 kHz, 175.3 seconds long, with different left and right channels. It measures -11.3 LUFS integrated with a 13 LU loudness range, and its peaks reach 0 dBFS. In 5-second windows it opens at -25 dB, reaches about -14 dB by the 90-second mark, peaks near -9 dB around 2 minutes 10 seconds, and falls from -26 dB to -57 dB over its last 8 seconds. A silence check found no gaps inside the track.

For the sound effect, the agent called flora_create_generations with type audio, model t2a-elevenlabs-sfx, duration_seconds 5 and the rope and waves prompt from step 2. It finished about 9 seconds after submission and was charged $0.144. The file is a 128 kbps stereo MP3 at 44.1 kHz, 5.04 seconds long, at -15.5 LUFS integrated with peaks at 0 dBFS. Measured in half-second windows, its level moves between -30 dB and -15 dB and drops to -45 dB in the last half second, so it tails off rather than cutting. No silent gaps, and the two channels differ.

These checks cover format, length and level. They say nothing about how the files sound, so listen before you mix. The Seedance 2.5 clip itself already carried a quiet ambient track, about -38 dB mean, which a mix can keep under the music.

What went wrong: we asked for a sparse cue for a 5-second clip and got a 2 minute 55 second track that climbs about 16 dB from its opening to its loudest stretch. The quiet opening is the part that fits a 5-second clip, so you have to trim. Both files also peak at 0 dBFS, so lower them on the timeline before mixing them with the clip. The fix is to ask for the length you need, such as "a 15-second cue", and to trim, fade and set levels on the FLORA timeline.

Music model

Lyria 3.5, id t2a-lyria-3-5-t2a

Music cost and time

$0.105, 65 s

Music output

MP3, stereo, 44.1 kHz, 2 min 55 s

Music loudness

-11.3 LUFS integrated, 13 LU range, peaks at 0 dBFS

Sound effect

t2a-elevenlabs-sfx, $0.144, about 9 s

Sound effect output

MP3, stereo, 44.1 kHz, 5.04 s, -15.5 LUFS

Music model or sound effects model: pros and cons

How do I brief AI music for a video?

Write it the way you would brief a composer for a cue.

• Mood in two or three words.
• Instruments, and what to leave out, such as drums or vocals.
• Tempo in beats per minute.
• Length, and how it should end.
• A frame from the video, when the model takes an image.

Do AI video clips come with sound?

Some do. Our Seedance 2.5 clips came back with an ambient audio track generated from the prompt. Music and effects go on separate tracks above it, so you can keep, lower or mute the clip's own sound.

Can I add a voiceover too?

Yes. FLORA has text-to-speech models from ElevenLabs and Google. Paste the script, pick a voice, and place the file on the timeline with the music and effects.

Frequently asked questions