How to Use MMAudio V2 in ChatGPT
MMAudio V2 adds synchronized sound effects and ambience to existing video, using a 44.1 kHz audio model. The authors recommend V2 and report better generalization to new footage than the earlier MMAudio checkpoint.
Prompts to try with MMAudio V2 in ChatGPT
Copy an example into ChatGPT, or ask in your own words.
Use MMAudio V2 on this 8-second beach video. Request gentle surf and distant gulls, with the waves as the main sound and no speech or music.
Use MMAudio V2 on this 8-second dog video. Request barking that follows the visible activity, with quiet outdoor ambience and no music.
Use MMAudio V2 to add sound to this 8-second golf shot. Request one sharp ball strike at contact and a quiet outdoor background.
Use MMAudio V2 on this 8-second video of someone opening a cardboard box. Request close cardboard rustling and soft handling sounds, with no voices or music.
Use MMAudio V2 on this 8-second rain video. Make two attempts with the same sound description, increasing how strongly the second follows the text, so I can compare them.
How to set up MMAudio V2 in ChatGPT
Pick the way that suits you. Each takes about two minutes.
Use the ChatGPT connector
- Add a custom MCP server
In ChatGPT open Plugins, then Add › Add custom MCP server. Name it Spiral, paste this URL and keep Authentication on OAuth. Tick I understand and want to continue, then click Create.
https://api.spiral.ai/mcp
- Click Connect
ChatGPT asks to connect Spiral. Connect takes you to Spiral in your browser.
- Sign in to Spiral and approve
Sign in, pick the organization ChatGPT should use and click Authorize. Spiral gives the connection its own API key; ChatGPT never sees it, and you can remove the connection under Settings › Connected apps.
Use the Spiral skill
- Send ChatGPT this message
Your agent reads Spiral's skill and sets itself up.
set up https://spiral.ai/SKILL.md
- Sign in or paste a key
Approve Spiral in the browser, or give the agent a key from the console.
Use the Spiral CLI
- Install and sign in
Installs the Spiral CLI, signs you in and adds its skill to your coding agents.
npx -y spiral-cli@latest init
Call the Spiral API
- Create an API key
In the console, under API keys.
- Run MMAudio V2
One request; the result comes back by polling or a webhook.
curl https://api.spiral.ai/v1/runs \ -H "Authorization: Bearer $SPIRAL_API_KEY" \ -H "Content-Type: application/json" \ -H "Idempotency-Key: $(uuidgen)" \ -d '{ "provider": "fal", "model": "fal-ai/mmaudio-v2", "input": { "video_url": "https://storage.googleapis.com/falserverless/model_tests/video_models/mmaudio_input.mp4", "prompt": "Indian holy music" } }'
Then paste one of the prompts above: ChatGPT finds MMAudio V2 on Spiral, shows the price and runs it.
MMAudio V2 providers and prices
Spiral runs MMAudio V2 through fal. Open a card for today's price, with every fee included.
Settings you can ask for
MMAudio V2 works from your video URL and prompt. Ask ChatGPT for any of these in your prompt.
| Setting | Allowed values | Default | Description |
|---|---|---|---|
| Negative prompt | what to leave out, in words | The negative prompt to generate the audio for. | |
| Seed | 0 to 65535 | The seed for the random number generator | |
| Number of steps | 4 to 50 | 25 | The number of steps to generate the audio for. |
| Duration | 1 to 30 | 8 | The duration of the audio to generate. |
| CFG strength | 0 to 20 | 4.5 | The strength of Classifier Free Guidance. |
| Mask away clip | on or off | off | Whether to mask away the clip. |
Every API field of MMAudio V2 · Video to Video is on its tool page: open its card above.
What can you do with MMAudio V2 in ChatGPT?
Turn a short video into a sound-design draft with MMAudio V2. Describe the sounds in conversation, then compare the result with the visible action.
Foley for silent AI footage
Add ambience and action sounds to short generated shots. Review each result against visible movement before using it in an edit.
Nature footage
Try environmental sound for short outdoor shots, describing the audible setting and the main visible action.
Animal clips
Generate barking or howling for footage with visible animal activity. Check that vocalizations occur at plausible moments.
Sports impact sounds
Add a draft sound pass to a short golf or racket-sport shot, then inspect the contact point for timing errors.
Percussion footage
Try synchronized hits for a short drum performance. Listen for missed strikes and extra hits before keeping the result.
Product handling videos
Prototype Foley for short shots of objects being opened, placed, or handled. Describe the materials and desired sounds as a sound-design brief.
Sound alternatives for an edit
Generate several interpretations of the same short shot by changing the sound description. Compare ambience, prominence of effects, and synchronization.
More on MMAudio V2 and ChatGPT
Overview
MMAudio V2 in other agents
Questions about MMAudio V2 in ChatGPT
Can ChatGPT use MMAudio V2?
Yes. Add Spiral to ChatGPT as an MCP server, and ChatGPT can find and run MMAudio V2 like any built-in tool.
How much does MMAudio V2 cost in ChatGPT?
$0.0013 per second. Billed on `duration`, default 8. You pay from your Spiral balance as you go, with no subscription, and a failed run is not charged. Your ChatGPT plan is separate.
Do I need a fal account to use MMAudio V2 in ChatGPT?
No. Spiral runs MMAudio V2 on its own fal account. You sign in to Spiral once and never paste a key.
How long does it take to set up MMAudio V2 in ChatGPT?
About two minutes: add a custom MCP server, then click Connect, then sign in to Spiral and approve.
Does ChatGPT wait while MMAudio V2 runs?
MMAudio V2 runs in the background. ChatGPT starts the run, checks on it and gives you the video when it's ready.
What is MMAudio V2 useful for?
Use it to add sound effects and ambience that follow visible action in existing footage. Its recommended checkpoint generates 44.1 kHz audio.
Can MMAudio V2 generate dialogue or a musical score?
Its focus is Foley: sound effects and ambience. Human speech requests can produce unintelligible mumbles, so use a dedicated speech or music model for those tasks.
What clip length should I start with?
Start with an 8-second shot, which matches the training duration. The authors warn that large departures from that length can reduce quality.
Will a 4K source produce better sound?
Higher source resolution does not improve the generated audio. Choose footage where the action is clear, and judge the result by listening to how the sound matches it.
How should I prompt MMAudio V2?
Describe the sounds you want, such as a sharp ball strike with quiet outdoor ambience. If the result follows the text too loosely, try increasing guidance strength and compare the timing again.
Start using MMAudio V2 in ChatGPT
One Spiral account for every tool your agent needs. No subscription; you pay per run.
Paste this into ChatGPT:
set up https://spiral.ai/SKILL.md


