Add Sound to Video
Generate Foley and ambient audio for a silent clip with HunyuanVideo-Foley. One flat price of 20 credits, optional prompt, 48 kHz output, videos up to 100MB.
Add Sound to Video generates Foley and ambient audio from the visuals in a clip. HunyuanVideo-Foley watches the on-screen action — footsteps, impacts, fabric, weather, room tone — and builds a synchronised 48 kHz track around it. A prompt is optional and the price is a flat 20 credits, so a long clip costs the same as a short one.
- Input: One video
- Output: 48 kHz audio
- Prompt: Optional
- Price: 20 credits
How to use Add Sound to Video
Start with a clip that looks right and sounds empty. The prompt steers the mix, but the visuals do most of the work.
- Pick a silent or sound-poor clip AI-generated scenes, drone footage and product turntables benefit most. Clean production audio can be muddied by an added Foley layer.
- Decide whether to prompt at all Leave the prompt empty when the visuals are unambiguous. Write one when you want a specific texture, material or atmosphere.
- Describe one scene and its layers Name materials and actions: rain on a tin roof, footsteps on gravel, a latch and a thud. Avoid listing unrelated environments.
- Fix the seed when the texture works A fixed seed makes the result reproducible, so a re-run with a changed prompt stays close to the approved sound.
- Mix the track under the picture Use the generated bed at a low level beneath music or dialogue. Keep music generation separate; this tool does not write songs.
Settings reference
Add Sound to Video has three inputs and one flat price: a video, an optional sound prompt and an optional seed. HunyuanVideo-Foley builds a 48 kHz Foley and ambience track synchronized to the picture.
| Setting | Value | Notes |
|---|---|---|
| Video | Required, up to 100MB | The model analyses motion and scene context to place sounds against visible actions. |
| Prompt | Optional | Name surfaces, materials and layers in one scene. Leave it empty for the model's default interpretation. |
| Seed | Optional | Fix a value to keep a texture you like while changing only the prompt. |
| Output | 48 kHz audio | A Foley and ambience bed designed to sit under the picture, not a full mix. |
| Output length | Follows the input clip | There is no duration selector; the track matches the video. |
| Price | Flat 20 credits per video | No per-second billing. A long clip costs the same as a short one. |
| Resolution | Not applicable | The tool returns audio only and does not change the picture. |
What video-to-audio does
The model analyses motion and scene context, then synthesises the sounds those visuals imply. A character walking across gravel gets footsteps with the right cadence; a door closing gets a latch and a thud; a rainy street gets rain, tyres and distant traffic. Because timing comes from the video, the track stays aligned with the action instead of being laid over it.
The output is an audio bed, not a replacement for dialogue or music. It is aimed at clips that are silent by design — AI-generated video, stock footage, screen recordings, rough cuts — or at adding texture beneath an existing edit. If the clip already has clean production audio, adding Foley on top can muddy it rather than improve it.
- Action-synchronised Foley
- Ambient beds and room tone
- 48 kHz high-fidelity output
- Works on silent or low-sound clips
- Not a dialogue or music tool
Pricing: one flat rate
Every run costs 20 credits, regardless of clip length or output duration. That is unusual among video models, which usually bill per second, and it makes the tool cheap to try on a long clip. The only input limit is a 100MB maximum file size; within that, there is no per-second surcharge.
Because the price does not scale, there is no cost reason to pre-trim a clip. Costs come from re-runs instead: each generation is another 20 credits, so use the seed field to reproduce a result you like and change the prompt when you want a different texture.
- Flat 20 credits per video
- No per-second billing
- 100MB input ceiling
- Seed for reproducibility
- Each re-run costs 20 credits
| Setting | What it does |
|---|---|
| Video | Required; up to 100MB |
| Prompt | Optional mood or sound description |
| Seed | Fixed value for reproducible output |
| Output | 48 kHz audio |
| Price | 20 credits per run |
When to use it
Use it when a clip looks right but sounds empty: AI-generated scenes, drone footage, product turntables, animation previz, or a social cut where the original sound was unusable. It is also useful for layering: generate a clean Foley pass, then mix it under music at a low level to make the scene feel grounded.
It is the wrong tool for cleaning up dialogue, removing wind noise or generating a song. Those need a speech enhancer, a denoiser or a music model. The model also cannot invent sounds that the visuals do not support — a static shot of a wall gives it very little to work with.
- Silent AI video and stock footage
- Product and object showcases
- Animation and previz
- A Foley layer under music
- Not for dialogue repair or music
Writing an optional sound prompt
The prompt is optional, but it steers the mix. Describe the layers you want and the surfaces involved: “rain on a tin roof, distant thunder, water dripping into a barrel” gives the model more to work with than “rainy mood”. Name materials and actions — gravel, wood, metal, fabric, footsteps, engines — because those are the cues the Foley engine maps to sound.
Keep it to one scene. A prompt that lists unrelated environments pushes the model to spread sounds across the clip, which can produce a busy, incoherent track. If you want a specific moment emphasised, say when it happens. Leave the prompt empty when the visuals are unambiguous and you want the model's default interpretation.
- Name surfaces and materials
- Describe layers, not just mood
- One scene per prompt
- Anchor important sounds in time
- Empty prompt for default interpretation
Limits and alternatives
The input is a video up to 100MB. The prompt is optional and the seed can be fixed for repeatable output. There is no duration selector or resolution setting; the audio follows the clip. The result is a 48 kHz track designed to sit under the picture.
For background music rather than Foley, use Mureka V9 BGM or MiniMax Music 3.0. For a talking performance, WAN 2.2 Speech-to-Video animates a portrait from an audio track. For editing the picture itself, Kling Video Edit and WAN 2.2 Video Edit handle prompt-driven changes.
- Video input, 100MB maximum
- Optional prompt and seed
- No duration or resolution controls
- Mureka for music
- Speech-to-Video for talking faces
Prompt examples for Add Sound to Video
- Rainy street: Rain on wet asphalt, soft footsteps, distant traffic and occasional thunder, a quiet city ambience with no music. Layer weather under the whole clip.
- Kitchen ASMR: Close-up cooking sounds: a knife chopping vegetables, oil sizzling in a pan, a spoon stirring, subtle room tone. Name each action for tighter sync.
- Forest walk: Footsteps on dry leaves, birds in the distance, a light breeze through branches, occasional twig snaps. Works well with slow camera moves.
- Product reveal: Soft mechanical clicks, a low whoosh as the product rotates, clean studio room tone, no music. Keep it sparse for a premium feel.
- Action impact: Heavy footsteps on concrete, fabric movement, a metallic impact at the moment of contact, debris settling. Anchor the impact to the visible hit.
- Cozy interior: Crackling fireplace, a clock ticking, rain against a window, quiet room ambience. Good for looping background beds.
Pro tips for Add Sound to Video
The flat price makes this cheap to try and expensive to over-iterate. The prompt is the main quality lever.
- The flat price removes the reason to pre-trim Every run is 20 credits regardless of clip length, so there is no per-second saving from cutting the video first.
- Each re-run is another 20 credits Costs come from repetition, not length. Keep the seed when you like the result and change only the prompt when you want a variation.
- A busy track usually means a busy prompt Unrelated environments in one prompt push sounds across the whole clip. Describe one place and let the visuals carry the rest.
- It is not a dialogue or music tool The model builds sound effects from visuals. Use a speech enhancer for dialogue repair and Mureka V9 BGM or MiniMax Music 3.0 for music.
- Static visuals give it little to work with A locked-off shot of a wall has almost no action cues. The model performs best when the clip shows movement, surfaces and impacts.
Frequently asked questions
How much does Add Sound to Video cost?
A flat 20 credits per video, regardless of clip length.
Is a prompt required?
No. The model can generate audio from the visuals alone. A prompt is useful for steering the mood, materials or specific sounds.
What is the maximum video size?
100MB. Within that limit there is no per-second charge or duration-based price change.
Does it generate music?
No. It generates Foley and ambient sound. Use Mureka V9 BGM or MiniMax Music 3.0 for music.
Can it replace or clean up dialogue?
No. It builds a new sound bed from the visuals; it is not a dialogue isolation or noise-removal tool.
What does the seed do?
A fixed seed makes the output reproducible, so you can keep a result you like and change only the prompt when you want a variation.
What quality is the output?
48 kHz audio, intended to sit under the picture as a Foley and ambience track.
Why does the result sound busy?
Usually because the prompt lists several unrelated environments. Describe one scene and the layers within it, and let the visuals carry the rest.
Pair it with other audio and video tools
- Mureka V9 BGM
- MiniMax Music 3.0
- WAN 2.2 Speech-to-Video
- Video Extend
- Kling Video Edit
- Browse all AI video models
Specifications
- Provider: Wavespeed
- Model type: Audio generation
- Credit pricing: 20 credits per video
- Capabilities: audio-sync
About Conut.ai
Conut.ai is an AI image, video and audio generation platform with 50+ creative tools. It supports text-to-image, image-to-video, face swap, motion control, relighting, inpainting and video effects through models including Nano Banana Pro, Seedream, GPT Image, Sora 2, Kling, WAN and Seedance. Start free — no credit card required.