Nodaro DocsVideo SFX
Generate sound effects, foley and ambience that follow the action in a video, from an optional prompt. The new sound replaces the video's own audio track.
The Video SFX node generates sound for a video: sound effects, foley and ambience that follow what happens on screen. It watches the clip, reads your optional prompt, and returns an MP4 with the generated sound. The node runs the MMAudio model.
The new sound replaces the audio track
Video SFX drops the video's original sound. To keep the original sound as well, wire the result and the original audio, for example from Extract Audio, into Merge Video & Audio.
When to use it
- Add foley or ambience to a silent generated video.
- Generate room tone, weather or other environmental sound for a short cinematic clip.
- Produce several alternate takes of the sound in one run, and keep the best one.
- Build a soundtrack in layers, with the effects from this node and the dialogue or music from other nodes.
To create a sound effect from a text description alone, without a video, use Text to Audio.
Quick start
Add the node
Press Tab on the canvas and choose Video › Sound for Video › Video SFX.
Connect the video
Wire the clip into the Video input. MP4, MOV and WebM clips of up to 300 seconds work.
Describe the sound
Type the sound in the prompt, for example rain on a metal roof. Leave the prompt blank for foley driven by the video alone.
Run it
Click Run on the node. The video with its new sound appears on the node.
Inputs
| Input | Accepts | What it does |
|---|---|---|
| Video | Video nodes, such as Generate Video and Upload Video | The clip that needs sound. Required. MP4 (H.264), MOV or WebM, up to 300 seconds. |
| Prompt | Text nodes, such as Text and Generate Text - Any LLM | The sound to generate. A wired prompt replaces the prompt typed in the settings panel. |
| Negative | Text nodes | Sounds to exclude. |
The output, Video, is an MP4 with the generated sound. The original sound is dropped.
Settings
| Setting | What it does |
|---|---|
| Prompt | The sound you want, for example footsteps on dry leaves, rain on a metal roof or engine revving. Leave it blank for foley driven by the video alone. |
| Negative prompt | Sounds to exclude. The default is music, which keeps the result free of music. Clear the field if you want music as the sound. |
| Versions | How many takes to generate in one run, from 1 to 4. Each version uses a different seed, and the cost grows with each version. A workflow run makes one version. |
| CFG strength | Under Advanced. How strictly the model follows the prompt, from 1 (loose) to 10 (strict). The default is 4.5. |
| Inference steps | Under Advanced. From 10 (fast) to 50 (clean). The default is 25. |
| Seed | Under Advanced. Leave it blank, or enter -1, for a random seed. The same seed and the same inputs give the same result. With several versions, each version adds its number to the seed, so every take is repeatable and different. |
| Pre & post text | Text that is always added before and after the prompt. It is hidden from people who use your workflow as an app. See Prompt pre and post text. |
Credits
The price depends on the length of the input video, rounded up to a bucket, and on the number of versions. Nodaro measures the video when the run starts, and the sound covers the whole clip. If the length cannot be read, the run is priced, and the sound made, for 8 seconds.
| Video length | Credits per version |
|---|---|
| Up to 8 seconds | 11 |
| Up to 15 seconds | 11 |
| Up to 30 seconds | 22 |
| Up to 60 seconds | 33 |
| Up to 120 seconds | 55 |
| Up to 300 seconds | 121 |
The total is the price per version times the number of versions. For example, a 12-second clip with 1 version costs 11 credits, a 31-second clip costs 33 credits, and a 60-second clip with 4 versions costs 132 credits.
A run from the node's Run button, a workflow run and a call from the API are all priced this way. In a workflow run, each Video SFX node makes one version, so it costs the price of one version.
Before a run, the Run button and the workflow estimate show the price for the length that the connected video node reports. On a video rendered by Apply EDL, they show the price for the render's estimated length, at most the 300-second price. This also applies through a node that keeps the render's length, such as the captioned episode of a Tighten Episode. A trailer cut by Edit Plan in Trailer mode is estimated at 40 seconds, the longest trailer the plan cuts, so it shows the 60-second price. A Trim Video, Loop Video, Combine Videos or Video SFX node in between passes on the length it makes. When no length is known yet, they show the 300-second price, 121 credits per version, which is the most a run can be charged. The run then charges the price for the length it measures.
The price that a published app or a template lists follows the same lengths. When the length can change from run to run, for example on a recording that the app's user replaces, the listed price is the 300-second price, 121 credits per version. A longer video is refused, so no run costs more than the listed price. On a trailer that Edit Plan cuts in Trailer mode, the listed price is the 60-second price, because a trailer is estimated at 40 seconds.
Limits
- Formats: MP4 (H.264), MOV and WebM.
- Length: up to 300 seconds. A longer video is refused with the error
video_duration_exceeds_limit. - Versions: up to 4 per run.
- Resolution: up to 1080p is recommended. Larger videos are untested and take longer to process.
Tips
- Describe the sound, not the scene.
crackling fireworks better thancampfire scene. - Test on short clips. A clip of up to 15 seconds costs the same as a clip of 8 seconds, so iterate on short clips.
- Layer the sound. For effects over music, generate the effects here, then mix in the music with Merge Video & Audio or Mix Audio.
- Fix the seed to compare prompts. With a fixed Seed, the only change between two runs is your prompt.
Frequently asked questions
Does Video SFX keep the original sound of my video?
No. The generated sound replaces the video's audio track. To keep both, wire the result and the original audio, for example from Extract Audio, into Merge Video & Audio.
Do I need to write a prompt?
No. Leave the prompt blank, and the model makes foley driven by the video alone. A prompt such as "rain on a metal roof" steers it toward a specific sound.
Can Video SFX generate music?
Only if you clear the Negative prompt, which is "music" by default. The default keeps the result free of music, so the node produces sound effects only.
How long can the video be?
Up to 300 seconds, which is 5 minutes. A longer video is refused before the run. The recommended resolution is up to 1080p.
How much does Video SFX cost?
The price depends on the length of the video and the number of versions. One version costs 11 credits for a clip of up to 15 seconds and 121 credits for a clip of up to 300 seconds.
Related
Merge Video & Audio
Text to Audio
Mix Audio
Extract Audio
Generate Video
Last updated on
Speaker Layout
Frame a whole podcast or interview recording by speaker in one node, with nothing cut. With several cameras, each speaker is shown on their own camera.
Merge Video & Audio
Add voiceover, music or sound effects to a video. Set each track's volume and start time, and keep, lower or mute the video's own sound.