zhenzhen-sonilo-video-to-music-fal

Every silent AI video eventually needs a soundtrack, and picking music that fits a clip's pacing is the kind of tedious job nobody wants to do by hand. Comfly_sonilo_video_to_music_fal is the zhenzhen pack's FAL wrapper for Sonilo's video-to-music model: feed it a video, get back generated music that matches the mood and rhythm. It lands as a real AUDIO tensor, so you can drop it straight onto your clip in the graph. The whole pitch is "the video is the prompt."

Mechanically it's the standard fal flow: the token goes in the node's api_key field, the video goes up through zhenzhen's proxy (https://ai.t8star.org/fal), the job gets submitted to fal-ai/sonilo/video-to-music, and the node polls until the audio is ready. Pay-per-call, prepaid-style credits.

The inputs that matter

• video_url (required) or video - the source. The required field is actually video_url (a public URL - when video is connected, the URL is ignored and the tensor is uploaded instead). So wire the video output of any generation node in, or paste a hosted URL.
• prompt - optional style/mood hint ("upbeat synth-pop, 120bpm"). Empty lets Sonilo infer everything from the video, which is usually the right call.
• num_samples - 1–3 variations, 1 default. More samples, more cost.
• start_offset - seconds into the video where scoring starts (0 = unset).
• duration - seconds of music, default 5.0 (a low-cost default); 0 = full remaining video. If you only need a sting, 5s keeps the bill down.
• video_way - upload (default, from the tensor) or video_url.

Outputs: audio (the tensor), audio_url, all_audio_urls (all samples if num_samples > 1), and response. Standard 6s × 600 polling defaults and the usual skip_error.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen

or ComfyUI Manager (search "Comfyui-zhenzhen"), restart. Paste a zhenzhen token (overseas https://ai.t8star.org) into api_key. No model downloads; audio handling is in the pack's normal dependency stack.

Common issues

• "requires a video input" - you left both video and video_url empty. Provide one.
• Music that ignores the pacing - Sonilo works best on video with clear cuts or beats; a static talking head gives it little to latch onto. Try giving it a prompt instead of relying on inference.
• Short clip, long silence at the end - check duration; if it's set to a fixed few seconds it won't score the whole video. Set 0 for full-remaining-video.

For the "just give me a soundtrack I don't have to think about" case, this node is the whole answer - and it slots into the same workflow as the pack's TTS nodes if you want voiceover and music on one clip.