Scoring AI Video — BGM and Sound Effects for the ComfyUI / Hailuo Era

How AI video gets made in 2026

It splits into two layers. One runs ComfyUI on a local GPU — Wan 2.2 (strong at big motion and camera work; the 5B build runs on modest VRAM), HunyuanVideo 1.5 for live-action, anime, and 3DCG, and LTX-2.5, which does synchronized audio and native multi-shot (one generation makes several connected cuts that hold characters and lighting). The other is the cloud/API layer, where MiniMax, Kling, Runway, Luma, Google's Veo, and ByteDance's Seedance generate cuts of a few seconds to tens of seconds from text or an image.

MiniMax's video model Hailuo 2.3 (October 2025) is strong at full-body motion, physics, and fine expression, and handles anime, ink-wash, and game-CG styles. MiniMax H3, released July 31, 2026, is a separate model that generates up to 15 seconds at 2K with stereo audio alongside the video, holding characters and style from reference images, video, and audio. A rough guide: Wan / HunyuanVideo to control style and characters locally, Hailuo 2.3 for speed and physics, and MiniMax H3, LTX-2.5, or Veo when you want the audio made in one pass too.

Silent models and audio-generating models coexist

Audio splits three ways. (1) Video only — Wan 2.2's standard generation, HunyuanVideo 1.5, Hailuo 2.3, Runway Gen-4.5, Luma, Pika. (2) Video driven by existing audio — Wan 2.2-S2V (it matches mouths and acting to a track; it doesn't make the sound itself). (3) Video and audio made together — MiniMax H3, LTX-2.5, Google Veo 3.1, Kling 3.0, ByteDance Seedance 2.5. Every tool in group (1) ships a separate 'Add Audio' step — itself the proof that video work still needs a post-production audio pass. OpenAI's Sora had audio, but its web app closed April 26, 2026, and the API is set to close September 24, 2026.

'Can generate audio' does not mean 'needs no BGM.' Veo 3.1 and MiniMax H3 can even generate music. You still lay a separate BGM — not because they can't make it, but because (1) each cut is only seconds to tens of seconds, and across stitched cuts the tone, level, and melody don't stay continuous, and (2) keeping dialogue, ambience, effects, and BGM as separate tracks lets you swap the BGM later or duck it under narration. That is exactly where an outside BGM like arcalune's earns its place.

The actual order for scoring a silent clip

In practice it usually goes in this order: generate cuts, pick the ones you'll use, lay one BGM across the whole thing, add ambience per scene (rain, waves, room noise), sync foley to the motion of people and objects (footsteps, cloth, impacts), drop a short effect on each screen transition, and duck the BGM only where there's narration or singing. Lay the BGM first as the foundation, and add effects last as points.

To export audio together in ComfyUI, expose an audio input on the Video Combine node from the add-on VideoHelperSuite and wire in a node that loads your music file (it's not a built-in node, so you install it first). You can also layer it later in a video editor. Either way, download the tracks here and use them as they are.

The patterns AI video on YouTube is built from

Faceless explainers and documentaries are built in the order script, synthetic voice, AI footage or images, captions, with a thin BGM under the voice. Cinematic shorts and trailers generate cuts from a storyboard and reference images, then build with a score, foley, and trailer-style impacts and risers (screenings at AI film festivals are the prime example). Shorts, memes, and montages that sell on tempo layer an up-tempo BGM and transition or impact effects onto a 5-to-15-second bit.

AI music videos fix the finished song first, generate many short clips per bar, and edit to the beat — they skip the model's own audio and keep the original track as the main audio. Product ads and brand films generate cuts from product images and logos as references, then add narration, a logo sound, and a short BGM afterward. AI ASMR generates footage and sound of cutting, crushing, or pouring materials that don't exist, swaps the off-sounding parts for real material sound, and re-edits into something long. Atmosphere and work-along videos loop or extend short clips and lay a long BGM plus ambience separately. The sections below line up which arcalune sound fits each pattern.

Cinematic shorts, AI films, and trailers

Generate cuts from a storyboard and reference images, then edit them into one piece. A continuous score, layered with foley and trailer-style impacts and risers.

1. 1

   壮大でドラマティックな劇伴

   Grand, dramatic cues to run under shots meant to feel like a film or trailer. Even a short built from a few-second cuts holds together as a piece when one score runs through it.

   Use it forAI films & shorts, Trailers, Openings

   These sounds from arcalune

   • Break the Gates3:02
   • The Road We Chose4:09
   • And That Was the Whole of It4:05
2. 2

   緊張と暗さの劇伴

   Darker cues that build unease and tension. For suspense scenes in an AI film, or the held beat just before a cut change in a trailer.

   Use it forSuspense, Trailer build-ups, Tense scenes

   These sounds from arcalune

   • Nobody Saw Them Coming3:12
   • The King Does Not Look Up3:29
   • Hold Until Dark3:12
3. 3

   予告編の一撃と上昇音

   A single impact to punch a trailer, and a rising swell of build-up. Placed on a title reveal or a cut change, they make the highlight land.

   Use it forTitle reveals, Cut changes, Highlight accents

   These sounds from arcalune

   • Impact Boom 10:02
   • Fly Away 10:03
   • Level Up 10:02

Explainers, documentaries, and faceless videos

Built in the order script, then synthetic voice, then AI footage or images, then captions. The voice leads, so you lay a thin BGM that doesn't cover it.

1. 4

   ナレーションを邪魔しないBGM

   Sparse, settled piano-led tracks. For explainer and documentary videos that carry a synthetic-voice narration (faceless videos), they sit underneath without swallowing the voice.

   Use it forExplainers & docs, Faceless videos, Under narration

   These sounds from arcalune

   • White Breath2:28
   • One Night Indoors4:18
   • Years Later, Briefly3:28

AI music videos and visualizers

Fix the finished song first, generate many short clips per bar, and edit them to the beat. The song is the main audio, so what you add is transition sound effects.

1. 5

   曲に合わせる転換の効果音

   An AI music video already has the finished song as its main audio, so what you add is transition sound effects, not more BGM. Clean whooshes and a single hit to land on beats or lyric breaks.

   Use it forBeats & lyric breaks, Visual transitions, Visualizers

   These sounds from arcalune

   • Whoosh 10:03
   • Fly Away 20:02
   • Impact Boom 20:02

Atmosphere, work-along, and loop backdrops

Loop or extend a short AI clip into something long. Combine a track that holds up over long play with environmental sound like rain or waves.

1. 6

   長時間流すアンビエント

   Longer ambient pieces with soft edges. For work-along and atmosphere videos that loop or extend a few-second AI clip into something long, they hold up under long play without tiring.

   Use it forLong work-along, Atmosphere, Loop backdrops

   These sounds from arcalune

   • The Trees Keep Count3:28
   • Nothing Holds Weight Here4:48
   • Whatever Is Listening4:28
2. 7

   雨と波の環境音

   Rain and wave ambience. Layered over a silent loop, they add the sense of being there that the picture alone can't carry.

   Use it forRain & wave scenes, Loop footage, Grounding ambience

   These sounds from arcalune

   • Rain 10:11
   • Ocean Wave 10:07
   • Rain 20:06

Shorts, memes, and tempo videos

A 5-to-15-second bit or a stitched montage. It lands on an up-tempo track paired with a sound effect that snaps the transitions and punchlines.

1. 8

   テンポの良い短尺BGM

   Bright, forward-moving tracks with a light step. They ride easily under Shorts that sell on tempo — a 5-to-15-second bit or a stitched montage.

   Use it forShorts & verticals, Montages, Tempo cuts

   These sounds from arcalune

   • Days of Youth2:05
   • The Big Chase2:37
   • Do Not Let Them Regroup3:21
2. 9

   切り替えと衝撃の効果音

   A flash for a cut change, a single impact, a clean whoosh. Dropped on a transition or a punchline in a tempo video or meme, they tighten the rhythm of watching.

   Use it forTransitions, Punchlines & memes, Impact accents

   These sounds from arcalune

   • FLASH CUT 010:02
   • Impact Boom 30:02
   • Whoosh 30:03

AI ASMR and texture videos

Generate footage and sound of cutting, crushing, or pouring materials that don't exist, then swap the off-sounding parts for real material sound. Here the sound effects lead.

1. 10

   質感・フォーリーの効果音

   Scissors cutting, running water, glass breaking. In AI ASMR and texture videos, use them to replace the spots where a generated clip's own sound feels off with a clear, defined material sound. A slice of the foley arcalune carries.

   Use it forAI ASMR, Texture accents, Replacing material sound

   These sounds from arcalune

   • SCISSORS 010:10
   • FAUCET 010:11
   • Glass Break 10:02