What Track Reels is
Track Reels takes one of your songs and produces social-ready video clips. Each clip is built from atmospheric AI b-roll matched to your song's energy and lyrics, with burned-in captions, a song-credit chyron, and the AetherWave watermark (Artist plan and up can remove the watermark). It's the budget alternative to a full music video - fast, cheap, and social-native.
The default setup gives you four 30-second 9:16 vertical clips that drop straight into TikTok, Instagram Reels, or YouTube Shorts. The Full Song option produces one continuous landscape music video matching the full length of your track.
Editorial intelligence (v2)
Under the hood Track Reels runs a two-pass AI editor. The first pass picks the most lyrically and emotionally interesting moments of your song and writes b-roll prompts. The second pass takes the actual generated b-roll clips and writes the cut sheet - which clip plays when, how long each shot is held, where the cuts land relative to the song's beat grid.
What that means in practice:
- Beat-synced cuts. Every cut after the first lands on a downbeat or beat from your song's rhythmic grid (within ±50ms). No more cuts on round seconds.
- Track-aware editorial style. Track Reels reads your song's tempo, energy curve, and section variance and picks an appropriate editorial pacing automatically. A driving but soft 100 BPM track gets a narrative arc edit (long opening shot, escalating density, rapid climax). A hypnotic 130 BPM track gets a motif-driven edit (one anchor clip recurring across structural beats). A true ballad gets slow burn (held shots, sparse cuts on downbeats only).
- Multi-clip variety. 15-second reels generate 2 b-roll clips, 30s generates 3, 60s generates 5. The editor cuts both within each clip (different in-out windows) and between them (clip changes). You see real visible cuts, not the same shot playing through.
- Section-aware density. When your song has a verse-chorus-verse arc, the editor matches it - longer held shots in the verses, rapid cuts in the choruses.
- Multi-reel differentiation. When you ask for 4 reels, each one gets a different editorial style assigned (rapid cuts, slow burn, narrative arc, motif-driven). Same song, same b-roll pool, four meaningfully different edits to choose from.
Quick path
- Open Track Reels from the grid menu under Music or Video.
- Drop an audio file. Supported: mp3, wav, m4a, flac, ogg. Up to 50 MB.
- Fill in the panels - defaults work fine if you just want to ship something.
- Click Analyze track. Claude reads the lyrics + energy curve and proposes your moments.
- Click Render. Wait the indicated time. Reels auto-save to your Gallery on completion.
Panel 1 dials
Reels (1 - 4)
How many separate clips to produce. More reels = more b-roll = higher cost. Fewer reels = budget-friendly.
When Length = Full Song, this locks to 1 (you don't want four copies of the same full music video).
Length (15s / 30s / 60s / Full Song)
How long each reel runs:
- 15 seconds - fast TikTok hook style, lowest-cost option (55 cr per reel)
- 30 seconds - standard social clip (100 cr per reel) - the default
- 60 seconds - extended scene, room for full chorus + verse (165 cr per reel)
- Full Song - single reel matching the entire length of your audio file. Claude writes a complete music video storyboard scene-by-scene across the song. The "Golden Eagle" full-length music-video-from-one-prompt option.
When you pick Full Song, the Aspect dropdown automatically flips to 16:9 landscape (you can change it back). Full Song forces Reels count to 1.
Aspect (9:16 vertical / 16:9 landscape)
- 9:16 vertical - default. Best for TikTok, Instagram Reels, YouTube Shorts. The phone-native frame.
- 16:9 landscape - best for full music videos, YouTube uploads, and desktop viewing. Auto-selected when you pick Length=Full Song.
The aspect drives Grok b-roll generation, the final composite dimensions (1080x1920 vs 1920x1080), and the gallery preview tile sizing.
Story Mode
Tells Claude what structural approach to take. Imagery describes the content; Story Mode controls the shape.
- Auto - Claude decides based on the imagery and song. Default behavior.
- Single narrative - same characters threaded across every clip. Continuous story. Pair with Character Consistency for best results.
- Multi-vignette - different characters or couples per clip, common emotional theme. Anthology style - like a music video where you see several different people experiencing the same feeling.
- Single performer - one figure throughout, often interpreted as the artist performing the song. Same character in every clip, varied poses and settings.
- Atmospheric only - no people, no faces. Pure mood - landscapes, light, weather, abstract textures.
- Abstract motion - pure motion graphics. Geometric forms, fluid simulations, neon ribbons. No representational content.
Style preset (10 options)
Cinematic, Cyberpunk Neon, Dark Moody, Natural Landscape, Epic Space / Sci-Fi, Retro 80s VHS, Anime, Fantasy Ethereal, Urban Street, Abstract Motion. Default Cinematic. The aesthetic wrapper applied across all b-roll prompts.
Imagery (with chips)
Free-text description of the recurring visual subject. Leave blank for default abstract atmospheric b-roll. Fill it in for grounded, specific, character-driven b-roll.
Below the textarea: four tap-to-fill chips with proven prompts (Single character / Multi-vignette / Atmospheric / Epic narrative). Lower the empty-field paralysis if you don't know what to type.
Character consistency (conditional)
Only appears when Story Mode is Single Narrative or Single Performer. When on, Claude defines a specific physical descriptor at the top of its interpretation (hair color, clothing, build, distinctive features) and prepends that EXACT description to every b-roll prompt. Best lever we have for character continuity given current model constraints - image-to-video continuity tools aren't strong enough yet to enforce identical faces.
Song title + Artist
Shown on the chyron card lower-left of every reel. Fades in twice (start and end). Optional but recommended.
Visual references (up to 10 images)
Optional. Upload up to 10 reference images to steer the look of your b-roll. Where Claude tags a moment as a visual match, that reference seeds an image-to-video (I2V) clip so your b-roll carries the characters, setting, or aesthetic you uploaded instead of a pure text-to-video guess. Leave it empty for fully AI-imagined b-roll.
Add images by uploading from your device or pulling from your Gallery. Click any reference thumbnail to open it full screen in a lightbox slideshow - use the on-screen arrows or your keyboard left/right arrows to flip through the set, the counter to see your place, and Escape (or a click on the backdrop) to close. The remove control on each thumbnail stays separate, so opening the viewer never deletes an image.
Remove watermark (Artist plan and up)
By default every reel carries the AetherWave watermark lower-right. On the Artist plan or higher (Artist, Producer, Mogul, Ultimate) a "Remove AetherWave watermark" toggle becomes active in the render options - flip it on before rendering and the output ships with no watermark. The toggle is visible on every plan, but locked on Free, Starter, and Studio with an upgrade prompt. The check is enforced server-side, so the watermark is only ever removed for eligible plans.
Burn lyric captions
Toggle off if you want clean video. Default on. Captions are 2-word UPPERCASE chunks generated from Whisper word-level timestamps.
Cost in credits
The cost label updates live as you adjust dropdowns. Reels are flat-rate per length tier; Full Song scales with your audio file's length.
Per-reel pricing:
| Length | Cost per reel |
|---|---|
| 15s | 55 cr |
| 30s | 100 cr |
| 60s | 165 cr |
| Full Song | 4 cr per second of audio |
Common combinations:
| Setup | Cost |
|---|---|
| 1 reel x 15s | 55 cr |
| 4 reels x 15s | 220 cr |
| 1 reel x 30s | 100 cr |
| 4 reels x 30s (default) | 400 cr |
| 1 reel x 60s | 165 cr |
| 4 reels x 60s | 660 cr |
| Full Song (3 min track) | 720 cr |
| Full Song (4 min track) | 960 cr |
Credits are deducted at render kickoff. If the render fails for any reason - Grok generation timeout, ffmpeg crash, etc. - the full amount is automatically refunded. Whisper transcription and Claude analysis are not separately charged.
What you get back
For every reel, the rendered output includes:
- 9:16 or 16:9 atmospheric b-roll matched to your song's mood and your direction
- Burned lyric captions (toggle-able)
- Song-credit chyron lower-left, fading in at the start and again near the end (stays visible through the final frame)
- AetherWave watermark lower-right (on by default; removable on Artist plan and up via the render toggle)
- Audio normalized to -14 LUFS / -1 dBTP (the social loudness standard)
- Auto-saved to your Gallery as type=video, visibility=private, with full metadata (song, artist, style, imagery, beat type, timestamp range)
Each output tile in Panel 3 has a Download button (blob fetch, no page navigation) and a Save button (manual gallery re-save). Both turn green with a checkmark when complete.
The Process another track button below the output grid clears the file + song title + render state but keeps your artist / style / imagery / story-mode / aspect / lyric-toggle so you can batch multiple tracks with consistent branding.
Recommended recipes
Quick TikTok teaser (cheapest)
- Reels: 1, Length: 15s, Aspect: 9:16, Story: Auto
- Imagery: optional
- Cost: 55 credits
Standard 4-reel social drop (default)
- Reels: 4, Length: 30s, Aspect: 9:16, Style: pick one matching the genre
- Imagery: a few words describing what should appear
- Cost: 400 credits
Full character-driven music video (Golden Eagle)
- Reels: 1 (auto), Length: Full Song, Aspect: 16:9 (auto), Story: Single Narrative, Character Consistency: on
- Imagery: detailed scene description with characters
- Cost: ~720 cr for a 3-minute song
Atmospheric instrumental video
- Reels: 1, Length: Full Song, Aspect: 16:9, Story: Atmospheric
- Imagery: optional
- Cost: ~720 cr for a 3-minute track
Limits
- Minimum song length: about 2 minutes for 4-reel mode. Track Reels picks four 30-second windows and needs enough song to choose from.
- Maximum audio file size: 50 MB uploaded. Re-encoded to 64kbps mono MP3 for Whisper, fits songs up to ~50 minutes.
- Render time: ~10 minutes for 4 reels at 30s. ~18-22 minutes for a Full Song reel on a 3-minute track (more Grok clips). Don't push deploys during a render - Railway restarts kill the Python subprocess.
- Vocal detection: Whisper handles most vocals. Heavily effect-processed, foreign-language, or fully-instrumental tracks may produce inaccurate captions. The pipeline still works on instrumentals (skips captions automatically).
Common questions
Why doesn't my full-length video have consistent characters? Image-to-video models can't enforce character continuity across separate generation calls. The only lever we have is repeating physical descriptors verbatim in every prompt. Turn on Character Consistency (under Story Mode = Single Narrative or Single Performer) - that tells Claude to define and repeat exact physical details across every clip's prompt. Results are improved but not perfect.
Do my reels overlap in time across the song? Sometimes. Each reel is a standalone social post, not a sequence - so Claude is allowed to overlap windows when the strongest editorial moments cluster together (chorus + drop). This is intentional.
The chyron shows "Untitled" or "Unknown Artist." Fill in the Song title and Artist fields before rendering. They're shown in the chyron exactly as typed.
Can I edit Claude's moment picks before rendering? Not yet. v1 uses Claude's picks directly. Drag-handle moment refinement is on the roadmap.
How do I get a different render of the same song? Click Process another track, re-upload the same song, and tweak Story Mode or Imagery for a different interpretation.
How do I remove the AetherWave watermark? Upgrade to the Artist plan or higher (Artist, Producer, Mogul, Ultimate). The "Remove AetherWave watermark" toggle in the render options is locked on Free, Starter, and Studio, and active on Artist and up. Turn it on before you render. On lower plans the watermark stays in the lower-right corner.
Can I preview my visual reference images? Yes. Click any reference thumbnail to open a full-screen lightbox, then use the arrows or your keyboard to step through all your uploaded references. You can add up to 10.