ANSWER · DJ EDITS FROM YOUR SONGS
The AI models behind Dropkit
“Which AI models does Dropkit use to make music?”
Dropkit uses Mureka V9.5 to render the music and vocals, Claude to write lyrics and suggest titles, Flux Pro 1.1 to generate cover art, and Demucs to separate stems for the vocal gate. All four run on the fal platform, so one infrastructure provider handles every job.
Music: Mureka V9.5
The song itself comes from Mureka's lyrics-to-song model. Dropkit sends it a style prompt, a lyric sheet, a target duration in seconds, and either a gender for the lead vocal or the ID of a cloned voice. The model returns a finished mix with a timing map of the lyric sections, which Dropkit uses to place the "vocal in" cue point.
Mureka is a generative model. It composes new audio rather than arranging pre-recorded loops, and it sings the lyrics you supply.
Lyrics and names: Claude
When you click "Write a draft for me" or ask for title suggestions, the request goes to Claude, Anthropic's language model. The prompt asks for club-friendly lyrics in your language with standard section tags, a strong repeated hook, under 1,200 characters, and no intro or outro sections, because those are instrumental. The draft lands in your lyrics box for editing before anything renders.
The same model writes the cover art concept from your title, genres and lyrics.
Cover art: Flux Pro 1.1
Covers come from Flux Pro 1.1, an image model. Dropkit asks for a square, high-contrast album cover with no text, letters or logos, so the title can be added later without clashing with baked-in typography.
Stem separation: Demucs
Before mastering, each take is split into vocals, drums, bass and other with the htdemucs model. Dropkit mutes the vocal stem during the intro and outro so the edges are truly instrumental, then sums the stems back into one file.
Mastering: not AI
Loudness normalisation, true-peak limiting, the mono bass crossover, silence trimming and MP3 and WAV encoding are deterministic audio processing performed with ffmpeg on a dedicated audio worker. The same settings produce the same result every time.
Why this matters to you
Knowing the models tells you what to expect. Vocal realism and genre range come from Mureka. Lyric quality comes from Claude and from how clearly you describe the song. Artwork style comes from Flux. The DJ-ready consistency, hard starts, mono bass, exact loudness and cue points, comes from Dropkit's own processing and does not vary between renders.
Frequently asked
Can I pick a different music model?
No. Every track is rendered with the same music model so results and mastering stay consistent.
Is the audio processing done by AI too?
The stem split uses the Demucs model. Loudness, mono bass, trimming and encoding are standard signal processing, not generative AI.