Recreate the sound of a song with AI by describing the recording, then generating from that description. What decides the result is where the drums sit, whether the vocal is dry or in a room, what job the bass has, and how the arrangement builds. Upload the reference to the AI Music Analyzer, keep the report lines that say those things, and press Use this prompt. That opens the AI Music Generator with the report's draft in the song description, ready to edit into a song to prompt.
The details that carry a song's feel
The detail that matters most is drum placement and the sound of the kit. A kit set back in a small room, soft beater, loose snare, is a different record from a dry close kick and a tight snare, even when both could be filed under the same decade. "Vintage" on its own is a useless word for this job. It names a period and leaves the microphone, the room, and the playing unnamed, so the prompt never chooses among them.
Vocal distance is the next choice that actually moves the result. A close, dry lead, breath still in the microphone and almost no room on the phrase, is a different picture from a voice down a hallway, and the word "intimate" covers both, so the prompt has to name the space. "Close, dry lead vocal, a little air on the consonants, no hall" puts the microphone somewhere, while "emotional vocal" only names a mood.
After the kit and the vocal, write the arrangement build, because that is what says when the other parts enter. If the reference stays as guitar and voice until the chorus, then adds a second guitar and a harmony, that order belongs in the prompt. A genre label leaves those entrances unspecified. Bass role belongs in the same breath: a bass locked to the kick, in a narrow range, holds the groove, while a bass that walks or answers the vocal becomes another lead. Tape and room are worth one concrete image, such as hiss, a short plate on the snare, or a small room on the drums. "Warm, nostalgic, cinematic" is the pile that means the observation stopped.
A common mistake is to keep every mood label and drop the production lines. Mood words are easy to pile up. The physical setup of the record is what deserves the space in the prompt.
For example: "Mid-tempo, drums set back in a small room, soft kick, loose snare with a short plate, close dry vocal, bass locked to the kick, arrangement starts with guitar and vocal and adds a second guitar and a harmony only on the chorus." The version that drifts is "vintage indie, emotional, cinematic, nostalgic." Same imagined song, and only the first one says where anything is.
Pick a reference and read its report
Work from one reference. Upload only a recording you have the right to process, and describe its production for a new track, leaving the melody and the lyric out of the prompt. Sign in, and upload one MP3, WAV, or M4A, up to 20 MB and 8 minutes. MP3 and WAV are the most reliable of the three. Analyze Music costs 2 credits, and those credits come back if the run fails. Packs and plans are on pricing.
The card is the page's Example report, a showcase sample, not an upload.
The useful part of the report is the specific phrasing in three fields: Rhythm & Groove, Arrangement, and Production. Here they give you a "four-on-the-floor pulse" and a "driving eighth-note bass pattern" for the groove, "adds drums and bass in stages" for the build, and "bright gated drums," "long reverb tails," and a "controlled low end" for the room. The example's own Generation Prompt falls back on vague words like "nocturnal nostalgic mood" and "polished retro production." Replace those with the phrases from the three fields before you send the prompt.
116 BPM and A minor at medium confidence are optional. Use them when you want the new track nearby, and drop them when the prompt is crowded. A finished report marks tempo and key high, medium, or low. Keep a high-confidence figure when you want the new track in the same ballpark, and leave a low-confidence figure out. If you want a separate BPM and Camelot reading of the same file, the Key & BPM Finder does that in the browser.
"No vocals detected" means keep the prompt instrumental.
Turn the report into a song to prompt
Use this prompt is on a finished analysis. It saves that Generation Prompt and opens the generator with the text in the song description. The example card has no Use this prompt button, so paste the text into Song Description yourself.
This edit keeps the instrumental, the pace, the arpeggios, the drums, the bass, and the guitar accents from that draft, and puts the rhythm, arrangement, and production lines where the mood stack was.
Create a cinematic synthwave instrumental at a steady mid-tempo pace with pulsing analog arpeggios, electronic drums, deep synth bass, and clean guitar accents. Keep a steady four-on-the-floor pulse, syncopated arpeggios, and a driving eighth-note bass pattern. Open with filtered pads, add drums and bass in stages, then expand into a wider final section before a short atmospheric outro. Bright gated drums, controlled low end, and long reverb tails.
If the prompt is too long for the field, trim it before you generate. Keep the instruments, the groove, and one production detail, and cut repeated mood words before anything else.
Listen for what the new track kept
Play the reference and the new track back to back once a take exists. The sample stops at the edited prompt, so the notes here are what to listen for on the take you make.
Judge where the kit sits and what it sounds like, whether any vocal is as dry or as wet as the reference, whether the arrangement builds the same way, and whether the bass still has the same job. On a track in the vein of Midnight Transit, that means the bright gated drums, the long reverb tails, drums and bass entering in stages, and the driving eighth-note bass. Matching the genre while missing the room means the sound has drifted. If the drums are right and a vocal has appeared, or the whole build arrives at once, change that clause and leave the kit sentence alone.
When the recording should be the source
This path writes a new track from a description of the sound. The reference file stays on the analyzer page, and the generator receives the text. When the job is to change a recording you already have, use a tool that starts from the file. The AI Music Remixer changes genre, mood, instrumentation, or arrangement while keeping the core melody recognizable. AI Song Cover keeps the song recognizable and changes the singing voice. Those are different jobs from a song to prompt.