Ready-made prompts
Background track for a video
Calm instrumental lo-fi: warm piano, a soft vintage drum loop, light vinyl crackle, 75 BPM, a cosy evening mood, no vocals.
Song with vocals
A warm pop ballad in English, female vocals, acoustic guitar and soft strings, a heartfelt chorus, 80 BPM.
Beat for Reels
An energetic electronic beat for short video: thick bass, bright synths, a tight groove, 124 BPM, no vocals.
Song lyrics
Write lyrics on my subject: two verses and a chorus, vivid images instead of generic words, an even line rhythm that is easy to sing.
Rock unplugged
An acoustic cover of a rock ballad: live guitar, light percussion, male vocals with a rasp, an intimate small-room sound, 70 BPM.
Examples created by our users
Why Moleculs.ai
A straightforward way to work with AI
One subscription, one payment
One balance for every model - no separate plan for each service
Works right in the browser
Nothing to install or configure - open a tab and start
Every AI model in one place
Text, images, video, and music - under one subscription
Clear, simple interface
Clear, native Russian across the entire product - no translation, no guesswork
Quick start
Sign up in a minute, no complicated setup - straight to work
No access restrictions
Nothing gets blocked - available anytime
What kinds of covers can you make with AI?
Want to hear a favorite song in a new way? AI re-sings it in a different voice or rearranges it in another style.
Voice Change
Re-sing the song in a male, female, or celebrity voice.
Genre Switch
Rock into jazz, pop into acoustic, rap into R&B.
Speed Up or Slow Down
Nightcore, slowed + reverb, and other remix styles.
Kids' Version
A favorite song in a soft, playful performance.
Acoustic Version
Live instruments instead of electronic sound.
Mashup
Combine the vocals from one song with another's arrangement.
How to create a cover
- Step 01
Upload the Original
Add the audio file of the song you want to rework.
- Step 02
Choose the Changes
A new voice, genre, tempo - specify what you want to change.
- Step 03
Cover Generation
The AI separates the vocals from the instruments and rebuilds the track to your specs.
- Step 04
Download the Result
Listen to the cover, adjust the balance, and save it.
What our users say
Real reviews from our Telegram community
Why choose Moleculs.ai
One access point for every model - without the limits of using just a single provider
 Moleculs.ai | ||
|---|---|---|
All the top models in one subscription | ||
Works right in the browser | ||
Professional roles and presets | ||
Free trial period | ||
24/7 support | ||
Projects and team collaboration | Coming soon | |
Support for multiple languages | Coming soon |
with Moleculs.ai, it's faster and more cost-effective
with Moleculs.ai you get dozens of models in one subscription, clear limits, and ready-made presets that save time. The tool helps you scale content, work as a team, and automate routine tasks without extra costs. All of it faster, simpler, and cheaper than piecing together separate tools.
Choose a plan for your needs
Answers to questions
Describe the genre, mood, tempo, and instruments - the AI generates a track in 30-60 seconds. You can set a style: pop, rock, electronic, jazz, hip-hop, classical, and more. The finished track is ready to listen to, download, and use in your own projects. Great for podcasts, videos, ads, and personal projects.
Try Moleculs.ai right now
Describe your task in one field - get a finished result in minutes. Moleculs.ai brings together the best AI models for text, images, video, and voiceovers in one interface. Start with a free generation, then subscribe to work without limits: write text, scale content, and switch between models without juggling extra tools.
What an AI cover actually does
Making an AI cover of a song is really two different operations people tend to lump together. The first: swap out the vocalist while keeping the melody, tempo, and arrangement. The second: rebuild the track in a different genre entirely - rock into acoustic, a pop ballad into electronic. The first keeps the original and changes the performer; the second changes the whole sound.
Which one you need decides what source file you need. For a voice swap, a track where the vocal is isolated - or at least clearly audible - works best. For a genre swap, almost any decent file is fine, since the model rebuilds the arrangement from scratch anyway.
Why the source matters more than the model
A phone recording from a concert crowd won't turn into a clean cover no matter how it's processed - the model inherits everything in the file, including room noise and distortion. A studio track or a good-quality file from a streaming service gives you something actually listenable. The gap between a good source and a bad one matters more than the gap between models.
How to brief the model
- Say what changes and what stays. "Keep the melody and tempo, swap the vocal for a female voice" - otherwise the model will rework both.
- Describe the voice by its qualities, not a name. "Low female voice, breathy, calm delivery" works; naming a specific artist is a bad idea on both quality and rights grounds.
- Name the whole genre, not just a mood. Not "make it acoustic" but "acoustic guitar, minimal percussion, tempo about ten percent slower."
- Handle tempo and key separately. Speeding up or slowing down changes a track's character more than you'd expect - ask for specific percentages.
- Expect more than one pass. The first result is almost never final - three or four runs is typical.
Voice swaps: where they usually fall apart
The weak spot is consonants and the transitions between words. On fast lines, the synthesis blurs word endings and the lyrics stop being intelligible. Slowing the track down a few percent, or splitting it into verses and processing them separately, fixes most of it. The other weak spot is high notes, where the voice starts to sound metallic. If the chorus climbs up high, it's worth dropping the key half a step to a full step, then shifting it back afterward.
Step by step: from file to finished cover
- Prep the source. Use the highest quality file you have. If it's noisy, clean up the noise first, before anything else.
- Separate the stems. Vocal on one track, instrumental on another. That way a voice swap doesn't touch the music, and a genre swap doesn't break the vocal line.
- Run a short clip first. One verse, not the whole song. Thirty seconds reveals everything that will go wrong over three minutes, and it takes a sixth of the time.
- Dial in the prompt on that clip. Voice, tempo, key, delivery. Usually two or three tries.
- Process the full track and mix it with the instrumental, matching the levels.
- Listen on headphones and on a phone speaker. Synthesis artifacts are obvious on headphones but disappear on a phone speaker - judge it by where people will actually listen.
How long it takes
The AI itself renders a cover in minutes, but getting it to a listenable state usually takes an hour to ninety minutes for the first track, and about half that for the next ones, once you've already worked out the right prompt.
Rights: a short but important paragraph
Generating an AI cover is technically simple - distributing it is a separate question. The melody and lyrics still belong to their authors, and a cover doesn't erase that. For personal listening, a gift, or a family video, that's a non-issue. For publishing on platforms, monetizing, or using it in advertising, you need rights to the underlying work.
One more note on voices: synthesis that imitates a recognizable performer is the area where complaints show up fastest. It's safer to describe the voice by its qualities than to name someone.
Other things people do with this tool
Beyond straight covers, the same process makes instrumental karaoke tracks, simplified kids' versions of songs with easier lyrics, mashups of two tracks, and translations of a song into another language while keeping the melody. That last one is especially tricky: another language's syllable structure rarely fits the original rhythm, so the lyrics have to be rewritten to fit the melody rather than translated word for word.
For a home recording, there's a simple trick: generate the AI cover in two or three different voices and have other people listen without telling them which one you like. Your own ear gets used to whichever version it hears first and stops noticing its flaws.
