Changelog

Music for the space you run

Three new entry points put the Music Agent to work on commercial background music. Describe your venue — who spends time there, the atmosphere you want, when the music plays — and the Agent proposes two distinct creative directions, then starts generating a first track immediately. One field is enough to begin, and all three pages are available in all nine languages.

• AI Background Music Planner → — plan scene-specific music around how a space actually operates
• Custom Music for Business → — original music shaped by your brand, customers, and operating hours
• Original Music for Shopping Malls → — build a coherent in-mall sound without commissioning every track one by one

The player feels the beat

While a track plays, the player now casts a soft ambient glow that pulses in time with the music. It follows the actual beat of each track rather than a fixed rhythm, so what you see matches what you hear.

Organize your lyrics history

Your lyrics projects now work the same way as your song projects. Star any project in the history panel to mark it as a favorite — the Starred tab collects them so the drafts you care about are always one click away. You can also delete projects you no longer need to keep the list tidy.

Open the lyrics generator →

Smaller things

• Fixed a broken Discord invite link on the Portuguese contact page.

Clear endings for every generation

Most song generations move cleanly from pending to ready. In rarer cases where a request fails before generation fully starts, its final status and notification can now arrive immediately instead of waiting for a later check.

This keeps the project view, notification, and final task state aligned across both successful and failed attempts.

Create a song →

Results stay responsive

Remote notification delivery remains separate from the generation result itself. A slow or unavailable notification service will not hold up the status shown in Lacuna or interfere with the rest of the music generation flow.

More control over AI data sharing

You can now withdraw AI data-sharing consent from Account > Data & AI in the Lacuna iOS app. New AI creation stays disabled until you choose to consent again, while your existing songs, lyrics, and chats remain available.

Clearer privacy and purchase terms

Our Privacy Policy and Terms of Use now explain data processing, retention, deletion, sign-in methods, and the differences between web and App Store purchases more clearly.

An idea, and a song

A song rarely starts with a lyric line. It starts with wanting a certain kind of song. That is what Text to Song takes: one sentence about what the song is about, how it should feel, and where it will be used. No prompt syntax and no music terminology — add tempo, era, or instruments only when those constraints actually matter.

Who writes the words is your call. Have the Agent develop them, add your own, or skip vocals entirely and keep the result instrumental.

Start from a written idea →

Lyrics, and a song

When the words are already written, what you need is the opposite: don't touch my words. Lyrics to Song keeps them in a field of their own and turns everything else into choices — six sound directions (pop anthem, acoustic ballad, R&B slow jam, indie rock, hip-hop, cinematic), a vocal direction, and a line of extra style notes.

Which model to pick depends on what the first useful version has to preserve. For exact wording, names, and narrative order — and for the widest coverage of sung languages — choose Reverie. To explore production directions and edit sections afterwards, choose Aether. For one structured three-minute draft to judge the overall form and pacing, choose Echo.

Start from lyrics →

What actually gets sung

Both pages share one rule, and it is the easiest thing to get wrong: only what sits in the lyrics field gets sung.

• Genre, instruments, tempo feel, era, vocal character — those belong in Sound direction. Put them in the lyrics field and they get sung as words.
• [Verse], [Pre-Chorus], [Chorus], [Bridge] and [Outro] guide the arrangement. They do not become lyric lines.
• Paste the chorus every time you want it to come back. "Repeat chorus" cannot express the order you have in mind.
• When two versions of a line are competing, generate them separately. Comparing separate takes is cleaner than putting alternatives inside one lyric block.

Sign in with Apple

You can now sign in to Lacuna with your Apple account. The updated sign-in experience also makes it easier to see how you can start creating for free, receive credits every month, and keep your work saved.

Sign in with Apple →

Clearer sheet music to MIDI workspaces

We refreshed the Sheet Music to MIDI and PDF to MIDI experience. Uploads, saved scores, playback previews, and MIDI editing are now easier to find, with more complete guidance for photos, scans, screenshots, and multi-page PDFs.

Convert sheet music to MIDI →

Convert PDF to MIDI →

New album cover size guide

Our new guide brings together the cover-art requirements for Spotify, Apple Music, major distributors, CDs, and vinyl, with practical explanations of resolution, DPI, bleed, and RGB versus CMYK.

Read the album cover size guide →

Point an assistant at Lacuna, no install required

Assistants that speak MCP can now connect to Lacuna over the network. There is nothing to install and no local process to keep alive: give your assistant the endpoint at https://lacuna.fm/mcp, authenticate once, and it can list the models, start a song and check on it from inside your chat.

Agent-to-agent setups are covered too — there is an A2A endpoint answering both dialects currently in circulation. Both are published where agents look for them, so an assistant pointed at lacuna.fm can work out what is on offer without being handed the URLs.

The tools available inside your browser went site-wide as well. They used to be registered only on the song creation form; now an agent working in your browser sees them on any page, including a new one that lists the models and what each is good at.

Read the MCP guide →

Every job tells you when it's finished

Song generation has notified you for a while. Everything else left you watching the tab. Stem separation, mastering, artwork generation and mixdowns now send the same notification when they land — including when they fail, which is the case where silence costs you the most time.

Chat that doesn't repeat itself

The chat leans less on stock phrasing when it writes lyrics for you, and the parallel directions it offers now have to be meaningfully different from one another rather than variations on a single idea.

Unread markers on your projects

A project you left running is marked unread in Recent Projects and in the history panel, so coming back after a while shows you what moved rather than what you already heard.

Smaller things

• Volume Booster joined the browser extensions in the footer, and the store links under each extension now line up.
• Wording and layout across the lyrics and LRC pages were tightened.

One song's vocal. Another song's beat.

Most mashup tools play two songs at once and leave the hard part to you. The hard part is that two tracks almost never share a tempo, a key, or a downbeat — and if any of the three is off, the result sounds broken no matter how good the source material is.

Pick two tracks and the new Mashup Maker separates the vocal from one and the instrumental from the other, then lines them up. Tempo is read at the right octave, so 70 BPM and 140 BPM are recognised as the same groove counted differently rather than a 100% mismatch. Bar lines are found before anything moves, because landing a vocal on beat three collapses the whole thing. Relative keys are understood — A minor and C major use the same notes, and treating them as three semitones apart would reject a pair that works.

When two tracks genuinely don't fit — too far apart in tempo or key, or no steady beat to lock onto — you get told so, and the credits go back. That is the deliberate difference: a refusal you can act on beats a render you have to listen to twice before realising it was never going to work.

The result plays against either original so you can hear what changed. Vocal and backing levels are yours to move, and the timing nudges by a fraction of a beat. Those re-renders are free — the separated parts are kept, so adjusting doesn't cost again.

A mashup is 60 credits, which covers both separations plus the analysis and the render. Bring a track you already split on the Stem Splitter and that side comes off the price.

Make a mashup →

The song generator has an API now

Everything the generator does from the page can now be done from your own code: send a prompt, get back a task, poll it, receive finished tracks. It is a plain REST API with a TypeScript SDK on top, and it works the same from a script, a backend, or a coding agent.

The quickstart lives at /docs, the full reference at /docs/api, and there is a separate guide for wiring it into an MCP-capable assistant at /docs/mcp. The overview page at /ai-music-api walks through what it's for before you read any of them.

The key management page was rebuilt alongside it. You can hold up to 20 active keys, so rotating one no longer means taking the old one out of service first, and the reveal flow now shows a new secret once, clearly, instead of leaving you unsure whether you copied it.

Smaller things

• Autotune's mixdown was reworked — the tuned vocal sits better against the original instrumental, and key detection is more reliable on dense mixes.
• Credit log entries that reset a balance rather than add to it now read as "Reset to", so a monthly refill no longer looks like a top-up you don't remember buying.
• The plan card now tells you the date your access ends, not just which plan you're on.

Loop a track without hearing the seam

The hard part of looping music is not repeating it — it is finding the one point where the repeat does not click, thud, or jump. Drop a track into the new Audio Looper and it reads the tempo, scores every bar-aligned section, and hands you a ranked list of places where the music lines up across the splice. Play each candidate and listen to the moment it wraps — that is the only test that matters. If it is not quite right, nudge either boundary a beat at a time until it is.

Then set the length you actually need — up to an hour — and export one seamless file, with the track's own intro and outro kept around the repeated middle so it still starts and ends like a song.

The whole thing runs in your browser. MP3, WAV, FLAC, M4A or OGG up to 50 MB, decoded on your device — nothing is uploaded, and it costs nothing.

Loop a track →

Separated vocals go straight into the lyric tools

Split a song on the Stem Splitter and the vocal no longer ends at a download button. It now carries directly into the LRC Generator for a synced lyric file, or into the Lyric Video Maker to put the words on screen — the same way karaoke and instrumental results already carry into mastering. Nothing is uploaded twice.

Split a song →

Smaller things

• Three pages have clearer addresses: the song generator now lives at /ai-song-generator, the lyrics generator at /ai-lyrics-generator, and the stem splitter at /stem-splitter. Every old link redirects, so nothing you have bookmarked or shared breaks.
• The pricing page's plan cards got a visual pass.

Autotune, for the song you actually have

Nearly every online tuner asks for an isolated acapella. Most people only have the finished track. Drop that in and Lacuna separates the vocal, tunes it, and mixes it back over the original instrumental. If you do have a dry vocal, choose that instead — it skips the separation, so it is faster and costs less. You can also record straight into the page, or pick something you already made on Lacuna.

Natural is the default: audibly in tune, still recognisably you. One click gets the hard T-Pain sound. Everything underneath stays open — strength, retune speed, humanize, tolerance, vibrato, formant and a safety limit — and each control comes with a plain description of what it does to a voice, plus three starting points for natural correction, melodic trap, and stacked doubles. If you do not know the key of your song, it is detected for you.

When the render lands you can A/B against the original with no gap and no jump in playback position, read the pitch curve to see which notes moved and by how much, and compare the two spectrograms. Export WAV 16, WAV 24 or MP3 320. No watermark on any of them.

Uploading, recording, changing every setting and reading the results are free. A dry vocal is 15 credits; a full song is 30 — that is the separation plus the tuning, each at the rate you would pay for it on its own.

Tune a track →

Karaoke and instrumental now have their own pages

Same separation underneath, two different jobs, so they are no longer one page with a compromise in the middle.

Karaoke Maker hands you the backing track with the lead vocal lifted out, and defaults to MP3 because that file is going straight to a phone, a karaoke player or a PA. When it is done it offers to put the words on screen — as a lyric video, or as a synced LRC file.

Instrumental Maker treats the result as a half-finished thing rather than a finished one, so it defaults to lossless WAV for a DAW or a video timeline, and its next step is mastering. The file carries across on its own; you do not upload anything twice.

Make a karaoke track → · Make an instrumental →

Your results sit next to the form now

The rap, country, birthday, diss track, 8-bit and jingle generators are now a two-panel workspace: the form on the left, everything you have made on the right. Play, download, star, share or delete from the panel without leaving the page. Drag the divider to whatever split suits you and it stays that way next time. On a phone the two panels become tabs. If you have not made anything yet, the panel plays examples instead of showing you an empty box.

A clearer set of models

The model list is now three: Aether, Echo and Reverie. Each card says how many languages it sings, with the full list one hover away — that is the difference that actually decides which one you want. Aether is the default: the broadest range of styles, two songs per generation, and the only one that opens up the advanced controls, where you can push how strange the result gets, how hard the style is applied, and name the things you want kept out. Echo writes structured full tracks up to three minutes. Reverie sings your lyrics exactly as written and reaches the widest set of languages, and it needs lyrics from you for anything with vocals.

Bring a file into the conversation

Start a song from a reference. Attach an audio file and Lacuna listens to it for style; attach a MIDI file and it reads the melody, along with the lyrics if the file carries them. Images and PDFs go into the conversation as they always did, for artwork and briefs.

Smaller things

• Stem results now end with a sensible next step rather than a download list, and the file travels to it for you.
• Preferences you keep repeating across sessions can now be recognised on their own and saved to your profile, where they stay editable, deletable, and switchable off entirely.
• The lyrics editor, the home input and the player bar all got a pass for spacing, focus and scrubbing.

Detection without the round trip

When you are working through a folder of tracks, the slow part is never the analysis — it is the uploading. The extension takes that step out. Drop a file into the popup and it is read on the spot: tempo, key, scale, and the Camelot code, which is the one you actually want if you are matching tracks for a set. Nothing is sent to a server.

It is free and there is no account to make. The detection is the same engine that runs on the BPM & Key Finder page on the site — reach for the page when you want several files side by side, or want to carry a result straight into another tool; reach for the extension when you want one number and you want it without leaving the tab you are in.

Get it on the Chrome Web Store →

This is the first of several. The next ones are already being built, and they will show up in the Extensions column in the footer as they land.

Radio

Sixteen stations that run continuously in the browser, free, without an account and without an app. Rain, ocean, forest, stream, wind, fire and crickets from field recordings. White, pink and brown noise, a room fan and an air conditioner for masking. A quiet café and a night train for rooms that feel lived-in. Two of the stations are not recordings at all but original instrumental music written for the job: low-motion keys and texture for deep work, and slow drum-free ambient for falling asleep.

Each station is built from long passages rather than a short loop, and every passage plays once before the station reshuffles, so nothing comes back around while you are listening. Playback follows you across the rest of the site and keeps going in a background tab. Search by sound or by mood if you already know what you are after.

Browse the stations →

MP3 to MIDI

Upload a song, a hummed melody or a single instrument recording and get back a real MIDI file. Lacuna detects the pitch, start and length of every note and writes them as MIDI events, so the result opens in any DAW, sampler or soundfont with every note editable — quantize it, transpose it, swap the instrument.

MP3, WAV, M4A, OGG and FLAC up to 30 MB and 10 minutes, 25 credits per conversion. The result plays back on an interactive piano roll before you download the .mid. Single instruments and lead melodies transcribe best; for a dense mix, separate the stems first and convert the part you want.

Try it →

A vocal remover of its own

Removing the vocals now has its own page, built around the one thing you came for: upload a track, get a clean instrumental to sing over and the isolated vocal on its own, download either as MP3 or lossless WAV. The stem splitter page has been rebuilt alongside it for the times you need the full breakdown — vocals, drums, bass, guitar and piano.

Remove vocals → · Split into stems →

Custom mixes, rendered back into a song

The stem mixer now has pan alongside volume, solo and mute, so you can place a part in the stereo field instead of only setting its level. When the balance is right, Generate custom mix renders it into a complete new track and adds it to the work — no charge, and no DAW.

Dropping a single part is quicker still. From a track you can ask for a version without the drums, or without the vocals, and get a finished track back. If that song has already been split into stems, the new version is free.

The workspace tells you what it will cost

Ask the music workspace to remix, continue, edit one section, split stems, redo the cover or master a track, and it answers with a card that names the source track, the price in credits and roughly how long the wait is, before anything starts.

• Batches. Ask for a set — the same style across several different songs, or several versions of one song — and you get a plan first: how many distinct songs, how many audio tracks that produces, what stays locked and what changes, and the total cost, all before you confirm.
• Straight answers about limits. Requests the models cannot deliver — seven independent lead vocalists, a single song over eight minutes, lyrics past 5,000 characters — are stopped before generation starts, and no generation credits are used. Requests that are best-effort, like a male/female duet or excluding a particular instrument, say so up front rather than after the fact.
• A clean slate when you move on. When a conversation that is full of one song starts drifting toward a new one, the workspace offers to carry on in a fresh context so the old material stops pulling on the new work.
• Memory that you control. Stable preferences — the language you write in, how you want your own lyrics handled — can be remembered across works, and every save shows a receipt you can undo. All of it is listed on your profile, editable, deletable, and switchable off entirely.
• A history you can keep tidy. Rename or delete projects from the history panel; deleting one also takes down its public share page.

Smaller things

• Lyric theme and forbidden words are now their own fields — name the premise the lyrics should stay on, list words that must not appear, and generation stops before credits are charged if any survive.
• Sound exclusions let you name instruments or arrangement elements to keep out of the track.
• The lyric video maker can generate a square cover for the video on the spot, or take an image you upload and crop it to fit. Either way the result is saved to your cover assets.
• MP3 to MIDI, PDF to MIDI and Radio joined the tool navigation.

Timed lyrics from any song

Upload a song and get its lyrics back with a timestamp on every word. If you already have the lyrics, paste them and they get lined up against the audio. If you do not, leave it empty and the song is transcribed first, then lined up. MP3, WAV, FLAC, M4A and OGG, up to 50 MB and 10 minutes, 50 credits per song.

Automatic timing gets close, rarely perfect, so the result opens in an editor. Every line carries a timestamp: click it to jump there, hover it to play from there. The current word fills in as it is sung, so drift is something you see rather than something you hunt for. Nudge a line a tenth of a second either way or snap it to the playhead, shift a few selected words 50 milliseconds as a group, split or merge lines, or retype the words and let the timings redistribute. Undo, redo and Reset to auto are all there, and edits save as you go.

The lyric can also be translated into one of ten languages for 40 credits — English, Chinese, Japanese, Korean, Spanish, Portuguese, French, German, Italian, Russian. Translations land on the same timestamps and ride along in the exports: LRC for players, word-by-word LRC for karaoke apps, SRT or VTT for video editors, plain text with no timing at all, and a bilingual LRC once a translation exists.

Try it →

Lyric videos

Pick a track from your library that has timed lyrics, or upload a song and time it on the spot for 50 credits. Choose a look — Clean, Cinematic, Storm, Disco or Dreamy — then take it apart: rain, snow or drifting smoke, lightning, disco beams, background blur and dimming, a slow zoom or pan on the cover, film grain. Lines already sung can dim, recede, stay lit, or fill in karaoke-style, and where the song has a clear beat the flashes and the cover pulse land on it. The preview is the real thing, frame for frame, and free.

Export in 9:16 for TikTok and Reels, 1:1 for feeds, or 16:9 for YouTube, at 720p or 1080p, with the lyrics kept clear of each platform's own buttons. Exports run in the background: submit it, close the tab, and you get a notification when it is ready. It costs 15 credits per minute at 720p and 30 at 1080p, rounded up to the next minute, and a failed export refunds itself. 1080p needs Pro, and removing the Lacuna badge needs Basic or above.

Try it →

Three new song generators

Each asks for what its genre needs instead of making you describe it in a prompt.

8-Bit Music Generator. Pick the scene you are scoring — boss battle, overworld, title screen, victory fanfare, dungeon crawl or speed run — and the console it should sound like: NES, Game Boy, 16-bit or arcade FM. There is a real tempo slider, and a loop-friendly switch that drops the intro and outro so a track can run under a level forever. Try it →

AI Diss Track Generator. Name who is getting it, say what they are to you, and hand over the material — bad habits, old failures, in-jokes. Heat runs playful, spicy or savage, there is a keep-it-clean switch, and six beats to pick from. Try it →

AI Jingle Generator. Give it a brand or show name and a tagline, and the tagline gets sung word for word as the closing line of the hook. Say what the spot needs to sell and where it plays, then pick how much of a spot you need: a sting, a classic spot, or a full jingle. Try it →

AI Mastering

Your mixes now have a mastering step. Point it at a reference — a song whose sound you are after — and it matches that track's loudness, tonal balance, and dynamics. If you do not have a reference in mind, start from a preset instead.

Every master comes back with the numbers that decide whether it is ready: integrated loudness, true peak, and dynamic range. You can check a track against streaming targets before you send it anywhere.

Try it →

Hear the difference, then commit

Mastering is a judgement call, so you can listen to versions back to back. A/B switching is gapless — the playhead holds its position and the audio crosses over without a dropout, which is the only way the difference is actually audible. Masters sit in your library next to the original and play through the same player as everything else.

Jobs that finish while you are gone

Mastering and generation take real time, so they no longer need you watching. The tab icon reflects progress when you are on another tab, and if you leave the site entirely, you still get a notification when the track is ready.

AI Birthday Song Generator

A birthday song with the right name, the actual details, and the in-jokes.

Try it →

This release leans into control and craft — a page built just for rap, two free tools you can use without an account, rhyme you can actually see while you write, and more you can do without leaving your project.

A generator built for rap

There's a new page dedicated to rap. Type an idea and get back a finished track — original bars, rap vocals with real flow, and a beat — in about a minute. Pick a direction like boom bap, trap, drill, West Coast, melodic, or old school, set the mood and tempo, choose your instruments, and either write your own bars or let the AI write them for you. Want just the beat? Make it instrumental. Everything you create is saved to a tracks panel where you can play, download, share, star, or open it in the full studio to keep editing.

Make a rap song →

Two free tools: name your song, tag your MP3

Two small tools, free and open to everyone — no account required.

The AI Song Name Generator turns a one-line description into a batch of title ideas, each with a note on the vibe it captures. Pick a genre and mood, ask for shorter, longer, or more poetic variations, pin your favorites, and jump straight into making the song once you've landed on the name.

The MP3 Tag Editor lets you fix up an MP3's details — title, artist, album, genre, year, and more — and swap or crop its cover art, then download the updated file. It all happens in your browser, so the file never leaves your computer.

Name a song → · Edit MP3 tags →

See your rhymes, in any language

The Lyrics Writer now shows your rhyme scheme as you type. It draws colored lines between the lines that rhyme, giving each rhyme group its own color so you can read your pattern at a glance, and it counts syllables per line to help you hold your meter. It detects the language you're writing in automatically — or you can lock it yourself — and it now understands eight languages, including Chinese, Japanese, and Korean, where before it only worked in English.

Open the Lyrics Writer →

Split any song into stems, on the canvas

Inside the studio you can now break a song into its parts without leaving your project. Choose how finely to split it — vocals and instrumental, a four-part split into vocals, drums, bass, and the rest, or a six-part split that also pulls out guitar and piano. A new card appears next to your song, and once it's ready you can play each part on its own.

Open the studio →

A studio chat that keeps up

The assistant in the studio got three upgrades. When you ask for something it can't do itself — like making a MIDI file or splitting audio into stems — it now hands you a card that takes you straight to the right tool. After it edits one of your tracks, it leaves a shortcut chip in the chat that reopens that exact edit later. And it now draws soft colored threads linking each finished track back to the recipe it came from, so you can trace how a song was built even in a long conversation.

Try it →

Smaller things

• Images across the site load faster, scaled to your screen and served in modern formats.
• A round of visual refinements to the full-screen player and the music workspace.

This release is about giving you more direct control over how a song gets made — a new hands-on way to create, a page built just for country, and a new voice to make it with.

Take full control with Custom mode

The studio now lets you choose how you work. Agent keeps the conversational flow where the assistant guides you step by step. The new Custom mode gives you a direct form instead: set the musical style, write or paste your own lyrics, name the track, choose vocals or instrumental, pick a model, and generate in one click. Your choice is remembered between visits, and inside a Custom project the form stays open in the workspace so you can adjust any field and regenerate — each new version lands right on the canvas.

Start creating →

A song generator built for country

There's a new page dedicated to country music. Describe a story, a place, or a feeling, pick a direction — heartbreak ballad, outlaw road anthem, small-town nostalgia, honky-tonk, or bluegrass — set the mood, tempo, and instruments like pedal steel, fiddle, and banjo, then generate a full song with lead vocals and a real band. Write your own lyrics with AI help, or keep it instrumental. Everything you make is saved to a My songs tab where you can play, download, share, or delete it. The page is live in all nine site languages.

Make a country song →

Meet Reverie

A new song model, Reverie, joins the model picker. It renders the exact lyrics you write in natural, expressive vocals, and returns two versions in a single generation so you have a choice to start from. The premium Nocturne model is also cheaper now, dropping from 200 to 180 credits.

Pick a model →

Smaller things

• Reuse lyrics you've saved — pick from your Lyrics Studio library right inside the song form, in both the music workspace and the country song generator.
• The model picker now shows how many songs each model produces and whether cover art is included.

This release is a broad refresh of how the product looks and feels — starting from the homepage, going deep on phones, and reaching into the chat where you make music.

A homepage that plays

The homepage has been rebuilt around actual music. The new hero puts an album wall and a player front and center, showcase songs are curated per language so you hear examples that fit how you'd actually use them, and a stems view lets you take featured songs apart layer by layer. New sections walk through what the studio does and who it's for.

A refreshed look, everywhere

The visual language has been upgraded across the site — type, color, spacing, and the small details in buttons, dialogs, and badges. Pages feel calmer and more consistent, and the same design carries through from the homepage to the workspace.

Built for your phone

The mobile experience got the most attention it's ever had. The player is now a compact dock that stays out of the way while you browse, navigation and the top header have been reworked for small screens, and the full-screen player is one tap away from anywhere.

Clearer answers in the music chat

The assistant you talk to while making music now answers more to the point — it's better at understanding what you're going for, and clearer about what it did and what to try next.

Pricing, spelled out

Plan cards have been reworked, and there's a new capability table that shows exactly what your credits translate to across every feature — songs, lyrics, MIDI, and more — so you can compare plans on what they produce, not just their numbers.

See plans →

Commercial licenses for your tracks

When you download a track you made, you can now download a commercial license document along with it — your name, the track, and the rights, in one PDF you can keep on file or hand to a client.

Smaller things

• Signing in is faster: Google users get a one-tap prompt instead of a full redirect.
• Tool hubs now share a tab bar, so switching between generators takes one tap.
• Public pages load noticeably faster.

Until now the workspace chat was where you talked to the AI, and the canvas was where the music showed up — so after asking for a song you had to look somewhere else to hear it. This release brings the results back into the conversation.

Hear your results where you asked for them

Every song you generate now appears as a card in the chat itself, placed in the order you created it rather than pushed to the bottom. While a song is still rendering, you can play an early preview of each take straight from its card; the moment it finishes, the card turns into the final, playable version in place — no jumping around, no hunting on the canvas. If a generation fails, the card says so right where you expected the song.

Start a new song →

Generate from a direction, right in the chat

When the AI lays out creative directions for you, each one now shows up in the chat as its own card — with its style, tempo, and vocal at a glance. Tap to generate a song from it on the spot, or open the matching card on the canvas to fine-tune it first. Directions carry the same number and color in the chat as they do on the canvas, so the two always line up.

A heads-up when results are ready

Kick off a few songs, scroll up to keep working, and the ones that finish out of view collect into a single reminder at the top of the chat: how many are ready, with a tap to preview each or jump to it. It clears itself as you catch up.

Smaller things

• Songs drafted in ChatGPT and sessions opened from someone else's link now display in your own language instead of defaulting to one.
• The scroll-to-bottom button in chat is easier to see against busy conversations.

Who is really making AI music

We read about 650,000 AI music generations from across the space to answer the question the "AI slop" headlines skip: who actually writes the music? On the creation side, people bring their own lyrics, mark up the structure section by section, cast a specific voice, and come back to do it again — one-click-and-leave is a small slice. It's our first full data study, in nine languages. Read it →

Smaller things

• The blog now has charts, callouts, and audio you can play right in the article.
• Polished the navigation bar, footer, and mobile menu.
• Tightened spacing and timeline details in the music chat.

Until now, a generated track was final — if the second verse bothered you, your only option was to regenerate the whole song. This release changes that.

Remix, extend, or rewrite a section

Every finished track in your workspace now carries three editing actions, each opening a new editing dock at the bottom of the canvas with the track's waveform front and center.

• Remix reinterprets the whole track in a new style while keeping its character. Instead of typing a style prompt, you pick from chips — we split out the original track's style tags and suggest three remix directions tailored to the song.
• Extend continues the song past its ending. Drag the start point on the waveform, and a ghost segment shows where the new section will go.
• Edit a section rewrites the lyrics of just one part. Drag to select the section on the waveform — everything outside stays untouched.

Edited results appear on the canvas as new tracks, linked back to their source with a colored ribbon so you can always trace where a version came from.

Open your workspace →

Lyrics that know where they are

When you select a section on the waveform, the lyrics editor shows the full original lyrics with your selection highlighted and editable — the surrounding lines stay dimmed for context, so you always see exactly which words will change. Move the selection and the editable part follows.

Behind this, every vocal track now gets time-aligned lyrics — so synced lyric scrolling in the fullscreen player works on all models, not just some.

An AI lyrics assistant, while you edit

While an editing dock is open, the workspace chat becomes a lyrics assistant for that edit. Ask it to rewrite a verse, polish the wording, or continue the lyrics — its edits stream straight into your draft, line by line. Three one-tap shortcuts (inspire, polish, continue) work on whatever you've selected.

Redo your cover art without leaving the workspace

A new cover dock lets you regenerate a track's album art in place — write a prompt or let AI suggest one based on the track's title and style, browse style ideas, generate up to four candidates, and apply the one you like. The new cover updates everywhere at once.

Your library, reorganized

The assets library now covers everything you've made across six tabs — songs, stems, lyrics, MIDI, sheet music, and covers — with search, starred filtering, and pagination. Songs can be grouped by project, and edited tracks carry a small chip showing how they were made.

Smaller things

• Every player control now has a proper tooltip, and volume moved into a compact popover with precise control.
• Like and dislike buttons stay visible in the player bar instead of disappearing after rating.
• The fullscreen player now slides up smoothly from the bottom bar.
• Clicking a waveform while paused cues the position instead of unexpectedly starting playback.
• Tracks generating on the canvas now appear where they'll end up, so nothing jumps around when they finish.

We spent this release sweating the details of the fullscreen player.

Lyrics that read better

When you open a track fullscreen, the lyrics follow along line by line, and now the top and bottom of the view fade out softly so your eye stays on the line that's playing.

Open the player →

A steadier audio visualizer

The visualizer behind a playing track is more resilient — if a track's audio can't be analyzed, it settles into a calm decorative pattern instead of getting stuck on a loading message.

Smaller things

• Refinements to the music workspace — the track panel, sidebar, and model picker.
• Each model now has its own accent color so you can tell them apart at a glance.

A credit history you can read at a glance

Your credit history used to paint nearly every entry red, which made an ordinary track generation look the same as an error. Now entries are colored by what they actually mean: top-ups and purchases stand out, everyday usage stays neutral, and only genuine failures show up in red. The amounts are quieter too — additions in green, everything else in plain text.

See your history →

Tidy away the examples once you know your way around

Every creation tool opens with a row of example showcases to give you a starting point. Once you've got the hang of it, you can dismiss them — click the close button in the corner, confirm once, and that page stays clean from then on. Each tool remembers its own choice.

Make music →

Smaller things

• Better screen-reader labels across the chat input and its buttons.

This release rebuilds how you make music in Lacuna — from a single back-and-forth into a workspace where you can explore several directions at once and shape each one on its own.

Explore several directions at once

Every project now supports multiple musical DNAs, so you can chase more than one idea without starting over. Pick references with the new context picker and let each direction evolve independently.

Open the workspace →

A canvas that remembers your layout

Your tracks live on a canvas you can arrange however you like, and your layout is saved between visits. Open any track for a richer detail panel that keeps everything in one place.

Real-time responses

The assistant now streams its replies as they are written, so you watch ideas take shape instead of waiting for the whole answer to land.

A smoother lyrics editor

Writing and refining lyrics is calmer now — your edits stay in sync as the assistant suggests changes, and syllable counts update as you type.

Write lyrics →

Browse your full history

Music and lyrics history is now paginated, so you can scroll back through everything you have made without the page slowing down.

Clearer plans, bigger batches

We redesigned the pricing page to make plans easier to compare, and you can now generate up to 16 songs in a single run — with larger batches on Pro and Ultra.

See plans →

Smaller things

• A new top navigation that is cleaner and easier to move around.
• Inline help hints next to less obvious labels, so you are never guessing what something does.
• Refreshed buttons, backgrounds, and polish across the landing pages.

A follow-up to the new workspace: now you can watch a track come together while it generates, instead of waiting for it to land.

Follow a track as it generates

Open a track while it's still generating and the detail panel keeps up in real time. It starts with a placeholder while your track is composed, switches to a playable streaming preview the moment one is ready, and seamlessly turns into the full track once it finishes — no clicking around to find the result.

Open the workspace →

The canvas stays with you

When you start a track, the canvas focuses on it so you can see progress where it's happening, then reframes to fit the finished result.

Know which model made each track

Every track now carries a small badge for the model behind it — Aether, Echo, or Nocturne — so you can tell at a glance how each one was made.

Smaller things

• Track styles now read as individual tags instead of one long line.
• When the assistant lays out your first DNA, it opens automatically so you can dive straight in.

Generation updates feel live again

Progress and "ready to play" state now reach the workspace immediately when a generation finishes, even after the server restarts. If you noticed the canvas occasionally getting stuck on "rendering" when the audio was already ready, that is fixed.

Hear every take while the rest is still rendering

When a generation produces multiple takes, all of them now appear as live preview buttons on the loading card the moment they are ready — not just the first one. Pick whichever take you like and start listening while the others continue rendering.

Version history points at the right snapshot

Music DNA and lyrics history now highlight the snapshot you actually restored, and update correctly after you create a new one. Restoring a lyrics version also no longer creates a duplicate entry in history.

Three Lacuna GPTs on ChatGPT

Prefer to work inside ChatGPT? Three custom Lacuna GPTs are now linked from the site footer:

• Lyrics Generator — write songs with AI
• Music Generator — turn ideas into AI songs
• Song Creator — lyrics and music, from idea to studio

Smaller things

• Share, pricing, recipe, and a couple of showcase dialogs no longer overflow off-screen on phones.
• Chat replies in the music workspace no longer occasionally show raw JSON for suggestion chips or DNA fields inline.

Stem separation

A new tool that splits any track into its core parts — vocals, drums, bass, and other instruments. Upload a song, or pick one from your library, and Lacuna gives each stem back as its own audio file.

Try it →

A live mixer

Splitting is just the start. The stems page comes with a built-in mixer — mute, solo, or rebalance each part while the song plays. The fastest way to hear what a track sounds like without vocals, with drums solo'd, or with everything but the bass dropped out.

We've also seeded a showcase library across genres so you can try the whole flow without uploading anything of your own first.

A reworked global audio player

The player at the bottom of the app has been rebuilt:

• Persistent across pages — keep listening while you browse to a different hub
• A new full-screen view with bigger artwork and cleaner controls
• Track-type badges so you can tell at a glance whether you're playing a generation, a stem, or an upload

Fork from ChatGPT

If you use the Lacuna GPT inside ChatGPT to draft lyrics or songs, you can now fork that session straight into a Lacuna workspace — lyrics, prompt, and context preserved — and continue refining from there.

Smaller things

• All hub cards now show relative times ("3 minutes ago") instead of static timestamps
• Polished the showcases, segmented controls, and scroll fades across the app
• Tightened the music workspace layout and global player bar

MIDI to sheet music

Upload a MIDI file and get a clean, printable score back — not just notes on a grid, but real engraved sheet music. Keep a history of conversions, preview the output inline, and export as PDF or image when you're ready.

Try it →

A better assets library

Your assets library now has full-screen preview dialogs for sheet music and album covers, so you can inspect anything without downloading first. Cover generations can be starred for quick recall later, and the grid has been tightened up for easier browsing.

Open library →

Smaller things

• Polished the top navigation, mobile menu, and music sidebar
• Refined the album cover generator hub

Album cover generator

A new tool dedicated to artwork. Describe the vibe, drop in lyrics, or pick a style — Lacuna composes a cover that matches the song you're writing. Try it →

Lyric muse

Stuck on a verse? The new lyric muse turns half-formed lines into something you can sing. It lives right inside the music creation flow — no extra page, no friction.

A redesigned music hero

The home and music pages now share a single, focused entry point:

• Inline lyrics expansion — paste or write lyrics without leaving the box
• Vocals vs. Instrumental — explicit toggle, no guessing
• Model picker — pick your preferred model up front
• Attachment menu — image, audio, and reference files in one place

Task status on your tab

Running a long generation in the background? Your browser tab favicon now turns blue while a task is running, green when it succeeds, red on failure. Switch tabs freely — Lacuna will quietly tell you when something's ready.

Smaller things

• A new Tools page collecting everything Lacuna can do
• Mobile menu redesigned with smoother transitions
• A redesigned 404 page — try playing with it
• Contextual nudges to help you discover features as you explore