v0.0.45

Windows: download Yeufonic-Setup-0.0.45.exe and run it over your existing install. Quit Yeufonic from its icon by the clock first (right-click, then Quit): closing the window leaves it running, and the installer waits for it to quit. Your library, settings, LoRAs and models are kept.

Docker: in the Yeufonic folder run git pull, then docker compose restart app and reload the page. No rebuild.

What is new in 0.0.45

• Hot keys help overlay. Pressing ? (or Shift + /) opens an overlay dialog displaying all application keyboard shortcuts grouped by section: main page playback and chosen take actions, the score editor and piano roll, and take comparison. The dialog closes on Escape, on clicking Close or the backdrop, or on pressing ? again. A Hot keys item with a keyboard icon is also in the top-left brand menu. Typing in text fields is left alone so question marks in lyrics, styles and titles are unaffected.
• Resilient fallback on external LLM and image model limits.
  • Take picture drawing: When set to The external LLM's image model, if an external draw request fails (HTTP 429 rate limits, quota/credit exhaustion, timeout), picture generation gracefully falls back to FLUX.2 [klein] on this PC if installed. If a song render is active on the GPU, drawing steps aside to the waiting queue so song renders are never interrupted.
  • Take picture description: When set to External LLM, if the description request fails, it seamlessly falls back to describing from the take itself (title, style tags and lyrics imagery) without failing the picture.
  • Lyric drafts from prompts: When LLM for writing lyrics & style tags is set to External LLM, if drafting fails or hits rate limits, Yeufonic falls back to drafting with local Gemma on the engine (if installed) and displays a clear warning in the status bar alerting you that external limits were hit.
  • Lyrics from recordings & style tags: If hearing lyrics with the external LLM fails, Yeufonic falls back to Whisper and flags the warning in the status bar. Corpus style tagging similarly falls back to local Gemma on the engine.
• Performance, deduplication and code maintainability.
  • Database transactions & query performance: wrapped startup initialization in atomic database transactions, optimized single-record lookups to .fetchone(), added thread-safe settings locking, and batched N+1 style resolution queries.
  • Canonical music theory & lyrics: centralized pitch, interval and ABC notation conversion in a shared app/music.py module, unified section tag synonyms in app/lyrics.py, and eliminated circular dependencies.
  • Audio & storage speedups: unified FLAC header parsing to read streaminfo directly across normalization, audio duration and tagging routines without spawning external ffprobe processes; cached LoRA metadata by file modification time.
  • Engine & frontend rendering: precalculated harmony bias penalties, cached NVML library handles to avoid reopening the shared library, recycled piano roll key DOM elements to eliminate layout rebuilds, paused mastering rack meter animations when closed, and throttled waveform playback updates.

UI enhancements

• The editor can fill the window. A button in its top right corner, after Close, makes the editor fill the whole browser window, and a second click puts it back to its usual size. The editor opens the way you last left it. Filling the window, the words column takes its full height: the lyrics box (once there are words in it) and the structure grow into the room, and the player stays at the foot of the column.
• The whole picture. Clicking the picture in the editor's Picture menu shows the whole of it, as large as the window allows; a click anywhere or Escape puts it away and leaves the editor open.
• Voice cues in the expanded lyrics editor. Buttons for + Male, + Female and + Both in the expanded lyrics window add [male voice], [female voice] or [both voices] on the line where the cursor is, and Done moves to the bottom status line beside the word count so the buttons fit on one line.

Earlier updates (0.0.44 to 0.0.40)

0.0.44 (2026-10-14)

• A picture for each take
• Hot keys
• Style LoRAs at install
• Mastering rack: a Boost preset
• The take that is playing pulses on its beat
• The editor has the take's own buttons
• …and 11 more

0.0.43 (2026-10-13)

• Mastering rack: a parametric equalizer
• Mastering rack: all of it in the window
• Mastering rack: three presets have a ratio the rack has a button for
• Mastering rack: a double-click on the compressor's OUTPUT goes to +6 dB
• Windows: Quit quits
• Sing again asks first

0.0.42 (2026-10-12)

• Saved styles
• The mastering rack opens from the editor window
• Mastering rack: the five presets are remade
• Mastering rack: the limiter limits, and MIX blends
• Mastering rack: the compressor's OUTPUT is all the gain it adds
• Mastering rack: saving a preset under a name already used replaces it
• …and 4 more

0.0.41 (2026-10-12)

• Piano roll
• Takes inside a take
• Windows

0.0.40 (2026-10-12)

• Takes made from a take are kept inside it (#48)
• Checkpoints closes the editor
• Take cards stand out from the page
• A mastered take's card says so
• Lyrics in the piano roll follow the tune of a song from a prompt
• Words matched to a score counted in thirty-seconds are in order
• …and 1 more

The full history is in the changelog.

Installer SHA-256: a7f0bcc58d7e632e2d039946067e1194e3da3b7063dc04e81ed30792f5293a63. The installer is unsigned, so Windows may show a SmartScreen warning.

v0.0.44

Windows: download Yeufonic-Setup-0.0.44.exe and run it over your existing install. Quit Yeufonic from its icon by the clock first (right-click, then Quit): closing the window leaves it running, and the installer waits for it to quit. Your library, settings, LoRAs and models are kept. Two new install options, both unticked: FLUX.2 [klein] (download size 8.2 GB), which draws each take's picture on your PC, and Style LoRAs (download size 3.8 GB), 39 style LoRAs by their authors.

Docker: in the Yeufonic folder run git pull, then docker compose restart app and reload the page. No rebuild. For pictures drawn on your PC run sh scripts/fetch-models.sh --pictures, and for the style LoRAs sh scripts/fetch-models.sh --loras, then docker compose restart app again.

What is new in 0.0.44

• A picture for each take. Each new take gets a picture of the song, made once no song is being made when it is drawn on this PC (so it never holds up a plan or a render) and while the take renders when an external LLM draws it, and shown faintly behind the take's card, across the top of the take panel when it is chosen, and behind the editor when the take is open there; a take made inside another shows that one's. It is described from the take's title, style and opening words with no LLM, drawn on white, and the app takes the white out: drawn on this PC a picture takes a few seconds and needs nothing but FLUX.2 [klein]. It is drawn on this PC by FLUX.2 klein 4B, a new install option (FLUX.2 [klein], unticked, about 8 GB; fetch-models.sh --pictures for Docker), or by the external LLM's image model where the provider lists one; with neither, Settings says No pictures, the default background, and how to have them, until something that can draw is there. The editor's Picture button makes a new one or goes back to the default background, for good. Settings turn pictures off, choose who draws (this PC first), who describes (the take itself by default, or Local Gemma or the external LLM, slower) and which image model. The description leaves out real names, which an image model will not draw, and asks for soft lettering.
• Hot keys. On the main page single keys act on the chosen take, the one in the left panel: M opens the mastering rack (again to close it), E opens the take in the editor, I shows its details, N normalises it or undoes it, V opens Variations, T Try more, S stars or unstars it, D deletes it after asking, and L opens or closes the logs. They do what the buttons do, they say in the status bar why when a take cannot, and they keep out of the way while you type, while a window is open, or with Ctrl, Alt or the system key held. A button's tip names its key, and the guide has a Hot keys section.
• Style LoRAs at install. A new, unticked install option, Style LoRAs (fetch-models.sh --loras for Docker), fetches 39 style LoRAs by their authors, 3.8 GB, each with a description giving its trigger word and suggested strengths: chanson, canzone, Bulgarian voices, qawwali, militant reggae, J-pop, dream pop, old school hip-hop and sixteen concept sliders. They are CC BY-NC 4.0, gathered unchanged in one Hugging Face repository whose README credits every author, and each is checked against its checksum. A LoRA already in the folder under the same name is left as it is.
• Mastering rack: a Boost preset. A sixth factory preset, made by ear: a strong lift of the lows and highs, a touch of the parametric equalizer, the compressor catching only what passes full scale, and no limiter. Unlike the others it comes out a good deal louder than the take it is given, and its loudest peaks clip, which is part of its sound.

UI enhancements

• The take that is playing pulses on its beat. Its edge widens and flashes the Play button's green on every other beat of its score's tempo (following the speed control, and back in step after a seek), with a slow breath for a take with no tempo, and the card sits a little larger than the takes at rest, so the take you are hearing is found at a glance. Reduced motion holds it still.
• The editor has the take's own buttons. Sing again, Variations, Try more, Inside, Take out, Normalise and Move sit in the window's top row, larger than on the card, lit as on the card and with tips that follow. The Picture button is an icon among them.
• The take panel shows the take's picture, as a banner across its top, behind the buttons and the title, fading out before the details. The title and details over it are bolder, the style chips sit on a soft backing, and the section headings (Style, Sound, Words, Score) on a strip of their own, so they read over any picture.
• Clearer, larger lettering. The text on every take card, and the small text in the editor, is a point larger, with a little more weight and a fine edge that keeps it sharp over a picture. A take's picture behind its card is less saturated, so it no longer competes with the card.
• Mastering rack: delete a preset from its bin. The preset box is a menu with a bin beside each of your own presets; deleting one leaves the rack sounding as it does. The old Delete button only showed while a preset was unchanged, and put the rack back to flat.
• The mastering rack follows the theme. In the light theme it was still black. Its panels are now brushed silver with dark lettering there, each unit keeping its own tint and coloured top edge; the knobs, the meter windows and the readouts keep the dark look of the hardware. The parametric equalizer's graph is drawn in light colours to match. The dark themes keep the rack as it was.
• Equalizer band tips. Each band of the parametric equalizer says what it is and how it is set when the pointer rests on it.
• Backdrop colour. A setting under Theme tints the page behind the takes with a hue you choose, at a strength from a faint wash to a rich colour, and the page changes as you drag. The take cards, the panels and the windows keep their own colours, and the high contrast theme stays black. Like every setting it is kept by the app, so every window onto it shows the same.
• The editor's player has its keys. Space and the arrows work the player for the take open in the editor, as they do the page's player; typing in a box is left alone.
• The editor is always three columns. The Editor layout setting, which offered the editor as steps one part at a time, is gone: the editor is three columns with everything in view, and the score on a tab of its own.
• The layout buttons say what they do. The buttons beside Starred read Comfy or Compact, and Fit or Wide, for the layout a click gives, rather than the one on show.
• Escape closes the logs before anything under them, and closes the popped-out log window.
• Reload to update now appears for a change to the page's look, as it did for a change to its scripts.

Earlier updates (0.0.43 to 0.0.39)

0.0.43 (2026-10-13)

• Mastering rack: a parametric equalizer
• Mastering rack: all of it in the window
• Mastering rack: three presets have a ratio the rack has a button for
• Mastering rack: a double-click on the compressor's OUTPUT goes to +6 dB
• Windows: Quit quits
• Sing again asks first

0.0.42 (2026-10-12)

• Saved styles
• The mastering rack opens from the editor window
• Mastering rack: the five presets are remade
• Mastering rack: the limiter limits, and MIX blends
• Mastering rack: the compressor's OUTPUT is all the gain it adds
• Mastering rack: saving a preset under a name already used replaces it
• …and 4 more

0.0.41 (2026-10-12)

• Piano roll
• Takes inside a take
• Windows

0.0.40 (2026-10-12)

• Takes made from a take are kept inside it (#48)
• Checkpoints closes the editor
• Take cards stand out from the page
• A mastered take's card says so
• Lyrics in the piano roll follow the tune of a song from a prompt
• Words matched to a score counted in thirty-seconds are in order
• …and 1 more

0.0.39 (2026-10-11)

• Voices follow the cues: a man, a woman or both, line by line
• A long song is no longer cut short by its cap
• The cap a new take starts with is a setting
• The mastering rack's preset box says what the rack holds

The full history is in the changelog.

Installer SHA-256: c074967521d1b535e1e907e9a0ea006be2c9d568db8c1899e1cf5443f1998455. The installer is unsigned, so Windows may show a SmartScreen warning.

v0.0.43

Windows: download Yeufonic-Setup-0.0.43.exe and run it over your existing install. Quit Yeufonic from its icon by the clock first (right-click, then Quit): closing the window leaves it running, and the installer waits for it to quit. Your library, settings, LoRAs and models are kept.

Docker: in the Yeufonic folder run git pull, then reload the page in your browser. This release needs no restart and no rebuild.

What is new in 0.0.43

• Mastering rack: a parametric equalizer. A new unit between the 1073 EQ and the compressor, with seven bands on a graph. Each band is a dot: drag it across for its frequency and up or down for its gain, turn the mouse wheel over it for its width, double-click it to reset it. A band can be a low cut, a low shelf, a bell, a notch, a high shelf or a high cut, and can be switched off; a cut waits, switched off, at each end. The graph shows each band's own curve in its colour, the curve of all of them together, and, while a take plays, the spectrum of what is heard behind them. Flat, beside the kinds, puts every band back to 0 dB and switches the cuts and notches off, leaving each band where it is. A button in the graph's corner sets how many decibels it shows (6, 12 or 18 either side), for fine moves or broad ones. The equalizer is part of the rack's settings: it is kept with the take, and a preset you save holds it. The factory presets leave it out and flat.
• Mastering rack: all of it in the window. With five units the rack is taller than most windows. It now takes all the height the window has and is drawn smaller, by as much as it needs for every unit to be seen at once, so there is nothing to scroll. Its lettering is set larger as the rack is drawn smaller, so labels and readouts stay close to their own size and stay readable. Full size in its header shows the rack at its own size instead, and scrolls; Fit goes back. In a very short window it stops shrinking at a little over half its size and scrolls the rest.
• Mastering rack: three presets have a ratio the rack has a button for. Vintage Tube Warmth, Vocal Air & Glue and Acoustic Clarity used compressor ratios between the rack's buttons, so none was lit when one of them was chosen. They are now 2:1, with threshold and output moved so that they come out as loud as before.
• Mastering rack: a double-click on the compressor's OUTPUT goes to +6 dB. That is where Reset leaves it; the double-click went to 0 dB, so the two disagreed.
• Windows: Quit quits. Choosing Quit from the icon by the clock asked "Quit Yeufonic?" every time. It now quits at once, and asks only when a render or a training run is going, which quitting would stop.
• Sing again asks first. One click on the card's Sing again button started a render at once. It now asks, saying what it will do, and does nothing if you cancel.

Earlier updates (0.0.42 to 0.0.38)

0.0.42 (2026-10-12)

• Saved styles
• The mastering rack opens from the editor window
• Mastering rack: the five presets are remade
• Mastering rack: the limiter limits, and MIX blends
• Mastering rack: the compressor's OUTPUT is all the gain it adds
• Mastering rack: saving a preset under a name already used replaces it
• …and 4 more

0.0.41 (2026-10-12)

• Piano roll
• Takes inside a take
• Windows

0.0.40 (2026-10-12)

• Takes made from a take are kept inside it (#48)
• Checkpoints closes the editor
• Take cards stand out from the page
• A mastered take's card says so
• Lyrics in the piano roll follow the tune of a song from a prompt
• Words matched to a score counted in thirty-seconds are in order
• …and 1 more

0.0.39 (2026-10-11)

• Voices follow the cues: a man, a woman or both, line by line
• A long song is no longer cut short by its cap
• The cap a new take starts with is a setting
• The mastering rack's preset box says what the rack holds

0.0.38 (2026-10-10)

• Vintage Stereo Imager & Spatial Processor
• A saved take sounds as it plays
• A rack nobody has touched does nothing
• A page left open across an update says so
• The limiter's Out is a true bypass
• Windows: Save asks where to put the file
• …and 2 more

The full history is in the changelog.

Installer SHA-256: 1888fd61bd298d7c75ad992f99ae16a827d3d356963a37c7d18c5f8a113f401b. The installer is unsigned, so Windows may show a SmartScreen warning.

v0.0.42

Windows: download Yeufonic-Setup-0.0.42.exe and run it over your existing install. Quit Yeufonic from its icon by the clock first (right-click, then Quit): closing the window leaves it running, and the installer waits for it to quit. Your library, settings, LoRAs and models are kept.

Docker: in the Yeufonic folder run git pull, then docker compose restart app, and reload the page in your browser. This release needs no rebuild.

What is new in 0.0.42

• Saved styles. A Save button beside the style menu keeps the style in the Style box under a name you give it. The menu's first entry is Saved styles, which opens onto the ones you kept, each with a bin to delete it, and the menu's search finds them with the canned styles. Saving under a name already used replaces that style, after asking. A style is kept without a style LoRA's trigger word, which is put back for the LoRA in use when the style is chosen. They are kept by the app, so the Windows window and any browser see the same ones.
• The mastering rack opens from the editor window. A Mastering button in the window's header, to the left of Advanced, is there when the take has sound, and is green while the rack changes how the take sounds. The rack opens over the window; Escape closes the rack first, and the window after. With the rack open there, the keys are the page's: space plays and pauses the take, the left and right arrows move through it, up and down set the volume.
• Mastering rack: the five presets are remade. The old ones made a take much louder than it came in and pushed it past full scale inside the rack, where it was cut off flat: that was the distortion. Each preset now turns the input down before it adds anything, comes out at about the loudness it was given (Radio Ready a little more), and leaves the peaks under full scale. A take already saved with one of the old presets keeps those settings, and the preset box shows it as Custom.
• Mastering rack: the limiter limits, and MIX blends. A peak over full scale used to be cut off flat by the stage before the limiter, so the limiter never saw it and Drive was a hard clip. An over now reaches the limiter whole, the tube stages bend it without a corner, and the ceiling is the most the rack puts out. The compressor's MIX used to thin the sound, because the compressed signal comes out a few milliseconds late and was blended with a dry signal that was not; the dry signal now waits for it. A take mastered with Drive up, or with MIX below 100%, will sound cleaner than it did.
• Mastering rack: the compressor's OUTPUT is all the gain it adds. The browser's compressor raises what it puts out by an amount of its own, more the lower the threshold and the higher the ratio, so a compressor with OUTPUT at 0 still made a take louder. The rack now takes that amount off, and OUTPUT at 0 leaves the quiet parts as loud as they were. Settings already kept with a take, and presets of your own, are read with that amount added to their OUTPUT, so they sound as they did and the knob shows the whole of it. The flat preset's compressor waits at +6 dB for the same reason: switched in, it compresses without a drop in level. The knob now goes up to +30 dB.
• Mastering rack: saving a preset under a name already used replaces it. It used to add a second preset with the same name. The rack now asks, then writes the preset over; presets left sharing that name by earlier saves become one.
• Mastering rack: opened for another take while one is playing, it plays that take. The take that was playing used to stop and leave silence; now the take the rack is set for is heard in its place. With nothing playing, opening the rack starts nothing.
• Mastering rack: Dock and Float mean what they say. A docked rack stays at the foot of the window and is no longer moved by dragging its header; Float lets it be dragged, and puts it back where it last floated. The first Float is centred by the rack's real height, where it used to land almost where the docked rack was.
• A take's title has the top of its card to itself. The small buttons that sat beside the title (Sing again, Variations, Try more, Inside, Normalise, the mastering rack, Move) are now on the status line, to the right, so a narrow card shows its title and not its first letter. A status too long to share the line takes it, and the buttons go under.
• The takes inside a take look like they belong to it. The cards in an open panel have the lighter shade of the take that holds them, where they used to look like more of the list.

Earlier updates (0.0.41 to 0.0.37)

0.0.41 (2026-10-12)

• Piano roll
• Takes inside a take
• Windows

0.0.40 (2026-10-12)

• Takes made from a take are kept inside it (#48)
• Checkpoints closes the editor
• Take cards stand out from the page
• A mastered take's card says so
• Lyrics in the piano roll follow the tune of a song from a prompt
• Words matched to a score counted in thirty-seconds are in order
• …and 1 more

0.0.39 (2026-10-11)

• Voices follow the cues: a man, a woman or both, line by line
• A long song is no longer cut short by its cap
• The cap a new take starts with is a setting
• The mastering rack's preset box says what the rack holds

0.0.38 (2026-10-10)

• Vintage Stereo Imager & Spatial Processor
• A saved take sounds as it plays
• A rack nobody has touched does nothing
• A page left open across an update says so
• The limiter's Out is a true bypass
• Windows: Save asks where to put the file
• …and 2 more

0.0.37 (2026-10-10)

• Vintage Mastering Rack & Channel Strip
• Piano Roll improvements
• Transport bar & take selection
• Modal and editor ergonomics

The full history is in the changelog.

Installer SHA-256: 38855b86fd677ca8aecefcb1d2a08920d2cb3e214679d10189a811458f2aa10b. The installer is unsigned, so Windows may show a SmartScreen warning.

v0.0.41

Windows: download Yeufonic-Setup-0.0.41.exe and run it over your existing install. Quit Yeufonic from its icon by the clock first (right-click, then Quit): closing the window leaves it running, and the installer waits for it to quit. Your library, settings, LoRAs and models are kept.

Docker: in the Yeufonic folder run git pull, then docker compose restart app, and reload the page in your browser. This release needs no rebuild.

What is new in 0.0.41

• Piano roll: saving a score no longer drops a chord that changes at a tie or part-way through a rest, and no longer moves a note by a semitone where a plain note follows a sharp or flat in the same bar.
• Takes inside a take: when one is playing and its panel is shut, the Play button on the take that holds it glows gold.
• Windows: the picture down the side of the installer says Yeufonic, where it still carried the app's earlier name. The installer on this page was replaced for this.

Earlier updates (0.0.40 to 0.0.36)

0.0.40 (2026-10-12)

• Takes made from a take are kept inside it (#48)
• Checkpoints closes the editor
• Take cards stand out from the page
• A mastered take's card says so
• Lyrics in the piano roll follow the tune of a song from a prompt
• Words matched to a score counted in thirty-seconds are in order
• …and 1 more

0.0.39 (2026-10-11)

• Voices follow the cues: a man, a woman or both, line by line
• A long song is no longer cut short by its cap
• The cap a new take starts with is a setting
• The mastering rack's preset box says what the rack holds

0.0.38 (2026-10-10)

• Vintage Stereo Imager & Spatial Processor
• A saved take sounds as it plays
• A rack nobody has touched does nothing
• A page left open across an update says so
• The limiter's Out is a true bypass
• Windows: Save asks where to put the file
• …and 2 more

0.0.37 (2026-10-10)

• Vintage Mastering Rack & Channel Strip
• Piano Roll improvements
• Transport bar & take selection
• Modal and editor ergonomics

0.0.36 (2026-10-09)

• CPU and GPU usage in the header
• Refreshed UI
• The end-of-render notes can be dismissed
• The section list says when a take ended short of its score

The full history is in the changelog.

Installer SHA-256: 69a2549eef3142a7d2f86ee23993f47797cce82321c55ad2335f1b38e2cb06ec. The installer is unsigned, so Windows may show a SmartScreen warning.

v0.0.40

Windows: download Yeufonic-Setup-0.0.40.exe and run it over your existing install. Quit Yeufonic from its icon by the clock first (right-click, then Quit): closing the window leaves it running, and the installer waits for it to quit. Your library, settings, LoRAs and models are kept.

Docker: in the Yeufonic folder run git pull, then docker compose restart app, and reload the page in your browser. This release needs no rebuild.

What is new in 0.0.40

• Takes made from a take are kept inside it (#48). Variations, Try more and Checkpoints each make several takes of one song, and they used to sit in the list beside the take they came from, told apart by the end of their titles. They are now kept inside that take, so a space lists songs rather than every render of each. A take that holds any, finished or still being made, has an Inside button on its card, lit orange, and its card is plainly lighter than the others (a warm tint in the light theme). It opens a panel under the card with the takes inside as cards of their own, each one a take like any other: play it, star it, save it, double-click it to open its song and score. A take shows there once it is finished. There is one level: a take inside another holds none of its own, and Take out on its card makes it a take of its own again. While a panel is open, Select all, Starred, the kind buttons, Compare, Delete and the player's previous and next work on the takes in it. A search finds takes wherever they are. Deleting a take deletes the takes inside it, after a warning that says how many, and moving a take to another space takes them along. Checkpoints, pressed with a take open in the editor, puts its takes inside that take; for a new song the take on the last checkpoint ticked holds the others, and it is not offered from a take that is itself inside another. Try more no longer ticks the takes it makes: with the panel open, Select all and Compare do that. Takes made before this stay as they are. An agent using the MCP server sees how many takes a take holds and can list them.
• Checkpoints closes the editor. Pressing Render in the Checkpoints window left the editor open over the library, which looked as if nothing had happened. It now closes, as it does when a plan is written, and the library shows the take being made.
• Take cards stand out from the page. A card had a faint outline on a page almost its own colour. In the dark and Studio themes it now has a clearer outline and a lit top edge; in the light theme a soft shadow lifts it; the high contrast theme, whose outline was already strong, is unchanged.
• A mastered take's card says so. The mastering rack's button on a card is lit green when the rack changes how that take sounds.
• Lyrics in the piano roll follow the tune of a song from a prompt. The roll shared the lyrics out section by section, and a plan's sections are seldom the lyrics' sections. A heading followed at once by a voice cue such as [male voice] counted as a section with no words, so the first sung phrase had none, and each block of the lyrics was squeezed into whichever section of the plan came next. Where the two do not count the same, every line is now fitted to the whole melody at once, as a cover's words already were, and a line that only says who sings is no longer a section. A score matched before this keeps the words it was given: Match Lyrics sets them again.
• Words matched to a score counted in thirty-seconds are in order. Matching a cover's or an import's words to its score read every length as a sixteenth. In a score that counts in thirty-seconds a bar then ran on into the next, the notes of neighbouring bars were taken out of order, and the words came out shuffled. Lengths are read in the score's own unit, fractions and whole bars of rest included.
• Match Lyrics leaves the score's sections alone. When the lyrics had more headings than the score had sung sections, the piano roll divided a long section at its pauses to share the words out, and wrote that division into the score as new % verse lines. The division is now only its own working.

Earlier updates (0.0.39 to 0.0.35)

0.0.39 (2026-10-11)

• Voices follow the cues: a man, a woman or both, line by line
• A long song is no longer cut short by its cap
• The cap a new take starts with is a setting
• The mastering rack's preset box says what the rack holds

0.0.38 (2026-10-10)

• Vintage Stereo Imager & Spatial Processor
• A saved take sounds as it plays
• A rack nobody has touched does nothing
• A page left open across an update says so
• The limiter's Out is a true bypass
• Windows: Save asks where to put the file
• …and 2 more

0.0.37 (2026-10-10)

• Vintage Mastering Rack & Channel Strip
• Piano Roll improvements
• Transport bar & take selection
• Modal and editor ergonomics

0.0.36 (2026-10-09)

• CPU and GPU usage in the header
• Refreshed UI
• The end-of-render notes can be dismissed
• The section list says when a take ended short of its score

0.0.35 (2026-10-08)

• Build a song's structure, then write the words
• How the lines sound
• Follow this structure exactly
• A chorus no longer opens the way the verse did
• Details shows the structure
• New songs start at Harmony Varied and interpretation Wide
• …and 7 more

The full history is in the changelog.

Installer SHA-256: 5f2b6762cdc01e845ffd5ec0422a9ef9aa11474a35b0eaebbb985d89ebe68b7f. The installer is unsigned, so Windows may show a SmartScreen warning.

v0.0.39

Windows: download Yeufonic-Setup-0.0.39.exe and run it over your existing install. Quit Yeufonic from its icon by the clock first (right-click, then Quit): closing the window leaves it running, and the installer waits for it to quit. Your library, settings, LoRAs and models are kept.

Docker: in the Yeufonic folder run git pull, then docker compose restart app, and reload the page in your browser. This release needs no rebuild.

What is new in 0.0.39

• Voices follow the cues: a man, a woman or both, line by line. A line in the lyrics such as [male voice], [female voice] or [both voices] says who sings the lines under it, until the next cue or the next section. A cue alone did not decide the voice: the render takes the singer mostly from how high the melody is written, and the planner tends to write a verse in two octaves, changing every four lines or so whoever was cued, so a block of eight lines for one singer could come out shared between two. With Voices follow the cues ticked (under the lyrics box, shown once the lyrics have a cue; on for a new take), each cued stretch of the plan's melody is moved by whole octaves to where that voice sings, low for a man and high for a woman; for both, the line is put where a woman sings and the second voice is left to the render. The lyric lines are matched to the phrases of the melody over the whole song; phrases that sit together move together, so a tune is not broken in the middle; a plan that is already as cued is left exactly as it is. Nothing else in the plan changes. It is done when the plan is written and again before each render, so it follows a change to the words or the score, and it works for a cover's words too. The Score page has Voices to cues, which does the same once to the score in the box, for a take made before the tick existed or with it off. Who sings a line is still partly chance: with the melody in the right place most lines go to the voice cued, and another seed can put right one that did not. Details shows the setting. The guide's chapter on writing a song from a prompt explains the cues.
• A long song is no longer cut short by its cap. The length cap of a song or an instrumental from a prompt now rises to fit its plan when the plan is written: to the plan's length with about ten seconds to spare, rounded up to ten, as a recording's score has always set a cover's. It is only raised, never brought down, and never past the 900 seconds the engine allows. A cap you typed yourself stays as you set it, and what is made from that take keeps it. A plan more than a third longer than its cap, and over eight minutes, is still written again rather than taken for a song.
• The cap a new take starts with is a setting. Settings has Length cap for a new take (4, 6, 8, 10 or 15 minutes; 6 as before), and DEFAULT_CAP in the environment sets what it starts at. A take made through the API or by an agent without a cap gets it too.
• The mastering rack's preset box says what the rack holds. When the rack takes up another take's settings, as it does when the take playing changes, the box names the preset those settings are, a factory one or one of your own, where it used to go on showing the last preset chosen. It shows Custom once a control has been moved away from a preset.

Earlier updates (0.0.38 to 0.0.34)

0.0.38 (2026-10-10)

• Vintage Stereo Imager & Spatial Processor
• A saved take sounds as it plays
• A rack nobody has touched does nothing
• A page left open across an update says so
• The limiter's Out is a true bypass
• Windows: Save asks where to put the file
• …and 2 more

0.0.37 (2026-10-10)

• Vintage Mastering Rack & Channel Strip
• Piano Roll improvements
• Transport bar & take selection
• Modal and editor ergonomics

0.0.36 (2026-10-09)

• CPU and GPU usage in the header
• Refreshed UI
• The end-of-render notes can be dismissed
• The section list says when a take ended short of its score

0.0.35 (2026-10-08)

• Build a song's structure, then write the words
• How the lines sound
• Follow this structure exactly
• A chorus no longer opens the way the verse did
• Details shows the structure
• New songs start at Harmony Varied and interpretation Wide
• …and 7 more

0.0.34 (2026-10-08)

• A corpus without vocals
• Render this score makes a new take
• Play after switching spaces

The full history is in the changelog.

Installer SHA-256: 446cc2a0f5f9785ef136d22ddc2bdc9a03e1b5b19303ebf12a5937ef4e67b97f. The installer is unsigned, so Windows may show a SmartScreen warning.

v0.0.38

Windows: download Yeufonic-Setup-0.0.38.exe and run it over your existing install. Quit Yeufonic from its icon by the clock first (right-click, then Quit): closing the window leaves it running, and the installer waits for it to quit. Your library, settings, LoRAs and models are kept.

Docker: in the Yeufonic folder run git pull, then docker compose restart app, and reload the page in your browser. This release needs no rebuild.

What is new in 0.0.38

• Vintage Stereo Imager & Spatial Processor. Added a skeuomorphic stereo imaging module to the Vintage Mastering Rack, positioned immediately before the master limiter:
  • Bigness width control: Seamless stereo soundstage width management from mono fold-down (0), transparent stereo (1), up to expansive spatial stereophony (9).
  • Bass punch filter: Dedicated 85 Hz mono-summed low-end punch filter keeping sub frequencies centred and punchy without stereo smear.
  • Analogue tube harmonics: Tube saturation circuit introducing gentle musical warmth across the stereo bus.
  • Range & Stage spatial network: Band-focusing filter network and all-pass phase rotation to shape stereo spread and depth.
• A saved take sounds as it plays. Saving a take the mastering rack changes renders the file in the page, through the chain that plays it: the same nodes and the same settings code, on an offline context at the file's own sample rate. The server converts and tags that file and no longer processes the sound itself, so there is no second version of the chain to drift from the first. If the render fails, nothing is saved and the status line says so. The take itself is left untouched.
• A rack nobody has touched does nothing. The compressor, the stereo imager and the limiter start switched out, and the rack is in the path only when something in it changes the sound, so a take plays and saves exactly as it was made until a control is used. Turning a control of a module that is out switches it in. Settings kept by earlier builds, which had all three in at their starting values, are read as out where they were never adjusted, and a take that was only played or saved no longer has settings written to it.
• A page left open across an update says so. A page loads its scripts once, and after the app behind it is updated and restarted it goes on running the old ones. The header now shows Reload to update when the page's scripts are not the ones the app would serve, and a save that an older page asks the server to master is refused with that advice rather than handed over unmastered.
• The limiter's Out is a true bypass, like the other modules', and a render keeps the take's sample rate rather than the sound card's.
• Windows: Save asks where to put the file. The app's window is told to ask on every start, so an install whose window was first opened by an earlier version asks too, where it used to save straight to Downloads.
• Mastered audio export cache prevention: Enforced Cache-Control: no-cache, no-store, must-revalidate headers across audio download and export endpoints so browsers never serve stale or unmastered cached audio.
• AudioContext autoplay reliability: Global user gesture listeners unlock and auto-resume the Web Audio context immediately on playback and transport actions, ensuring audio begins without first having to open the mastering rack.

Earlier updates (0.0.37 to 0.0.33)

0.0.37 (2026-10-10)

• Vintage Mastering Rack & Channel Strip
• Piano Roll improvements
• Transport bar & take selection
• Modal and editor ergonomics

0.0.36 (2026-10-09)

• CPU and GPU usage in the header
• Refreshed UI
• The end-of-render notes can be dismissed
• The section list says when a take ended short of its score

0.0.35 (2026-10-08)

• Build a song's structure, then write the words
• How the lines sound
• Follow this structure exactly
• A chorus no longer opens the way the verse did
• Details shows the structure
• New songs start at Harmony Varied and interpretation Wide
• …and 7 more

0.0.34 (2026-10-08)

• A corpus without vocals
• Render this score makes a new take
• Play after switching spaces

0.0.33 (2026-10-07)

• Make an instrumental is now available for covers as well as for songs from a prompt

The full history is in the changelog.

Installer SHA-256: a4503d8599aa32c323023a27cec227ac034c4e0181f1ab6fb07f037f1d25c37a. The installer is unsigned, so Windows may show a SmartScreen warning.

v0.0.37

Windows: download Yeufonic-Setup-0.0.37.exe and run it over your existing install. Quit Yeufonic from its icon by the clock first (right-click, then Quit): closing the window leaves it running, and the installer waits for it to quit. Your library, settings, LoRAs and models are kept.

Docker: in the Yeufonic folder run git pull, then docker compose restart app. This release needs no rebuild.

What is new in 0.0.37

• Vintage Mastering Rack & Channel Strip. A skeuomorphic studio mastering rack with ultra-low latency real-time Web Audio DSP and matching export rendering:
  • 1073 Equalizer: Preamp drive gain (-15 to +20 dB), high-pass low cut (20 to 300 Hz), low shelf (35 to 220 Hz, ±16 dB), mid peaking bell (360 Hz to 7.2 kHz, ±18 dB), 12 kHz high shelf (analog 6.8 kHz corner slope, ±16 dB), phase invert, and output trim.
  • Vintage Compressor: Fast FET-style compressor with stepped ratios (2:1, 4:1, 8:1, 12:1, 20:1, and "All Buttons In" mode), threshold, attack (1 to 80 ms), release (50 to 1200 ms), makeup gain, parallel mix blend, and an analog gain reduction VU meter.
  • 670 Variable-Mu Master Limiter & Tube Warmth: Clean input drive, tube warmth harmonic saturation curve, true peak ceiling control, and release timing.
  • Chassis & Workflow: Draggable floating or bottom-docked chassis, take header selector with quick-access M icon on each take card, factory and user preset manager, and master bypass A/B auditioning.
  • Export Rendering: All mastering rack settings are non-destructively saved per take and baked into exported FLAC, WAV, and MP3 renditions.
• Piano Roll improvements:
  • Follow playback auto-scroll: The piano roll view smoothly scrolls to follow the active bar as audio plays, keeping the playback position in view.
  • Voice toggle: Switch quickly between Vocal and Instrumental voice parts in the Piano Roll (with dedicated button and keyboard shortcut).
  • Key signatures and accidentals: Corrected note pitch resolution and accidental tracking in flat keys (such as F, B♭, and E♭), respecting measure barlines and key signatures.
  • Score saving: Score edits now auto-save on modal exit and editing, and the score modal has an explicit Save button.
• Transport bar & take selection:
  • Selecting or highlighting a take now links its recording to the transport bar, so pressing Spacebar immediately plays the selected take.
  • Selecting another take does not abruptly stop active playback; playback continues undisturbed until Spacebar or Play is clicked.
  • Double-clicking any take card immediately opens the Take Editor.
• Modal and editor ergonomics:
  • The song and score editor modals remain open when clicking outside on the backdrop, preventing accidental lost state.
  • Relocated the Take Editor Close button to the top-right header.

0.0.36 (2026-10-09)

• CPU and GPU usage in the header
• Refreshed UI
• End-of-render notes dismissible
• Section list drift warning

The full history is in the changelog.

Installer SHA-256: 3f491f7aa53dacb4cbb7684cb64d6da29c5b2c835a70e4765bbe0ff15491288e. The installer is unsigned, so Windows may show a SmartScreen warning.

v0.0.36

Windows: download Yeufonic-Setup-0.0.36.exe and run it over your existing install. Quit Yeufonic from its icon by the clock first (right-click, then Quit): closing the window leaves it running, and the installer waits for it to quit. Your library, settings, LoRAs and models are kept.

Docker: in the Yeufonic folder run git pull, then docker compose restart app. This release needs no rebuild.

What is new in 0.0.36

• CPU and GPU usage in the header. Next to the version number, two subtle indicators show current GPU and CPU utilisation in percent, updating live with the app's state polling.
• Refreshed UI. Take cards, action buttons and waveforms have a refreshed look across all themes, and the README screenshots have been updated to match.
• The end-of-render notes can be dismissed. "Ran to the length cap" and "Stopped before the last section" have a × at their end, as the weak-render note has, for when the end sounds fine. A new take made from one starts with the note again.
• The section list says when a take ended short of its score. A render that is more than about 5 percent shorter than its score (and at least 6 s) now gets a line under the section list saying by how much, and that the list can be out of step with the audio, more so later in the song. The score's section times are right for the first part of every take and for takes that render in full; they drift after a render has compressed or skipped something.

Earlier updates (0.0.35 to 0.0.31)

0.0.35 (2026-10-08)

• Build a song's structure, then write the words
• How the lines sound
• Follow this structure exactly
• A chorus no longer opens the way the verse did
• Details shows the structure
• New songs start at Harmony Varied and interpretation Wide
• …and 7 more

0.0.34 (2026-10-08)

• A corpus without vocals
• Render this score makes a new take
• Play after switching spaces

0.0.33 (2026-10-07)

• Make an instrumental is now available for covers as well as for songs from a prompt

0.0.32 (2026-10-07)

• A Details button on a take
• Make an instrumental from a song's or cover's score
• Switching a corpus song's words between versions no longer drafts them again
• Each interpretation shows what it does as a tooltip in the menu
• Stems, vocal separation and lyric hearing run on the GPU
• The take card's Sing again, Variations and Try more icons have their tooltips back

0.0.31 (2026-10-06)

• The guide says where to find style LoRAs
• The Corpora pages share LoRAs
• New cover starts with no recording
• Writing a plan closes the editor
• A planned instrumental lists its sections to rearrange, as one from a recording does
• A take that ends before its score's last section is flagged, and retried once
• …and 2 more

The full history is in the changelog.

Installer SHA-256: 60584e15e47cfccc13970e9a1601ae915ec90c7a151fa36d0a859968c2f2037a. The installer is unsigned, so Windows may show a SmartScreen warning.