Skip to content

Producer studio: song planner, remix takes, vocal balance, library backup, style packs - #12

Open
IsaacNewtonne wants to merge 19 commits into
ServeurpersoCom:masterfrom
IsaacNewtonne:producer-studio
Open

IsaacNewtonne wants to merge 19 commits into
ServeurpersoCom:masterfrom
IsaacNewtonne:producer-studio

Conversation

@IsaacNewtonne

@IsaacNewtonne IsaacNewtonne commented Sep 26, 2026 •

Copy link
Copy Markdown

Summary

Producer-focused WebUI and server upgrades on top of ea07706. No model or inference defaults changed; the planner compiles direction into the existing YuE2 style prompt.

What is included

  • Song planner and voices (singing amount, vocal presence, 1-4 voices, tempo, arrangement, 3-take compare mode) — docs/PRODUCER_CONTROLS.md
  • Remix with original seeds and variation control, take-group compare in Library
  • Vocal-level adjust (/vocal-balance, auto and manual -6..+6 dB) + setup-vocal-separation.ps1
  • Library backup/restore (ZIP), favourites, artwork attach, ComfyUI artwork hooks
  • Ollama lyric assistant, style blend favourites, influence ring, studio rail + signal visualizer
  • 100+ music style packs (music_styles/) and lyric style packs (lyric_styles/) with thumbnails
  • Server: /vocal-balance endpoint, mastering profiles, smoke test (tests/server-smoke.mjs), WebUI tests

Validation

  • Node regression tests in tools/webui/tests/, Svelte check clean, Vite + native Release builds, isolated Chrome checks of controls and sequential fresh-seed takes. Browser generation uses mocked audio.

Excluded

Local run logs, agent notes (AUDIT_REPAIRS.md, PROJECT_BUILD_PLAN.md, pinokio_agent/) and docs/lyric-style-pack/ working docs are intentionally not part of this PR.

…ckup, style packs

Song planner and voices, 3-take compare mode, remix with original seeds, vocal-level adjust, library backup/restore, Ollama lyric assistant, ComfyUI artwork hooks, server smoke test, 100+ music styles and lyric style packs. See docs/PRODUCER_CONTROLS.md.
@ServeurpersoCom

Copy link
Copy Markdown
Owner

Thanks for the work! yue2.cpp stays a minimal engine on purpose, so I won't merge an app layer like this one into it: style packs, library, agents and integrations belong in a project built on top of it.

That's exactly what the engine is made for, and I'll happily list your project in the upcoming Third-party projects section, like YuE2 Studio.

Add theme-specific typography, surfaces, panel geometry and navigation. Improve mobile navigation and light-theme action contrast, and rebuild the embedded UI.

Validation: production UI build and svelte-check passed; six-theme browser review covered desktop, tablet and phone layouts.
…t ceiling

Auto duration now stops at 1.2x the lyric estimate plus 15 s (client
estimate via duration_estimate, else the server estimate). Scores with no
sung bars or a missing chorus are redrawn up to score_retries (default 2)
times before rendering; later choruses reuse the first chorus melody when
their words match (chorus_lock, default on). Adds test-score-check.
vocal_polish off/mix/stems: EQ, de-ess and gentle compression on the
mix, or on the BS-RoFormer vocal stem (opt-in, ~640 MB first-use
download). Releases Ollama models (keep_alive 0) before a job so the GPU
has room.
Style becomes a tag line of 300 characters or less in the official order;
lyrics are sent with standard section labels, no title, voice or speaker
names. Editor text is unchanged. Prompt format clean/legacy is stored for
A/B tests.
Single Song length control (Auto fits the lyrics) with a green/amber/red
fit meter; section map and planner coloured by section (verse blue,
chorus gold, pre-chorus orange, bridge violet, intro/outro slate,
instrumental teal) and by voice; stage colours in Activity; quality
controls (vocal polish, -14 LUFS, chorus lock, score retries, prompt
format) with engine settings in a drawer; pre-flight chips by Generate;
12 px minimum text, contrast and 44 px touch targets; compact phone header.
studio.ps1 prints a YUE2 STUDIO gradient banner (24-bit colour with a console-colour fallback), checks the server, build, model files, VRAM and Ollama, starts the server via server.cmd or the newest build exe, waits for the port and opens the browser. Supports -Port, -Host and -NoBrowser / YUE2_NO_BROWSER. Pure ASCII; Windows PowerShell 5.1 and PowerShell 7. studio.cmd is a double-click wrapper.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants