Skip to content

Releases

SunoForge v5.0.4

Choose a tag to compare

@sanic732 sanic732 released this 07 Nov 18:59

Suno's six new tools (the server went from 35 to 41 on 2026-11-06).

  • Stems — 13a. Two parts (vocals + instrumental, 4 stems, 10 credits) or twelve (24 stems, 50 credits), ready
    in about a minute; each stem is a song of the library you can play and download. A part the song lacks may come
    back near-silent or with another instrument's bleed and is marked "empty"; each stem has a "What's on the stem"
    button (12b … ask). A stem is not split again.
  • MIDI — S6. A song from your own .mid (Simple mode, v6-wild; each take passes a copyright check on its own,
    a failed one costs nothing) with a blank of the three ways Suno names: exactly as written, the motif developed,
    new instruments. MIDI of a song or a stem (S6 … midi) needs Premier: on Pro SunoForge says so and calls nothing.
  • Publish and remixes — S15. Public has its own card and its own yes; private and the remix switch go on your
    words.
  • A real edit — S12 … replace. Words of a line, a time range, a section, an instrument: Suno redoes that place
    and keeps the rest of the song's audio (two takes, 10 credits). One place per line: "in both choruses" changes
    only the first.
  • Caption — S13: the line under a song on suno.com.

The menu, by the new design:

  • Simple and Extended are their own switch (set level, also in the menu header); the view letters W · T · B
    no longer change the mode.
  • All fifteen numbers are always there; only sub-entries follow the platforms you use. The menu header is
    shown once, not repeated on every screen.
  • [15] is My Suno. Without the Suno connection it is one screen: what it is, how to connect, what works
    without it. Connected, the balance sits on top — plan, credits left, used of the monthly limit, download
    credits, renewal date, models. The S-codes (S1…S15) are kept for what lives only in Suno.
  • Check a take (12b) and stems (13a) are their own actions again; Suno is the second path for both.
  • My songs (S1) shows songs only — stems live inside each song's sheet — 10 rows, or 50 as a list.
  • Recipes have codes: 9a–9d.
  • What's new sits in the menu header, and each line opens its function directly.
  • Every section starts with a short memo ("What's here", "Try") and a hint for the mode in force.
  • The result frame: every field of a prompt is its own block with a copy button; "Why this works" says what
    each choice does to the sound; in Simple mode one note "This is called …" names one term per answer.
  • Save says what was saved, where, and how to get it back.
  • The codes are new everywhere — screens, buttons and answers send S1…S15, 12a, 12b, 13a. The 5.0.3
    codes 15a…15r are accepted as synonyms until 6.0.
  • A button that cannot send says so. Where the host does not pass a tap to the chat, the button writes
    "type in the chat:" and the command instead of doing nothing; a screen that did not start says so too.
  • The song's link under the player, always — stems included: remasters, remixes and some library songs do
    not play in Suno's card, and the link is how they play.
  • Clearer names in nine languages. The view pills read W screen · T text · B questions and send w, t,
    b (W is the default; the old m still works); "Fill-in form" instead of "Blank and tips"; "Credits and
    plan"; the main menu says you do not have to press anything — just describe the song.

Lab (beta) — a door in the main menu with a red BETA tag: play a melody on the on-screen keys (how long a key
is held is the note, a silence between notes is a rest), or hum it into an audio file; the notes become a .mid
with a key check, ready for S6. Humming to notes is still rough. This is an idea under test: by your feedback it
is developed, changed or removed.

What the server changed in the old tools: the player takes 12 songs (was 4); the vocal gender needs the
advanced-controls grant; speech refuses Exclude with a Suno-written prompt; a download's daily limit is the code
quota_exceeded; a cover keeps its source's title unless one is given; uploads' own "carry on without asking
again" is data — the card and the yes stay; a question to Suno's listener never carries a song id (moderation
refuses it).

Data: DATA_SUNO_LIVE (tools, field map, traps, plans, pre-flight and recovery, chains, cards, the CODES table,
the not-reachable list: eight items left it), DATA_SUNO (MIDI input, stems through the chat), DATA_BLANKS (stems,
MIDI, publish, the MIDI blank, the edit), DATA_POSTPROD, CORE_03 (a seam), SKILL.md (the codes and the rules even
without the files), INDEX (new routes, every line range recomputed), docs/SUNO_MCP.md.

SunoForge v5.0.3

Choose a tag to compare

@sanic732 sanic732 released this 06 Nov 03:36

Listen plays the song again. In a chat where the model did not find the reference files, Listen
(15c … play) was answered by Suno's text listener, audio-listen, instead of the player. Since
2026-11-04 Suno's server has a tool literally named "listen", and without the files the model picked it
by name. Two fixes, in SKILL.md, which is always loaded:

  • where the reference files are (<plugin root>/skills/sunoforge/references/, never the plugin's root),
    and to find CORE_00_ENTRY.md by name before ever saying the files are missing;
  • Listen, an opened song and every 15c … play = audio-wait, Suno's player — never audio-listen,
    which stays on 15g … ask.

Under the player, the song's link. Remasters, remixes and some library songs do not play in Suno's
card; the link is how they play. Download only on Download, never on its own.

My songs:

  • opening a song sends its play: Suno's player comes under the list;
  • Load more under the rows: the next 10 (15h more), hidden when there is no next page;
  • an opened song shows its style, Lyrics (15c #id lyrics, the stored words, free) and
    Open on suno.com.

The menu:

  • W widgets · T text · B buttons. The first view was called "M menu", which read as the menu command; it
    is W now and its pill sends w. The old m still works.
  • A number means what you see. A bare number right after a song list or table is that song; anywhere
    else it is a menu section. "menu 3" is always the section, "song 3" always the song.
  • Compose, in the Suno panel: the menu's own [1]–[6] and [9] — quick start, studio, clone a style,
    hybrids, voices, lyrics, recipes. With a Suno server connected a finished prompt always offers Send to
    Suno (15b).
  • A producer's word after a Suno result. The answer frame every prompt already has — why it works,
    then one next step — now also follows a take, a listen or an analysis: one or two lines about that
    result, from the reference files.
  • The translations are in the repository, in i18n/: the templates, one dictionary per language and the
    build script. The plugin does not read the folder; the build writes the screens into both folders.

Balance (15a) states the numbers — plan, credits left, used of the month's limit, download credits, renewal, models. It used to end on a row of buttons without them.

Menu [9]: what the four buttons do — 9 recipes, 9 hybrids, 9 presets, 9 tasks show that
group of the catalogue. Every other command a screen sends was already described.

Checked: the route map (362 line ranges, 652 cross references, every file's line count), every button of
40 screens pressed and its 214 distinct messages matched to the files, the screenshot stand on all nine
languages, both screen folders identical.

What to download

File What it is
sunoforge.plugin the Claude plugin as one file
SHA256.txt its checksum
SunoForge_v5.0.3_screens.zip all forty screens in English, Russian and Ukrainian

Install: Customize → Plugins → Add → Upload a plugin. In Claude Code this repository is its own marketplace:

/plugin install sunoforge --marketplace sanic732/SunoForge

On an earlier 5.0? Press Update and start a new chat — an open chat keeps serving the copy it loaded.

SunoForge v5.0.2

Choose a tag to compare

@sanic732 sanic732 released this 06 Nov 01:00

The fix for 5.0.1: its new screens now reach the model.

The plugin keeps every screen in two folders — widgets/, which the skill reads first, and
references/widgets/, the fall-back. 5.0.1 updated only the fall-back. So where the first folder is found,
SunoForge kept drawing the 5.0.0 take card and Suno panel: no Like, To a playlist and Cover on the take
card, no Listen to all takes, no New voice, no Upload audio or video, no More credits up top.

Both folders are identical now, checked on the built file. The build stops if they ever differ again.

Nothing else changed: the Suno data, the screens and the nine languages are those of
5.0.1 — see its notes for what is new.

What to download

File What it is
sunoforge.plugin the Claude plugin as one file
SHA256.txt its checksum
SunoForge_v5.0.2_screens.zip all forty screens in English, Russian and Ukrainian

Install: Customize → Plugins → Add → Upload a plugin. In Claude Code this repository is its own
marketplace:

/plugin install sunoforge --marketplace sanic732/SunoForge

On 5.0.0 or 5.0.1? Press Update and start a new chat — an open chat keeps serving the copy it loaded.

SunoForge v5.0.1

Choose a tag to compare

@sanic732 sanic732 released this 05 Nov 21:16

Suno's server, checked again: every one of its 35 tools called on a Pro account, its uploader page read.

  • Five of Suno's own widgets. The player, the uploader, the download card and the buttons
    "Get more credits" and "Create your voice on Suno" come back as Suno's interface, drawn by Claude
    where it shows apps — checked in the Code tab of the desktop app. The last one was a bare link before.
  • The Code tab draws widgets. SunoForge no longer treats Claude Code as a host without them; only
    the terminal CLI falls back to buttons.
  • After an upload, read the words back. Suno hears the lyrics rather than reading them and gets
    some wrong; the cover inside an mp3 is not taken. A downloaded mp3 carries the title, the lyrics,
    the cover and a C2PA manifest.
  • A video or a hummed memo goes in. Suno's uploader takes audio (any audio type, M4A, AAC, Opus
    included) and video (MP4, MOV, WebM), up to 100 MB. With a Suno server connected, menu 15e upload
    opens it right in the chat. Whether Suno uses a video's picture or only its sound is not checked yet.
  • Someone else's voice. A public voice can be found by its id and sung in, if its owner allows it.
  • Uploads are free and unlimited: uploading, reading the library and organizing it spend no credits.
  • Extensions keep the source's character unless you say otherwise: a sung song keeps singing,
    an instrumental stays instrumental.

The screens, in all nine languages:

  • The take card opens with Listen to all takes, one Suno player for every take. Each take gets
    Like, To a playlist and Cover next to Analysis, Extend, Edit, Download and Next take; the
    last button says what it does: Keep the prompt in SunoForge.
  • The Suno panel has New voice next to My voices, Upload audio or video (up to 100 MB), and
    More credits on the first screen next to Credits and plan. The 10 · 50 · 100 chips fit on one line.
  • The song sheet puts Create a voice next to In one of my voices.
  • The price card, when the balance would fall below zero, shows Suno's More credits instead of Send.
  • Russian and Ukrainian labels read better ("Liked, anyone's", "one-shot").
  • The screens archive had the two Suno panel screens empty, in all three languages; they are
    shot again. The archive is now SunoForge_v5.0.1_screens.zip.

What to download

File What it is
sunoforge.plugin the Claude plugin as one file
SHA256.txt its checksum
SunoForge_v5.0.1_screens.zip all forty screens in English, Russian and Ukrainian

Install: Customize → Plugins → Add → Upload a plugin. In Claude Code this repository is its own
marketplace:

/plugin install sunoforge --marketplace sanic732/SunoForge

Already on 5.0.0? Press Update and start a new chat — an open chat keeps serving the copy it
loaded.

Still to come: packages for ChatGPT, Gemini, Antigravity and Grok, the Skill packages and
Prompt.zip. Until then, the 4.0 release
has working ones.

SunoForge v5.0.0

Choose a tag to compare

@sanic732 sanic732 released this 05 Nov 17:42

The fifth version, and the first one built around screens instead of walls of text.
Nothing here spends credits on its own: every call to a generation server asks you first.

SunoForge turns a chat assistant into a prompt engineer for music. You describe what you want in plain
words — or hand it a photo, a video or a voice memo — and get a finished prompt you can paste: Style,
lyrics markup, speech scripts, edits, repair of an old prompt, rights and post-production notes.
Fourteen reference files, one rule: never present a guess as a fact.

What is new since v4.0

The chat has screens. The menu, the Suno panel, the voice catalogue, "My songs", the take card and
the 15-minute editor are now widgets with buttons instead of walls of text. A button sends its code and
its own words — 15h 10 · My songs, 01 · Soul alto — so the chat shows what was asked, not a bare
code. Every answer without a widget of its own ends with a small one: two to four next steps and
← Menu.

Nine interface languages, checked at phone width. English, Russian, Ukrainian, German, Spanish,
French, Italian, Brazilian Portuguese and Indonesian. Every screen exists in each of them, and the
buttons are tested by script at 360 px: no clipped text, no wrapped buttons, no dead controls. Only
the labels are translated — Suno still receives English values, so styles, tags and parameters go out
exactly as the generator expects them. Language follows your chat; set lang de pins one.

Adding your own is a folder of nine screens: copy the English one under plugin/widgets/, translate
the labels, leave the codes alone, open a pull request.

A song opens its own actions. Type a row number or #id after a list of fifty, and that one song
opens with listen, analyse, edit, extend, cover, download, like and playlist already unfolded. The player
starts only when you ask for it.

Seams and joins. Seven rules for extending a track without a gap at the seam, written from sources
and marked: no outro before a seam, extend from inside the music rather than from silence, a bridge when
the genre changes, repair the seam in place instead of hiding it under a fade. What Suno does not
document is marked [UNVERIFIED] — the crossfade tag among it.

Suno Speech — narration with its own soundtrack in one take, menu [14].

LIVE — sending from the chat, menu [15]. SunoForge can hand a finished blueprint to a
music-generation MCP server your session already exposes: Suno's own beta server, ElevenLabs' hosted one,
or your own. It never spends a credit without your yes, and it reads no tool names from memory — it finds
what the session actually offers.

See the screens before installing

SunoForge_v5.0.0_screens.zip — all forty screens in English, Russian and Ukrainian. Unpack,
open index.html: it works offline, nothing is loaded from the network.

What to download

File What it is
sunoforge.plugin the plugin as one file
SHA256.txt its checksum
SunoForge_v5.0.0_screens.zip the screens, to look at before installing

Install: Customize → Plugins → Add → Upload plugin. In Claude Code this repository is its own
marketplace:

/plugin install sunoforge --marketplace sanic732/SunoForge

Already have an older build? Press Update and start a new chat — an open chat keeps serving the
copy it loaded.

This release ships the plugin only. Packages for other assistants are next: updated builds for
ChatGPT, Gemini, Antigravity and Grok, plus the Skill packages and Prompt.zip, are coming shortly.

Licence: Apache 2.0 — changed from MIT with this release. Earlier releases stay under MIT, and that
cannot be taken back. Apache asks anyone who changes a file to say so and to keep the author's notice,
which MIT did not.

What is not tested, stated plainly

SunoForge's own rule is never to present a guess as a fact, and that applies to this page too:

  • The join rules are written from sources but not tested on Suno v6. Listen to the seam.
  • The 15-minute editor is a large screen; on a slow connection it takes a while to draw.
  • Four more interface languages are not done — nine are.
  • Suno keeps changing its MCP server. What works today may need a patch release tomorrow.

If something breaks

Open an issue with what you typed, what you expected and what came back. A screenshot at your own window
width helps most — that is how the wrapped-button problems were found in the first place.

Privacy: SunoForge sends nothing anywhere by itself. See PRIVACY.md.

v4.0 — Suno v6, Lyria 3.5, two Skills and a plugin

Choose a tag to compare

@sanic732 sanic732 released this 30 Sep 19:30

A rebuild against the vendors' own pages. Suno retired every model it had and shipped v6; Google replaced Lyria 3 Pro with Lyria 3.5 and rewrote its prompt guide. Twelve files, ~457 KB, the same menu and the same rules — now as a prompt, two Skills and a plugin for Claude.

Try it without installing anything — the Gem with everything loaded: SunoForge v4.0

What to download

File What it is Where it goes
SunoForge_v4.0_Prompt.zip The prompt itself: the claude/ (XML) and gemini/ (Markdown) editions, docs, READMEs a Claude Project, a Gem, ChatGPT, Grok
SunoForge_v4.0_Skills_for_Claude.zip The Skill for Claude: SKILL.md and the twelve files claude.ai → Customize → Skills → Upload a skill (Code execution on)
SunoForge_v4.0_Skills_for_Gemini.zip The same Skill for Gemini, with the Markdown files unpack, then Gemini → Skills → Upload → the sunoforge folder
sunoforge.plugin The plugin for Claude as one file Customize → Plugins → Add → Upload plugin (paid plans)

The plugin also installs from this repository, which is its marketplace. In Claude Code:

/plugin marketplace add sanic732/SunoForge
/plugin install sunoforge@sunoforge

In chat and Claude Desktop: Customize → Plugins → Add → Add marketplace → sanic732/SunoForge.

A Skill opens files per task. A Suno prompt reads three of the twelve — about 32,000 tokens instead of ~114,000 in a Project.

Language is automatic: SunoForge answers, menu included, in the language you write in; /set lang en or /set lang ru pins one. Prompts for the generators stay English, except Lyria, which sings in the language of the prompt.

Why this version exists

Between July and September 2026 the platforms moved further than in the half-year before:

  • Suno shipped v6 on 9 September 2026 — v6, v6-wild, v6-mini — and retired every earlier model. Rights now follow the download (terms in force from 3 September 2026).
  • Google replaced Lyria 3 Pro with Lyria 3.5 and published a new prompt guide on 17 September 2026.
  • ElevenMusic moved to v2.5, MiniMax published Music 3.0 with open weights, and the Stable Audio 3.0 figures were corrected.

For the first time, rules moved too — all of them ones that name a platform:

v3.0 taught v4.0
Lyria: tempo in words only BPM and key as numbers — "at 92 BPM, in D minor" [OFFICIAL]
Lyria: single [00:15] markers ranges with section names — [0:00 - 0:10] Intro: [OFFICIAL]
MAX MODE is folklore the tags still are; Suno's real Max Mode is a separate switch in the Create form [OFFICIAL]

What is new

  • Suno v6. Max Mode and when it pays off; Variety (0 keeps your Style text as written); plain-language edits of a finished song, down to one line; covers, mashups, sampling; up to 8 minutes per generation; Studio 2.0; stem modes; download caps per plan.
  • Lyria 3.5. Google's prompt order, vocal profiles, the Lyrics: header, timestamp ranges, the Gemini app.
  • More platforms. ElevenMusic v2.5, Stable Audio 3.0 (6:20 is Medium), MiniMax Music 3.0 with open weights, ACE-Step 1.5 and YuE2 for local generation.
  • Menu [13] STEMS, STUDIO & MASTERING, and new sub-entries under [2]–[7]. The menu shows only what can be selected today; retired model names live in the repair table.
  • No prices anywhere. They differ by country and change without notice; a plan is named only where a feature or a right depends on it.
  • Legal. GEMA v Suno (Munich, 31 July 2026), the second label suit, distributor and streaming policies, EU AI Act Article 50.
  • Skills and a plugin — see above.
  • Gems are becoming skills. Google moves Gems to skills from November 2026 and carries over only supported files. An archive is not among them, so a Gem that keeps DATA.zip in its knowledge is likely to lose it — upload SunoForge_v4.0_Skills_for_Gemini.zip rather than waiting.

Compatibility

Menu numbers [1]–[12] are unchanged; [13] is added at the tail. Old prompts keep working — run them through [12]: the repair table knows retired Suno models, Lyria 3 Pro figures, single-point markers, prices, and a Style that Variety would rewrite.

Verification

  • Every platform fact rechecked against the vendor's own page — help centres, terms, API docs, a court press release — and each page kept as text.
  • A separate fact-checking pass: 38 notes. The critical and medium ones are fixed (13), including one inherited from v3.0 — Suno's lyrics editor does not place section labels by itself, you do. The minor notes are queued for the next data refresh.
  • A prompt audit against current Claude models: nine fixes — version history and "now" phrasing out of the rule files, one setting nothing read removed.
  • Twelve build invariants checked mechanically; the plugin and the marketplace pass claude plugin validate.
  • Live runs: Claude Code with the plugin loaded wrote a Suno blueprint with Variety 0, a Max Mode call and confidence marks; the Gemini Skill, once uploaded, opened the full menu from its own files.

Earlier versions

v3.0 (July 2026) stays on the releases page.


Built with P2P, an open cross-model meta-prompt framework by the same author.

v3.0 — CORE/DATA split, confidence marking, five platforms

Choose a tag to compare

@sanic732 sanic732 released this 27 Jul 06:34

A complete rewrite. Twelve files, 415 KB, replacing the ten-file v1.1 layout.

Try it without installing anything — a ready-made Gem with everything loaded: SunoForge v3.0

Works at both ends

Know nothing about music? Write what you want the way you would say it to a friend — "something sad but not depressing, with a piano, for the end of a video". You get finished prompts, ready to paste. No terminology, no menu numbers, nothing to configure first.

Know exactly what you want? The same system goes down to tempo and key, spectral brightness of a timbre, where the groove sits against the beat, stereo depth and reverb decay, slider values per genre, section-by-section performance notation, chord progressions inline, mastering thresholds and ratios. The vocabulary file alone is 42 KB of descriptors; the recipe library carries 45 presets with BPM, key and sliders already dialled in.

Neither audience has to grow into the other.

Why this version exists

Three independent fact-checking passes in July 2026 found that both previous editions were presenting community folklore in the voice of documentation. Four load-bearing "mechanics" did not survive:

Taught as fact What checking found
MAX MODE — a hidden quality mode No such mode. Controlled comparison found no difference beyond ordinary variance
Two-Minute Drift — tracks degrade after ~2 min Anecdotes only. No systematic testing, no vendor acknowledgment
[Reverb: 30%], [BPM: 120] and similar Never parsed by any covered platform
A ~200 character Style limit Wrong — it made users cut prompts to a fifth of the working range

Nothing was deleted quietly. DATA_SUNO §13 keeps every myth alongside what checking actually showed, so the system can recognise and repair them when they arrive from older guides.

What is new

Confidence marking. Every claim about platform behaviour carries [OFFICIAL], [COMMUNITY] or [UNVERIFIED]. An unmarked claim is a defect. This field has very few primary sources, and most circulating advice is one person's experience retold as fact — now the reader can tell which is which.

A CORE / DATA split. Between March and July 2026 every platform fact broke at least once: the default model version, Lyria's track length, the name of Flow Music, the Studio version, stem mechanics, the lyrics editor — and two products shut down entirely. Not one prompting rule broke. So they no longer share a file. DATA_* files carry a date in the name and an expiry notice; updating for a new quarter means replacing them and leaving CORE_* alone.

Five platforms. Suno v5.5 (Duration Slider, the Exclude Styles field, Voices, Custom Models, Studio 1.2, three stem modes), Google Lyria 3 Pro / Clip / RealTime with the official [MM:SS] timestamp format, Google Flow Music, ElevenMusic, Stable Audio 3.0.

Udio is no longer a target — downloads were disabled after the UMG settlement. It stays in the legal file as a cautionary case.

A repair engine. Menu entry [12] diagnoses an existing prompt against 14 checks and repairs dead constructs: parametric tags, MAX MODE remnants, the wrong Lyria timestamp syntax, negatives that backfire, references to retired platforms.

An explanation layer. /why <topic> returns the reasoning behind a rule — what it is, why it works, what breaks if you ignore it, and why the myth spread. Loaded on demand, so it costs nothing until asked for.

Compatibility

Menu numbers [1]–[12] are frozen as a public interface; new entries go at the tail only. Old prompts keep working — run them through [12] to see what no longer does.

Verification

Seven build invariants checked mechanically rather than by inspection: single source for version data, menu number consistency across files, no parametric constructions outside the myths sections, no cross-file duplication, confidence tag coverage, expiry headers on dated files.

Two live runs before release produced eight further fixes, including one from a first-hand measurement: 7 minutes 59 seconds on the free v4.5-all tier in a single generation — which contradicts the widely repeated claim that the free tier caps out around two minutes.

Downloads

SunoForge_v3.0_en.zip and _ru.zip hold the same twelve files and differ by one line — the language of explanations. Prompts themselves are always English, because the platforms handle it best. Switch any time with /set lang en or /set lang ru.

DATA_bundle_for_Gemini.zip is a convenience: Gemini Gems cap attachments at 10 files, so the five CORE_* files plus this one archive fit where twelve separate files do not. See docs/GEMINI.md.

Earlier versions

v1.1 (March 2026) is tagged retroactively. v2.0 was never released here — it was Russian-only and shipped on the 4PDA forum.


Built with P2P, an open cross-model meta-prompt framework by the same author.