MiniMax Music 3 training tournament round 1

A central comparison of MiniMax Music 3 language-model LoRA tournament round 1 runs. The previous campaigns are preserved below, followed by the Caption v2 training-window and singer-tag comparisons. Tournament and prompt sections are collapsed so individual audio comparisons can be opened as needed.

The card stores 2,144 FLAC renders directly in this repository. Each comparison holds its lyrics, caption conditions, LoRA-strength sweep, and random stream constant within that tournament.

Tournament overview

Machine-readable comparison manifest

Previous tournament: original objective comparison (256 renders)

The original 1,000-step Vanilla, XM, NextLat, and NextLat + XM comparison.

All renders use a 40-second maximum duration, 30 flow steps, flow CFG 1.7, and seed 20260822. MiniMax Music 3 may end a song before the duration cap.

Training legs

The source model repos were consolidated into this repository; audio renders are stored below each comparison.

• Vanilla: soad-mm3-vanilla-20260822
• Explorative Modeling (XM): soad-mm3-xm-20260822
• NextLat: soad-mm3-nextlat-20260822
• NextLat + XM: soad-mm3-nextlat-xm-20260822

No trigger word / off-genre descriptors

Caption: Alternative pop-jazz, 102 BPM, brushed drums, upright bass, clean electric piano, airy male vocals, smooth verse-chorus songwriting, restrained dynamics, polished intimate mix.

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00

Trigger word only

Caption: system of a down, soad style music

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00

Matching descriptors, no trigger word

Caption: Lilting hypnotic arpeggios, a gentle melodic mid-tempo groove with layered harmonies, sudden explosive distorted riffs, unstable meter changes, theatrical male vocals, tightly syncopated bass and drums, abrupt shifts between fragile verses and frantic heavy choruses.

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00

Trigger word + matching descriptors

Caption: system of a down, soad style music, lilting hypnotic arpeggios, a gentle melodic mid-tempo groove with layered harmonies, sudden explosive distorted riffs, unstable meter changes, theatrical male vocals, tightly syncopated bass and drums, abrupt shifts between fragile verses and frantic heavy choruses.

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00 Full song without regularisation (576 renders)

AdamW 8e-5, full-song training windows, no caption dropout, and no regularisation set.

All renders use a 40-second maximum duration, 30 flow steps, flow CFG 1.7, and seed 20260822. MiniMax Music 3 may end a song before the duration cap.

Training legs

• Vanilla: soad-mm3-vanilla-20260823-10k-adamw8e-5-fullsong-nocapdrop
• Explorative Modeling (XM): soad-mm3-xm-20260823-10k-adamw8e-5-fullsong-nocapdrop
• NextLat: soad-mm3-nextlat-20260823-10k-adamw8e-5-fullsong-nocapdrop
• NextLat + XM: soad-mm3-nextlat-xm-20260823-10k-adamw8e-5-fullsong-nocapdrop

Run note: These four runs resumed their learned LoRA weights from checkpoint 500 with fresh optimizer, sampler, and RNG state after the original local trainer state was lost.

No trigger word / off-genre descriptors

Caption: Alternative pop-jazz, 102 BPM, brushed drums, upright bass, clean electric piano, airy male vocals, smooth verse-chorus songwriting, restrained dynamics, polished intimate mix.

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00

Trigger word only

Caption: system of a down, soad style music

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00

Matching descriptors, no trigger word

Caption: Lilting hypnotic arpeggios, a gentle melodic mid-tempo groove with layered harmonies, sudden explosive distorted riffs, unstable meter changes, theatrical male vocals, tightly syncopated bass and drums, abrupt shifts between fragile verses and frantic heavy choruses.

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00

Trigger word + matching descriptors

Caption: system of a down, soad style music, lilting hypnotic arpeggios, a gentle melodic mid-tempo groove with layered harmonies, sudden explosive distorted riffs, unstable meter changes, theatrical male vocals, tightly syncopated bass and drums, abrupt shifts between fragile verses and frantic heavy choruses.

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00 Random windowing + regularisation (432 renders)

AdamW 8e-5, random 128-frame windows, 50% parent-teacher regularisation, and no caption dropout.

All renders use a 40-second maximum duration, 30 flow steps, flow CFG 1.7, and seed 20260822. MiniMax Music 3 may end a song before the duration cap.

Training legs

• Vanilla: soad-mm3-vanilla-20260823-10k-adamw8e-5-randwin128-reg128-nocapdrop
• NextLat: soad-mm3-nextlat-20260823-10k-adamw8e-5-randwin128-reg128-nocapdrop
• NextLat + XM: soad-mm3-nextlat-xm-20260823-10k-adamw8e-5-randwin128-reg128-nocapdrop

No trigger word / off-genre descriptors

Caption: Alternative pop-jazz, 102 BPM, brushed drums, upright bass, clean electric piano, airy male vocals, smooth verse-chorus songwriting, restrained dynamics, polished intimate mix.

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00

Trigger word only

Caption: system of a down, soad style music

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00

Matching descriptors, no trigger word

Caption: Lilting hypnotic arpeggios, a gentle melodic mid-tempo groove with layered harmonies, sudden explosive distorted riffs, unstable meter changes, theatrical male vocals, tightly syncopated bass and drums, abrupt shifts between fragile verses and frantic heavy choruses.

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00

Trigger word + matching descriptors

Caption: system of a down, soad style music, lilting hypnotic arpeggios, a gentle melodic mid-tempo groove with layered harmonies, sudden explosive distorted riffs, unstable meter changes, theatrical male vocals, tightly syncopated bass and drums, abrupt shifts between fragile verses and frantic heavy choruses.

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00 Caption v2: training-window comparison (400 renders)

AdamW 8e-5 and no regularisation, comparing Vanilla and NextLat + XM with random 128-frame and full-song windows.

All renders use a 40-second maximum duration, 30 flow steps, flow CFG 1.7, and seed 20260822. MiniMax Music 3 may end a song before the duration cap.

Training legs

• Vanilla, random window: soad-mm3-vanilla-randwin128-20260823-5k-adamw8e-5-capv2-noreg
• NextLat + XM, random window: soad-mm3-nextlat-xm-randwin128-20260823-5k-adamw8e-5-capv2-noreg
• Vanilla, full song: soad-mm3-vanilla-fullsong-20260823-5k-adamw8e-5-capv2-noreg
• NextLat + XM, full song: soad-mm3-nextlat-xm-fullsong-20260823-5k-adamw8e-5-capv2-noreg

Run note: Checkpoint coverage differs by leg; empty table cells were not rendered.

No trigger word / off-genre descriptors

Caption: Alternative pop-jazz, 102 BPM, brushed drums, upright bass, clean electric piano, airy male vocals, smooth verse-chorus songwriting, restrained dynamics, polished intimate mix.

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00

Trigger word only

Caption: system of a down, soad style music

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00

Matching descriptors, no trigger word

Caption: Lilting hypnotic arpeggios, a gentle melodic mid-tempo groove with layered harmonies, sudden explosive distorted riffs, unstable meter changes, theatrical male vocals, tightly syncopated bass and drums, abrupt shifts between fragile verses and frantic heavy choruses.

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00

Trigger word + matching descriptors

Caption: system of a down, soad style music, lilting hypnotic arpeggios, a gentle melodic mid-tempo groove with layered harmonies, sudden explosive distorted riffs, unstable meter changes, theatrical male vocals, tightly syncopated bass and drums, abrupt shifts between fragile verses and frantic heavy choruses.

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00 Caption v2: singer-tag comparison (480 renders)

The same four Caption v2 legs evaluated with explicit Serj and Daron section assignments.

All renders use a 40-second maximum duration, 30 flow steps, flow CFG 1.7, and seed 20260822. MiniMax Music 3 may end a song before the duration cap.

Training legs

• Vanilla, random window: soad-mm3-vanilla-randwin128-20260823-5k-adamw8e-5-capv2-noreg
• NextLat + XM, random window: soad-mm3-nextlat-xm-randwin128-20260823-5k-adamw8e-5-capv2-noreg
• Vanilla, full song: soad-mm3-vanilla-fullsong-20260823-5k-adamw8e-5-capv2-noreg
• NextLat + XM, full song: soad-mm3-nextlat-xm-fullsong-20260823-5k-adamw8e-5-capv2-noreg

Run note: Checkpoint coverage differs by leg; empty table cells were not rendered.

No trigger word / off-genre descriptors

Caption: Alternative pop-jazz, 102 BPM, brushed drums, upright bass, clean electric piano, airy male vocals, smooth verse-chorus songwriting, restrained dynamics, polished intimate mix.

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00

Trigger word only

Caption: system of a down, soad style music

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00

Matching descriptors, no trigger word

Caption: Lilting hypnotic arpeggios, a gentle melodic mid-tempo groove with layered harmonies, sudden explosive distorted riffs, unstable meter changes, theatrical male vocals, tightly syncopated bass and drums, abrupt shifts between fragile verses and frantic heavy choruses.

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00

Trigger word + matching descriptors

Caption: system of a down, soad style music, lilting hypnotic arpeggios, a gentle melodic mid-tempo groove with layered harmonies, sudden explosive distorted riffs, unstable meter changes, theatrical male vocals, tightly syncopated bass and drums, abrupt shifts between fragile verses and frantic heavy choruses.

LoRA strength 0.25 LoRA strength 0.50 LoRA strength 0.75 LoRA strength 1.00

Evaluation lyrics

The original comparison lyrics are shared by the first four tournaments. The latest singer-tag comparison uses its own section-assigned lyric sheet.

Show lyrics

[Intro]
Count the teeth in the traffic light
Paint the silence ultraviolet

[Verse 1]
Paper crowns on the factory floor
Everybody wants a little more
Wind-up prophets in a checkout line
Selling borrowed seconds back as time

[Pre-Chorus]
Turn the dial, divide the sky
Ask the wires to tell us why

[Chorus]
Wake up the walls, let the ceiling shake
We built a feast from a plastic lake
Hands in the air while the engines stall
Who signed the order to swallow it all?

[Verse 2]
Velvet sirens in the avenue
Paint every warning a pleasant blue
I fed my shadow to the voting machine
It came back polished, quiet, and clean

[Bridge]
Softly, softly, count to four
Kick down the arithmetic door
One for the hunger, two for the show
Three for the ones who already know

[Final Chorus]
Wake up the walls, let the ceiling shake
We built a feast from a plastic lake
Hands in the air while the engines stall
Who signed the order to swallow it all?

[Outro]
Count the teeth
Kill the light
Carry the silence
Into the night

Show singer-tag comparison lyrics

[Verse 1: Serj Tankian]
Why don't you ask the kids at Tiananmen Square?
Was fashion the reason why they were there?
They disguise it, hypnotize it
Television made you buy it

[Chorus: Daron Malakian]
I'm just sitting in my car and
Waiting for my

[Verse 2: Daron Malakian]
She is scared that I will take her away from there
Dreams and her country left with no one there
Mesmerized, the signal bends the truth
The broadcast paints a blindfold on the youth

[Chorus: Serj Tankian]
I'm still waiting by the dashboard
Watching for the signal
I'm still waiting by the dashboard
Waiting for my girl

[Instrumental Break]

[Chorus: Daron Malakian]
I'm still waiting by the dashboard
Waiting for my girl
I'm still waiting by the dashboard
Waiting for my girl

Inference notes

The four LoRA strengths were evaluated in one batched pass with a shared random stream. NextLat's predictor and XM's route embeddings are auxiliary training modules and were not used at inference. The generated audio remains hosted in the source model repositories linked above.