Model reference · open weights

acestep-captioner

Audio ACE-Step Music / audio 1 build Open weights 1k dl/mo

acestep-captioner is an open-weight audio or speech model from ACE-Step. acestep-captioner (BF16) weighs 22.4 GB; the smallest configuration that runs it is RTX 5090 32 GB.

  • ACE-Step Captioner is a 10.7B parameter model developed by ACE-Step for text-to-audio tasks.
  • It generates detailed, structured descriptions of audio content, including musical style, instrument recognition, and structural analysis.
  • The model is released under the MIT license.

Summary of the ACE-Step/acestep-captioner model card, 2026-10-01

What it is

Released byACE-Step
Released2026-01-23
Parameters10.7B
VRAM22.4 GB for the weights

What it runs on

Memory and cards for acestep-captioner (BF16)

22.4 GBweights, file size
1.6 GBruntime overhead
CardOne streamMemory
RTX 3060 12 GB … RTX 4090 24 GBdoes not fit
RTX 5090 32 GBfits31.0 GB
L40S 48 GBfits44.0 GB
A100 80 GBfits78.2 GB
H100 80 GBfits78.1 GB
RTX PRO 6000 Blackwell 96 GBfits93.8 GB
DGX Spark (GB10) 128 GB unifiedfits107 GB
H200 141 GBfits138 GB
B200 180 GBfits176 GB

From the model card

What ACE-Step says about acestep-captioner

Read the model card

Description

ACE-Step Captioner is the annotation model used by ACE-Step v1.5 for training data labeling. It is a professional-grade music captioning model that generates detailed, structured descriptions of audio content.

Performance

🏆 Accuracy surpasses Gemini Pro 2.5 in music description tasks

Key Features

  • 🎼 Musical Style Analysis - Identifies genres, sub-genres, and stylistic influences
  • 🎸 Instrument Recognition - Detects and describes 1000+ instrument types and combinations
  • 🎭 Structure & Progression - Analyzes musical arrangement including intro, verse, chorus, bridge, climax, and outro
  • 🔊 Timbre Description - Captures tonal qualities, textures, and sonic characteristics
  • 📝 Rich Vocabulary - Supports 1000+ descriptive terms for comprehensive music annotation

Usage

The usage is the same as Qwen2.5 Omni-7B.

Prompt Format

Use the following prompt to caption audio:

*Task* Describe this audio in detail

Output Format

The model generates natural language descriptions covering multiple aspects of the music.

Example Output

A melancholic indie folk track featuring fingerpicked acoustic guitar
as the primary instrument. The song opens with a sparse, contemplative
intro before the vocals enter with a breathy, intimate delivery.
The arrangement gradually builds through the verse, adding subtle
string pads and a gentle kick drum. The chorus lifts with layered
harmonies and a warmer, fuller texture. The bridge introduces a
key change and emotional climax before returning to the stripped-down
acoustic arrangement for the outro.

Descriptive Capabilities

Musical Styles (Examples)

CategoryStyles
ElectronicAmbient, Techno, House, Drum & Bass, Synthwave, IDM, Downtempo
RockAlternative, Indie, Post-Rock, Progressive, Psychedelic, Grunge
PopSynth-pop, Electropop, Dream Pop, Art Pop, Indie Pop
ClassicalOrchestral, Chamber, Minimalist, Neo-Classical, Cinematic
WorldLatin, African, Middle Eastern, Asian Traditional, Celtic
JazzFusion, Smooth, Bebop, Modal, Free Jazz
Hip-HopTrap, Boom Bap, Lo-fi, Instrumental, Cloud Rap

Instruments (1000+ Supported)

CategoryExamples
StringsAcoustic Guitar, Electric Guitar, Violin, Cello, Bass, Harp, Mandolin
KeysPiano, Synthesizer, Organ, Rhodes, Wurlitzer, Mellotron
PercussionDrums, Electronic Drums, Congas, Bongos, Timpani, Vibraphone
WindSaxophone, Trumpet, Flute, Clarinet, Oboe, French Horn
ElectronicSynth Bass, Pad, Lead, Arpeggiator, Sampler, 808, 303

Structure Analysis

  • Intro / Outro - Opening and closing sections
  • Verse / Pre-Chorus / Chorus - Main song structure
  • Bridge / Break - Transitional sections
  • Build-up / Drop / Climax - Dynamic progression
  • Interlude / Solo - Instrumental passages

Timbre Descriptions

DimensionDescriptors
TextureWarm, Bright, Dark, Crisp, Muddy, Clean, Distorted, Saturated
SpaceReverberant, Dry, Spacious, Intimate, Cavernous, Tight
DynamicsPunchy, Soft, Aggressive, Gentle, Compressed, Dynamic
CharacterEthereal, Gritty, Smooth, Raw, Polished, Organic, Synthetic

Use Cases

  • Music AI Training - Generate high-quality captions for music generation models
  • Music Information Retrieval - Create searchable metadata for audio databases
  • Content Moderation - Analyze and categorize music content
  • Music Education - Provide detailed analysis for learning purposes
  • Audio Production - Document and describe sound design elements

Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.

Running it yourself

Run it on a rented GPU

Rent a machine by the hour. How to run this model is on its model card.

© 2026 AxForge · EU-hosted AI infrastructure Pricing Docs Trust Privacy Terms