Orpheus Music Transformer
SOTA 8k multi-instrumental music transformer trained on 2.31M+ high-quality MIDIs
🔥[2026]🔥 Now featuring large optimized model!
🪬 User Guide 🪬
🤗 Duplicate 🤗
❤️ Models ❤️
🚀 Spaces 🚀
🦖 Dataset 🦖
Key Features
Efficient Architecture with RoPE
: Large optimized 748M full attention autoregressive transformer with RoPE.
Extended Sequence Length
: 8k tokens that comfortably fit most music compositions and facilitate long-term music structure generation.
Premium Training Data
: Trained solely on the highest-quality MIDIs from the Godzilla MIDI dataset.
Optimized MIDI Encoding
: Extremely efficient MIDI representation using only 3 tokens per note and 7 tokens per tri-chord.
Distinct Encoding Order
: Features a unique duration/velocity last MIDI encoding order for refined musical expression.
Full-Range Instrumental Learning
: True full-range MIDI instruments encoding enabling the model to learn each instrument separately.
Natural Composition Endings
: Outro tokens that help generate smooth and natural musical conclusions.
Best Practices Tips
Good prime seed MIDI is everything!!!
Trim the seed MIDI to exact (or at least - approximate) musical phrase.
30sec-1min (1024-1536 tokens) run time is ideal.
Remove excessive instruments. 4-5 most pronounced instruments work best.
Do not be discouraged by generated contunuations! Sometimes you need several tries to get it right!
Do not strive for perfection! Instead, try to have fun and enjoy the music!
Upload seed MIDI or select prime instruments or simply click 'Generate' button for random output
PLEASE NOTE:
Orpheus Music Transformer is a primarily continuation/co-composition model!"
The model works best if given some music context to work with
Random generation from SOS token/embeddings may not always produce good results
Input MIDI
Drop File Here
- or -
Click to Upload
Generation options
Prime instruments (select up to 5)
NOTE: Custom MIDI overrides prime instruments
Number of prime tokens
↺
16
6656
Number of tokens to generate
↺
16
1024
Model to use
Use medium model when speed is important, use large models when quality is important
Model temperature
Increase for more creative output, decrease for more repetitive output
↺
0.1
1
Model sampling top p value
1 == Disabled
↺
0.1
1
Add drums
Add an outro
Generate
Batch Previews
Batch # 0
Batch # 1
Batch # 2
Batch # 3
Batch # 4
Batch # 5
Batch # 6
Batch # 7
Batch # 8
Batch # 9
Batch # 0
Batch # 1
Batch # 2
Batch # 3
Batch # 4
Batch # 5
Batch # 6
Batch # 7
Batch # 8
Batch # 9
Batch # 0 MIDI Audio
Batch # 0 MIDI Plot
Batch # 0 MIDI File
Add/Remove Batch
Batch number to add/remove
↺
0
9
Add batch
Remove batch
Clear
Final MIDI audio
Final MIDI plot
Final MIDI file