Dataset Viewer
Auto-converted to Parquet Duplicate
text
stringlengths
0
96
Music: Eric Skiff - "Resistor Anthems" (10 tracks)
Licence: CC BY 3.0 (https://creativecommons.org/licenses/by/3.0/) - use allowed, credit required
Source: Free Music Archive, via huggingface.co/datasets/benjamin-paine/free-music-archive-full
Original site: https://ericskiff.com/music/
fzgchip_00 A Night Of Dizzy Spells
fzgchip_01 Underclocked (underunderclocked mix)
fzgchip_02 Chibi Ninja
fzgchip_03 All of Us
fzgchip_04 Come and Find Me
fzgchip_05 Searching
fzgchip_06 We're the Resistors
fzgchip_07 Ascending
fzgchip_08 Come and Find Me - B mix
fzgchip_09 Arpanauts
fzgchip, chiptune, 8-bit, instrumental, square wave lead, arpeggios, retro video game music
fzgchip, chiptune, 8-bit, instrumental, square wave lead, arpeggios, retro video game music
fzgchip, chiptune, 8-bit, instrumental, square wave lead, arpeggios, retro video game music
fzgchip, chiptune, 8-bit, instrumental, square wave lead, arpeggios, retro video game music
fzgchip, chiptune, 8-bit, instrumental, square wave lead, arpeggios, retro video game music
fzgchip, chiptune, 8-bit, instrumental, square wave lead, arpeggios, retro video game music
fzgchip, chiptune, 8-bit, instrumental, square wave lead, arpeggios, retro video game music
fzgchip, chiptune, 8-bit, instrumental, square wave lead, arpeggios, retro video game music
fzgchip, chiptune, 8-bit, instrumental, square wave lead, arpeggios, retro video game music
fzgchip, chiptune, 8-bit, instrumental, square wave lead, arpeggios, retro video game music

Fizgig ACE-Step Chiptune Starter

A ready-to-train music dataset for your first ACE-Step 1.5 XL LoRA in Fizgig, and the LoRA it made, so you can hear where you're heading before you start.

  • songs/: 10 instrumental chiptune tracks (2 to 3.5 minutes each, one album) with a caption beside each.
  • fzgchip_acestep15xl_e800_ema.safetensors: the LoRA from this dataset, epoch 800, trained with EMA 0.98. Works really well at strength 0.5. Try anything between 0.5 and 1, but it tends to work better towards 0.5.

Use the LoRA

Load it on ACE-Step 1.5 XL Turbo in ComfyUI or Fizgig at around strength 0.5 and start your caption with the trigger word fzgchip:

fzgchip, chiptune, 8-bit, upbeat, fast tempo, square wave lead, rapid arpeggios, retro video game music

Towards 1 it leans closer to the training songs; around 0.5 it writes new music in their sound. Add lyrics and a voice to the caption for songs, e.g. robotic vocals, vocoder voice.

Train your own

  1. Download songs/ and pick it as the training folder on Fizgig's Start tab.
  2. Choose ACE-Step 1.5 XL (music) and the ✨ ACE-Step Official LoRA (rank 64, 800 epochs) preset.
  3. Start. The preset trains with EMA 0.98 and saves a checkpoint and a preview every 40 epochs. The whole 800-epoch run takes about 2 hours on an RTX 5090, previews included.

Expect the previews to sound really bad for a while: as the model moves from what it started with towards your style, the quality drops. It can sound like overtraining, but it isn't. Once it has moved over to the style it starts to improve again, and with this dataset the best checkpoints came between epochs 740 and 800. Compare late checkpoints at strength 0.5.

Each caption is fzgchip, chiptune, 8-bit, instrumental, square wave lead, arpeggios, retro video game music. To train on your own music, copy the layout: one audio file and one caption .txt per song, the same trigger word in every caption.

Credits

Music: Eric Skiff, Resistor Anthems (ericskiff.com/music), licensed CC BY 3.0, from the Free Music Archive. Credit Eric Skiff wherever you share these songs or music made with this LoRA. Track list in CREDITS.txt.

Downloads last month
119