A lora model trained on the Engram dataset, 5 songs split, converted, captioned and trained.
use the genres.txt and instruments.txt for the identifiers - wedgeewoo/Riffusion-Textual-Inversion-template: Templates for musical textual inversion for riffusion (github.com)
Example - Stream Sprouts #11 by Wedgeewoo | Listen online for free on SoundCloud
Tips and Tricks - Use a weight of .8, the results come out much more cohesive when using the genres.txt identifiers.
Example 2 - Stream Sprouts #14 by Wedgeewoo | Listen online for free on SoundCloud
Prompt -
Complextro Tropical house song , reese bass, neuro bass, tuba, drop <lora:Engram:.75>
Steps: 30, Sampler: Euler a, CFG scale: 8.5, Seed: 3779202002, Size: 512x512, Model hash: 99a6eb51c1, Variation seed: 1752705851, Variation seed strength: 0.25, Tile X: True, Tile Y: False, Start Tiling From Step: 15, Stop Tiling After Step: -1, Wildcard prompt: "__genres__, reese bass, neuro bass, tuba, drop <lora:Engram:.75>"
Wedgewoo - Engram : LoRa model for Riffusion. - YouTube