Add LEGO mode: generate instrumental stems over references#19
Merged
Add LEGO mode: generate instrumental stems over references#19
Conversation
… docs Co-authored-by: lmangani <1423657+lmangani@users.noreply.github.com>
Copilot
AI
changed the title
[WIP] Implement LEGO mode support for acestep.cpp
Add LEGO mode to dit-vae via --lego flag
Mar 10, 2026
Copilot stopped work on behalf of
lmangani due to an error
March 10, 2026 11:22
…ate, lego) Co-authored-by: lmangani <1423657+lmangani@users.noreply.github.com>
Copilot
AI
changed the title
Add LEGO mode to dit-vae via --lego flag
examples/lego.sh: self-contained 3-step lego test pipeline
Mar 10, 2026
…io_cover_strength only Co-authored-by: lmangani <1423657+lmangani@users.noreply.github.com>
Copilot
AI
changed the title
examples/lego.sh: self-contained 3-step lego test pipeline
Fix lego.json and README: strip request to minimal required fields
Mar 10, 2026
Co-authored-by: lmangani <1423657+lmangani@users.noreply.github.com>
Copilot
AI
changed the title
Fix lego.json and README: strip request to minimal required fields
Fix lego mode: minimal request JSON and WAV output to prevent crash
Mar 10, 2026
…t root Co-authored-by: lmangani <1423657+lmangani@users.noreply.github.com>
Copilot
AI
changed the title
Fix lego mode: minimal request JSON and WAV output to prevent crash
Fix lego mode: WAV output, base model parameters, and models.sh working directory
Mar 10, 2026
lmangani
changed the title
…12% (#20) * perf: improve batch generation in step 1 by 3-12% * remove comments * remove comments
Adds --lego <track> to dit-vae for ACE-Step lego mode, which generates a new instrument stem layered over an existing source audio file. - tools/dit-vae.cpp: --lego <track> CLI flag; passes instrument name into the DiT context alongside --src-audio; requires the base model - examples/lego.json: request with caption, audio_cover_strength=1.0, inference_steps=50, guidance_scale=7.0 (base model settings) - examples/lego.sh: self-contained 3-step pipeline (download base model from project root, generate source track, apply lego guitar); models.sh is always invoked from the project root via subshell (cd .. && ./models.sh) - README.md: lego mode section with usage, available track names, and correct base-model parameters - .github/workflows/ci-build.yml: exclude mp3/ from clang-format check Co-authored-by: lmangani <1423657+lmangani@users.noreply.github.com>
Copilot
AI
changed the title
WIP: Lego mode support
Add LEGO mode: generate instrument stems over a backing track
Mar 10, 2026
Removed example JSON and bash commands from README.
lmangani
changed the title
lmangani
pushed a commit
that referenced
this pull request
Mar 10, 2026
lmangani
added a commit
that referenced
this pull request
Mar 10, 2026
* Add LEGO mode: generate instrumental stems over references (#19) * Add LEGO mode: --lego <track> flag for dit-vae, example files, README docs * Remove base model check from lego.sh Removed the echo statement for ensuring the base model. * Refactor lego.sh by removing echo statements Removed echo statements for steps in the script. * Implement error check for --lego with DiT model Add error handling for --lego option requiring base DiT model * Move lego mode from `--lego <track>` CLI flag to `"lego"` JSON request field (#21) * apply requested changes --------- Co-authored-by: Copilot <198982749+Copilot@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Implements ACE-Step LEGO mode in
dit-vae: given a source audio file, generate a new instrument stem conditioned on it via--lego <track>+--src-audio. Requires the base model (acestep-v15-base); turbo/SFT do not support it.tools/dit-vae.cpp—--lego <track>flag; builds a"Generate the {track} track based on the audio context:"instruction injected into the DiT alongside--src-audioexamples/lego.json— request preset (audio_cover_strength=1.0,inference_steps=50,guidance_scale=7.0)examples/lego.sh— end-to-end demo: download base model → generate source track → apply legoREADME.md— usage docs and full track name reference./build/dit-vae \ --src-audio backing-track.wav \ --lego guitar \ --request examples/lego.json \ --text-encoder models/Qwen3-Embedding-0.6B-Q8_0.gguf \ --dit models/acestep-v15-base-Q8_0.gguf \ --vae models/vae-BF16.gguf \ --wavAvailable tracks:
vocals,backing_vocals,drums,bass,guitar,keyboard,percussion,strings,synth,fx,brass,woodwinds