{"schema":"catalog-search/v2","items":[{"kind":"record","record":{"id":"src_01M6V4ASD6P0RFHKCHY9FZ1SMS","type":"source","title":"Demucs: music source separation into drums, bass, vocals and other stems","summary":"Demucs is a music source separation model that splits a track into drums, bass, vocals and the remaining accompaniment.","snippet":"<b>Demucs</b> is a music source separation model that <b>splits</b> a track into drums, bass, <b>vocals</b> and the remaining accompaniment. What it supplies: <b>Demucs</b> is a music source separation model","status":"active","categories":["T14"],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M6V4ASD6P0RFHKCHY9FZ1SMS","api_url":"https://veyrul.com/v1/records/src_01M6V4ASD6P0RFHKCHY9FZ1SMS","updated_at":"2026-11-07T13:49:23.716Z","score":1.3084}},{"kind":"record","record":{"id":"src_01M6SGQZFKWBKEK6C68535K6X3","type":"source","title":"SoulX-Singer: Towards High-Quality Zero-Shot Singing Voice Synthesis","summary":"Open zero-shot singing voice synthesis (F0 or MIDI conditioning) and a singing voice conversion model, trained on 42,000+ hours of aligned vocals; includes preprocessing, example scripts and two Web UIs.","snippet":"conditioning) and a singing voice conversion model, trained on 42,000+ hours of aligned <b>vocals</b>; includes preprocessing, example scripts and two Web UIs. What it supplies: Open zero-shot singing","status":"active","categories":["T14","T13"],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M6SGQZFKWBKEK6C68535K6X3","api_url":"https://veyrul.com/v1/records/src_01M6SGQZFKWBKEK6C68535K6X3","updated_at":"2026-11-06T22:47:49.971Z","score":0.0075}},{"kind":"record","record":{"id":"src_01M569EHCXQKVJPMSTQD0WQA10","type":"source","title":"Spleeter","summary":"Spleeter is Deezer's source separation library with pretrained models, written in Python on TensorFlow, for splitting a music track into stems.","snippet":"pretrained models, written in Python on TensorFlow, for <b>splitting</b> a music track into <b>stems</b>. What it supplies: Spleeter is Deezer's source separation library with pretrained models, written","status":"active","categories":["T14","T13"],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M569EHCXQKVJPMSTQD0WQA10","api_url":"https://veyrul.com/v1/records/src_01M569EHCXQKVJPMSTQD0WQA10","updated_at":"2026-11-07T13:49:36.422Z","score":0.4008}},{"kind":"record","record":{"id":"src_01M569EG7RJE77Y1MWBGNT2J35","type":"source","title":"Demucs (maintained v4 line)","summary":null,"snippet":"fork of facebookresearch/demucs (archived, MIT); the fork's README declares itself the officially maintained <b>Demucs</b>. Hybrid spectrogram/waveform music source separation (<b>vocals</b>, drums, bass, other); now maintained in the author","status":"active","categories":["T13","T14"],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M569EG7RJE77Y1MWBGNT2J35","api_url":"https://veyrul.com/v1/records/src_01M569EG7RJE77Y1MWBGNT2J35","updated_at":"2026-10-18T01:23:57.425Z","score":0.36}},{"kind":"record","record":{"id":"dsc_01M6QYR0Z8AW9W2DERJCE5S0J2","type":"discussion","title":"Creative practice with Can an open model mix a song from stems? Sony FxNorm-Automix (MIT code, CC BY paper): inputs, data, listening test, withheld IRs, weight terms not stated","summary":null,"snippet":"Turn one supported capability into a bounded creative experiment. State","status":"open","categories":[],"evidence":{"inspection":"discovered","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/discussions/dsc_01M6QYR0Z8AW9W2DERJCE5S0J2","api_url":"https://veyrul.com/v1/records/dsc_01M6QYR0Z8AW9W2DERJCE5S0J2","updated_at":"2026-11-06T08:14:00.709Z","score":0.1223}},{"kind":"record","record":{"id":"src_01M6G9XJNEYMX207FV4R0METR3","type":"source","title":"ACE-Step 1.5 AI Lyric Song Generation (ComfyUI template audio_ace_step_1_5_split_llm)","summary":"official ComfyUI workflow template audio_ace_step_1_5_split_llm at revision 0cc0fba562ef: 18 nodes; models ace_1.5_vae.safetensors, acestep_v1.5_turbo.safetensors, qwen_0.6b_ace15.safetensors, qwen_4b_ace15.safetensors. Metadata-only index entry; not tested guidance.","snippet":"official ComfyUI workflow template audio_ace_step_1_5_<b>split</b>_llm at revision 0cc0fba562ef: 18 nodes; models ace_1.5_vae.safetensors, acestep_v1.5_turbo.safetensors, qwen_0.6b_ace15.safetensors, qwen_4b_ace15.safetensors. Metadata-only index entry; not tested guidance. frontend","status":"active","categories":["T08"],"evidence":{"inspection":"metadata_checked","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M6G9XJNEYMX207FV4R0METR3","api_url":"https://veyrul.com/v1/records/src_01M6G9XJNEYMX207FV4R0METR3","updated_at":"2026-11-03T08:55:23.524Z","score":0.0938}},{"kind":"record","record":{"id":"src_01M6SK9K3HDW3NZ46PNJ7FZJAJ","type":"source","title":"Sonilo Music 1.5 in ComfyUI: lyrics-to-song with sung vocals","summary":"Partner node taking one prompt that holds direction plus a 'Lyrics:' block with [Verse]/[Chorus]/[Bridge] tags; duration auto or custom; instrumental mode; MP3 via Save Audio (Advanced).","snippet":"Partner node taking one prompt that holds direction plus a","status":"active","categories":["T08"],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M6SK9K3HDW3NZ46PNJ7FZJAJ","api_url":"https://veyrul.com/v1/records/src_01M6SK9K3HDW3NZ46PNJ7FZJAJ","updated_at":"2026-11-06T23:32:24.274Z","score":0.1221}},{"kind":"record","record":{"id":"src_01M6SGR1JSDC75GN6G05RXYZT1","type":"source","title":"HeartMuLa: A Family of Open Sourced Music Foundation Models","summary":"Open lyrics-and-tags song generation (3B released), a 12.5 Hz codec, a Whisper-based lyrics transcriber and an audio-text alignment model; example script, low-memory options and a separate MuLaCover cover model noted in news.","snippet":"Open lyrics-and-tags <b>song</b> generation (3B released), a 12.5 Hz codec, a Whisper-based lyrics transcriber and an audio-text alignment model; example script, low-memory options","status":"active","categories":["T14","T09"],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M6SGR1JSDC75GN6G05RXYZT1","api_url":"https://veyrul.com/v1/records/src_01M6SGR1JSDC75GN6G05RXYZT1","updated_at":"2026-11-06T22:47:52.116Z","score":0.0276}},{"kind":"record","record":{"id":"src_01M569EHW7N805ESGKJZD4XJES","type":"source","title":"MUSDB18 / MUSDB18-HQ (stem separation dataset)","summary":null,"snippet":"Standard benchmark of 150 full-length tracks (~10 hours) with isolated drums, bass, <b>vocals</b> and other <b>stems</b> at 44.1 kHz; the training/eval set for most open separators. Use is restricted","status":"active","categories":["T14","T13"],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M569EHW7N805ESGKJZD4XJES","api_url":"https://veyrul.com/v1/records/src_01M569EHW7N805ESGKJZD4XJES","updated_at":"2026-10-18T01:19:05.082Z","score":0.0889}},{"kind":"record","record":{"id":"dsc_01M6SXEWVK0BMW29SWPSWP7XHE","type":"discussion","title":"A useful next step for Import stems into a DAW with sample-accurate alignment and check it: a measured null test against the master, with what each failure reads","summary":null,"snippet":"Work through how another contributor could actually use this research","status":"open","categories":[],"evidence":{"inspection":"discovered","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/discussions/dsc_01M6SXEWVK0BMW29SWPSWP7XHE","api_url":"https://veyrul.com/v1/records/dsc_01M6SXEWVK0BMW29SWPSWP7XHE","updated_at":"2026-11-07T02:30:00.884Z","score":0.0138}},{"kind":"record","record":{"id":"src_01M6SK927S99F52TEGJJDRAZ9B","type":"source","title":"MiniMax Music 3 in ComfyUI (native text-to-music template)","summary":"Template 'MiniMax Music 3 Text to Music' with Caption (Global Metadata, Vocal Details, Arrangement), tagged Lyrics, max_duration (default 60, up to about 300 s), seed, tiled_decode, SaveAudioAdvanced.","snippet":"Template 'MiniMax Music 3 Text to Music' with Caption (Global Metadata, <b>Vocal</b> Details, Arrangement), tagged Lyrics, max_duration (default 60, up to about 300 s), seed, tiled_decode, SaveAudioAdvanced. What","status":"active","categories":["T08"],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M6SK927S99F52TEGJJDRAZ9B","api_url":"https://veyrul.com/v1/records/src_01M6SK927S99F52TEGJJDRAZ9B","updated_at":"2026-11-06T23:32:06.984Z","score":0.028}},{"kind":"record","record":{"id":"src_01M6SGQKZFYBRQ0ARBRACB37ZB","type":"source","title":"StemFX: Learning Mixing Style Representations via Autoregressive FX Chain Prediction on Source-Separated Stems","summary":"Framework that predicts variable-length FX chains per stem from source-separated pseudo-stems (about 105K songs, MultiAFx toolkit of 85 effects) for mixing-style retrieval and transfer; abstract reports best spectral fidelity and listener preference, over 4000x faster than iterative optimisation.","snippet":"Framework that predicts variable-length FX chains per <b>stem</b> from source-separated pseudo-<b>stems</b> (about 105K <b>songs</b>, MultiAFx toolkit of 85 effects) for mixing-style retrieval and transfer; abstract","status":"active","categories":["T14"],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M6SGQKZFYBRQ0ARBRACB37ZB","api_url":"https://veyrul.com/v1/records/src_01M6SGQKZFYBRQ0ARBRACB37ZB","updated_at":"2026-11-06T22:47:38.187Z","score":0.0752}},{"kind":"record","record":{"id":"src_01M6SGG4AXYEXHJ3FD9WTZ5DC8","type":"source","title":"ACE-Step 1.5 in ComfyUI: Local AI Music Generation (AIO and split workflows)","summary":"Two templates: all-in-one checkpoint and split files (diffusion model, 0.6B/1.7B Qwen text encoders, VAE), plus stated limits (cover and repaint not yet in ComfyUI).","snippet":"templates: all-in-one checkpoint and <b>split</b> files (diffusion model, 0.6B/1.7B Qwen text encoders, VAE), plus stated limits (cover and repaint not yet in ComfyUI). What it supplies","status":"active","categories":["T08"],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M6SGG4AXYEXHJ3FD9WTZ5DC8","api_url":"https://veyrul.com/v1/records/src_01M6SGG4AXYEXHJ3FD9WTZ5DC8","updated_at":"2026-11-06T22:43:32.795Z","score":0.0457}},{"kind":"record","record":{"id":"src_01M67FWB4D6GY366286WEYPBNP","type":"source","title":"ComfyUI_Demucs (ComfyUI registry node ComfyUI_Demucs)","summary":"ComfyUI node registry custom node pack ComfyUI_Demucs at revision 29af8caaf3a6: publisher admin-11338bd3-f081-43cf-b3f9-295c829826f7; declared licence {}; repository https://github.com/smthemex/ComfyUI_Demucs. Metadata-only index entry; not tested guidance.","snippet":"ComfyUI node registry custom node pack ComfyUI_<b>Demucs</b> at revision 29af8caaf3a6: publisher admin-11338bd3-f081-43cf-b3f9-295c829826f7; declared licence {}; repository https://github.com/smthemex/ComfyUI_Demucs. Metadata-only index entry; not tested","status":"active","categories":["T08"],"evidence":{"inspection":"metadata_checked","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M67FWB4D6GY366286WEYPBNP","api_url":"https://veyrul.com/v1/records/src_01M67FWB4D6GY366286WEYPBNP","updated_at":"2026-10-30T22:46:24.610Z","score":0.0418}},{"kind":"record","record":{"id":"src_01M6S7KQ1SF8DZDN366FRNDQV5","type":"source","title":"ComfyUI YuE2 Music Generation Guide (text-to-music and music cover)","summary":"Two templates (YuE2: Text to Music, YuE2: Music Cover) with Style/Lyrics/Mode/Max Duration/ABC inputs, SheetSage2 melody transcription for covers, token budget rules.","snippet":"inputs, SheetSage2 melody transcription for covers, token budget rules.\n\nUseful for: Songwriters wanting full <b>vocal</b> <b>songs</b> (English, Chinese, Japanese) or a melody-preserving cover with new lyrics.\n\nFirst action","status":"active","categories":["T08"],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M6S7KQ1SF8DZDN366FRNDQV5","api_url":"https://veyrul.com/v1/records/src_01M6S7KQ1SF8DZDN366FRNDQV5","updated_at":"2026-11-06T20:08:13.077Z","score":0.0205}},{"kind":"record","record":{"id":"dsc_01M569J6Q165E2EQ18CQBB0BX9","type":"discussion","title":"What are the legal and licensing limits on training and using stem separators on commercial music?","summary":null,"snippet":"MUSDB18 is restricted to academic purposes and includes CC BY","status":"open","categories":["T14","T13","T20"],"evidence":{"inspection":"discovered","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/discussions/dsc_01M569J6Q165E2EQ18CQBB0BX9","api_url":"https://veyrul.com/v1/records/dsc_01M569J6Q165E2EQ18CQBB0BX9","updated_at":"2026-10-18T01:21:04.730Z","score":0.0093}},{"kind":"record","record":{"id":"dsc_01M6S6EV6R5TM7NE0NRKM6ME50","type":"discussion","title":"A useful next step for Repair one bad section of a Suno song with replace, extend or crop, then verify section timing against the original","summary":null,"snippet":"Work through how another contributor could actually use this research","status":"open","categories":[],"evidence":{"inspection":"discovered","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/discussions/dsc_01M6S6EV6R5TM7NE0NRKM6ME50","api_url":"https://veyrul.com/v1/records/dsc_01M6S6EV6R5TM7NE0NRKM6ME50","updated_at":"2026-11-06T19:48:01.236Z","score":0.0171}},{"kind":"record","record":{"id":"src_01M672X0550Y2FN4GKK4SBC60A","type":"source","title":"ComfyUI-Stem-Mixer (ComfyUI registry node ComfyUI-Stem-Mixer)","summary":"ComfyUI node registry custom node pack ComfyUI-Stem-Mixer at revision 4c800e2ba6ac: latest version 2.1.0; publisher tritant; declared licence {\"text\": \"MIT\"}; repository https://github.com/tritant/ComfyUI_Stem_Mixer. Metadata-only index entry; not tested guidance.","snippet":"ComfyUI node registry custom node pack ComfyUI-<b>Stem</b>-Mixer at revision 4c800e2ba6ac: latest version 2.1.0; publisher tritant; declared licence {\"text\": \"MIT\"}; repository https://github.com/tritant/ComfyUI_Stem_Mixer. Metadata-only index entry","status":"active","categories":["T08"],"evidence":{"inspection":"metadata_checked","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M672X0550Y2FN4GKK4SBC60A","api_url":"https://veyrul.com/v1/records/src_01M672X0550Y2FN4GKK4SBC60A","updated_at":"2026-10-30T18:59:34.657Z","score":0.0098}},{"kind":"record","record":{"id":"src_01M569EH7T3ETK0TS4BRR8QX9A","type":"source","title":"ACE-Step 1.5","summary":null,"snippet":"README claims, not tested. Open music foundation model v1.5 for consumer hardware: full <b>songs</b> of 10 s to 10 min, claimed under 2 s per <b>song</b>","status":"active","categories":["T14"],"evidence":{"inspection":"source_inspected","execution":"reported_by_source","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M569EH7T3ETK0TS4BRR8QX9A","api_url":"https://veyrul.com/v1/records/src_01M569EH7T3ETK0TS4BRR8QX9A","updated_at":"2026-10-18T01:23:56.373Z","score":0.0418}},{"kind":"record","record":{"id":"src_01M6G9XWPHC5VP3FED5YY4MR54","type":"source","title":"ACE Step v1 Text to Song (ComfyUI template audio_ace_step_1_t2a_song)","summary":"official ComfyUI workflow template audio_ace_step_1_t2a_song at revision 0cc0fba562ef: 11 nodes; models ace_step_v1_3.5b.safetensors. Metadata-only index entry; not tested guidance.","snippet":"official ComfyUI workflow template audio_ace_step_1_t2a_<b>song</b> at revision 0cc0fba562ef: 11 nodes; models ace_step_v1_3.5b.safetensors. Metadata-only index entry; not tested guidance. frontend version: 1.51.9\nindex title","status":"active","categories":["T08"],"evidence":{"inspection":"metadata_checked","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M6G9XWPHC5VP3FED5YY4MR54","api_url":"https://veyrul.com/v1/records/src_01M6G9XWPHC5VP3FED5YY4MR54","updated_at":"2026-11-03T08:55:33.805Z","score":0.0193}},{"kind":"guide_preview","preview":{"schema":"guide-preview/v1","id":"art_01M6TW9HMPHNPA9HTRNGY94N05","type":"article","revision":1,"title":"Split a song into stems or vocals with Demucs","coverage":"For anyone who wants drums, bass, other and vocals, or vocals against the rest, from a finished song using the Demucs command line. You end up with separate audio files you can remix or study. Based on the project README; nothing was run, and the original repository is unmaintained.","example":"A not-run example separates vocals from one placeholder-named song with the two-stem option; the guide says the second output file name is not documented.","tips":["The README contradicts itself on the default model; name the model explicitly if the choice matters.","Two-stem mode still runs the full separation, so it saves neither time nor memory.","Listen to every stem against the original; automatic rescaling to avoid clipping can change the balance between stems."],"categories":["T14","T13"],"tags":["demucs","stem-separation","vocal-isolation","music-source-separation","htdemucs","karaoke","local-audio"],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed","review_revision":null},"media":[]}},{"kind":"guide_preview","preview":{"schema":"guide-preview/v1","id":"art_01M5NPD89RP6Y42DBTBB0S0QZD","type":"article","revision":1,"title":"Suno stem separation: the three split modes, stem counts and the help pages' warnings","coverage":"For musicians and editors who want to remix or mute parts of a Suno-generated song. It summarises the three split modes, the stem counts and the limits Suno's help pages state. Based on help text only; no separation was tested and no quality figures are given.","example":null,"tips":["Automatic labelling can be wrong, so listen to each stem instead of trusting its name.","Requesting an instrument or vocal that is not in the song may still use credits and give no usable stem.","Stems come from a stereo mix, so expect some bleed; that is an inference, not a Suno figure."],"categories":["T14"],"tags":[],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed","review_revision":null},"media":[]}},{"kind":"guide_preview","preview":{"schema":"guide-preview/v1","id":"art_01M5NGX6KWBHP9R9W1866ZGVB4","type":"article","revision":1,"title":"Which artefacts Demucs documents, and which common failure modes its README does not","coverage":"For audio editors and producers splitting stems from mixed tracks who want to know what the project itself documents. It lists the artefacts the README describes and treats phase smearing and vocal remnants as unmeasured expectations. Based on the README only; no model was run and the README is undated.","example":null,"tips":["Output stems are rescaled to avoid clipping, so do not assume stem levels still match the original mix.","The experimental six-source model is described as weak for piano; guitar is called acceptable on limited testing.","A problem missing from one README does not prove it never happens; label unmeasured failure modes as expectations."],"categories":["T13","T14"],"tags":[],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed","review_revision":null},"media":[]}},{"kind":"guide_preview","preview":{"schema":"guide-preview/v1","id":"art_01M5NP9WVHBCDFFFHPN1MEJNN1","type":"article","revision":1,"title":"What Demucs can and cannot separate when you need a dialogue or M&E stem","coverage":"For dub, remix and post-production people who only have a mixed track. It states which stems Demucs outputs, why film dialogue and effects do not map onto them, and the clipping and licence points from its README. Documentation-based; the model was not run and no dialogue-extraction quality was measured.","example":null,"tips":["Demucs is trained on music stems (drums, bass, other, vocals), so its vocals output targets singing, not speech dialogue.","Output stems are rescaled to avoid clipping, which can change relative levels between stems.","A separated sum of stems is a reference or rough-cut aid, not a delivery-grade music-and-effects mix."],"categories":["T09","T13","T14"],"tags":[],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed","review_revision":null},"media":[]}}],"truncated":false,"has_more":true,"next_cursor":"eyJyZWZlcmVuY2UiOiJNakEiLCJwcmV2aWV3IjpudWxsLCJxIjoiU3BsaXQgYSBzb25nIGludG8gc3RlbXMgb3Igdm9jYWxzIHdpdGggRGVtdWNzIiwidHlwZXMiOltdLCJjYXRlZ29yaWVzIjpbXX0","ordering":"bounded_reference_and_preview_lanes"}