Fix ACE-Step community UI parameter wiring
This commit is contained in:
Mikei386
@@ -31,10 +31,13 @@ Im Athena-Dashboard stehen **LLM-Betrieb**, **Musikstudio**, **Audio trennen**,
|
|||||||
verbindliche Produktionspfad.
|
verbindliche Produktionspfad.
|
||||||
- **Community UI · experimentell** öffnet `fspecii/ace-step-ui`. Die
|
- **Community UI · experimentell** öffnet `fspecii/ace-step-ui`. Die
|
||||||
CPU-leichte React/Express-Anwendung hält Bibliothek, Playlists und
|
CPU-leichte React/Express-Anwendung hält Bibliothek, Playlists und
|
||||||
Einstellungen in `/data/music/ace-step-ui`. Ein noch nicht übernommener
|
Einstellungen in `/data/music/ace-step-ui`. Sie verwendet die offizielle
|
||||||
Upstream-Kompatibilitätsfix für die aktuelle 72-Felder-Gradio-API ist lokal
|
`/release_task`-API mit benannten Parametern und ist damit unabhängig von der
|
||||||
zurückportiert; normale Generierung funktioniert, Cover und Remix gelten bis
|
Reihenfolge der Gradio-Felder. Normale Generierung funktioniert; Cover und
|
||||||
zu eigenen Ende-zu-Ende-Tests weiterhin als experimentell.
|
Remix gelten bis zu eigenen Ende-zu-Ende-Tests weiterhin als experimentell.
|
||||||
|
Vor dem Start zeigt sie die übertragenen Werte an. Referenzaudio beeinflusst
|
||||||
|
nur Klang und Produktion, während Quellaudio Melodie, Rhythmus und Akkorde
|
||||||
|
erhält.
|
||||||
|
|
||||||
Die Community-Oberfläche ist im WireGuard-Netz unter
|
Die Community-Oberfläche ist im WireGuard-Netz unter
|
||||||
`http://192.168.1.212:7861`, die originale Gradio-Oberfläche unter
|
`http://192.168.1.212:7861`, die originale Gradio-Oberfläche unter
|
||||||
|
|||||||
@@ -50,9 +50,11 @@ gemeldete maximale CUDA-Allokation lag bei 9,38 GiB. Es gab weder OOM noch
|
|||||||
CUDA-Fehler. Der technische Test ist damit bestanden; die subjektive
|
CUDA-Fehler. Der technische Test ist damit bestanden; die subjektive
|
||||||
Die originale ACE-Step-Gradio-Oberflaeche ist der stabile Produktionspfad. Die
|
Die originale ACE-Step-Gradio-Oberflaeche ist der stabile Produktionspfad. Die
|
||||||
persistente `fspecii/ace-step-ui`-Oberflaeche bleibt bis zur Abnahme aller
|
persistente `fspecii/ace-step-ui`-Oberflaeche bleibt bis zur Abnahme aller
|
||||||
Audio-zu-Audio-Modi experimentell. Der lokale Backport des offenen Upstream-PR
|
Audio-zu-Audio-Modi experimentell. Am 10. September 2026 wurde der zerbrechliche
|
||||||
[#109](https://github.com/fspecii/ace-step-ui/pull/109) wurde gegen die 72
|
Aufruf der positionsabhaengigen `/generation_wrapper`-Schnittstelle entfernt.
|
||||||
Parameter des live veroeffentlichten `/generation_wrapper`-Schemas getestet.
|
Die Community-UI nutzt nun `/release_task` mit benannten Parametern; das
|
||||||
|
versionierte Worker-Derivat reicht dabei auch Referenz-/Quellaudio, Cover-
|
||||||
|
Staerke, Thinking, AI Enhance und die XL-SFT-Werte weiter.
|
||||||
Ein zehnsekündiger FLAC-Textauftrag lief am 8. September 2026 erfolgreich durch
|
Ein zehnsekündiger FLAC-Textauftrag lief am 8. September 2026 erfolgreich durch
|
||||||
UI, Express-Backend und Gradio-API und wurde in der persistenten Bibliothek
|
UI, Express-Backend und Gradio-API und wurde in der persistenten Bibliothek
|
||||||
gespeichert; dieser Test belegt Cover und Remix ausdrücklich noch nicht.
|
gespeichert; dieser Test belegt Cover und Remix ausdrücklich noch nicht.
|
||||||
|
|||||||
@@ -38,15 +38,24 @@ WireGuard-Netz unter `http://192.168.1.212:7862` erreichbar.
|
|||||||
|
|
||||||
`music-ui` baut [fspecii/ace-step-ui](https://github.com/fspecii/ace-step-ui)
|
`music-ui` baut [fspecii/ace-step-ui](https://github.com/fspecii/ace-step-ui)
|
||||||
reproduzierbar von Commit `a1fdf91829ec6f7b98844f80e323529cd155dbf2`.
|
reproduzierbar von Commit `a1fdf91829ec6f7b98844f80e323529cd155dbf2`.
|
||||||
Ein lokaler Backport des noch offenen Upstream-PR
|
Die Community-Oberflaeche verwendet nicht mehr das positionsabhaengige
|
||||||
[#109](https://github.com/fspecii/ace-step-ui/pull/109) gleicht das Frontend an
|
Gradio-Schema. Ihr Express-Dienst ruft die offizielle `/release_task`-API mit
|
||||||
das aktuelle 72-Felder-Gradio-Schema an. Athena-spezifisch sind nur die
|
benannten Feldern auf; das kleine Worker-Derivat erweitert diese Route um die
|
||||||
persistente Ablage, die XL-SFT-Anzeige und die gemeldeten Laufzeitlimits.
|
im installierten `GenerationParams` bereits vorhandenen Felder fuer Referenz-,
|
||||||
|
Quell- und Coveraudio sowie XL-SFT-Parameter. Athena-spezifisch sind ausserdem
|
||||||
|
die persistente Ablage, die XL-SFT-Anzeige und die gemeldeten Laufzeitlimits.
|
||||||
Schlaegt bei einer spaeteren Upstream-Fassung ein Patch-Anker fehl, bricht der
|
Schlaegt bei einer spaeteren Upstream-Fassung ein Patch-Anker fehl, bricht der
|
||||||
Image-Build ab. Die Community-UI unter `http://192.168.1.212:7861` ist bis zu
|
Image-Build ab. Die Community-UI unter `http://192.168.1.212:7861` ist bis zu
|
||||||
vollstaendigen Ende-zu-Ende-Tests von Cover und Remix als experimentell
|
vollstaendigen Ende-zu-Ende-Tests von Cover und Remix als experimentell
|
||||||
gekennzeichnet.
|
gekennzeichnet.
|
||||||
|
|
||||||
|
Die Community-UI startet XL-SFT mit 50 Schritten, Guidance 7, Shift 1, FLAC
|
||||||
|
und aktivem Thinking ueber den 1,7B-Planer. `AI Enhance` wird als `use_format`
|
||||||
|
uebertragen. Vor jedem Auftrag zeigt sie die tatsaechlich gesendeten Parameter
|
||||||
|
und faengt offensichtliche Widersprueche ab. Referenzaudio steuert nur Klang
|
||||||
|
und Produktion; nur **Quellaudio / Cover** erhaelt Melodie, Rhythmus und
|
||||||
|
Akkorde.
|
||||||
|
|
||||||
## Start
|
## Start
|
||||||
|
|
||||||
Vor dem Start muessen das aktive llama.cpp-Profil und Qwen3-TTS beendet sein.
|
Vor dem Start muessen das aktive llama.cpp-Profil und Qwen3-TTS beendet sein.
|
||||||
|
|||||||
@@ -27,98 +27,202 @@ patch('server/src/services/acestep.ts', (text) => {
|
|||||||
'persistent generated audio',
|
'persistent generated audio',
|
||||||
);
|
);
|
||||||
|
|
||||||
const oldArgs = ` params.instruction || 'Fill the audio semantic mask with the style described in the text prompt.', // 17: Instruction
|
text = text.replace("import { handle_file } from '@gradio/client';\n", '');
|
||||||
params.audioCoverStrength ?? 1.0, // 18: Audio Cover Strength
|
text = text.replace('getGradioClient, ', '');
|
||||||
0.0, // 19: Cover Noise Strength (ACE-Step v1.5 new param, default 0.0)
|
|
||||||
(params.taskType === 'audio2audio' ? 'cover' : params.taskType) || 'text2music', // 20: Task Type
|
|
||||||
params.useAdg ?? false, // 21: Use ADG
|
|
||||||
params.cfgIntervalStart ?? 0.0, // 22: CFG Interval Start
|
|
||||||
params.cfgIntervalEnd ?? 1.0, // 23: CFG Interval End
|
|
||||||
params.shift ?? 3.0, // 24: Shift
|
|
||||||
params.inferMethod || 'ode', // 25: Inference Method
|
|
||||||
params.customTimesteps || '', // 26: Custom Timesteps
|
|
||||||
params.audioFormat || 'mp3', // 27: Audio Format
|
|
||||||
params.lmTemperature ?? 0.85, // 28: LM Temperature
|
|
||||||
isThinking, // 29: Think
|
|
||||||
params.lmCfgScale ?? 2.0, // 30: LM CFG Scale
|
|
||||||
params.lmTopK ?? 0, // 31: LM Top-K
|
|
||||||
params.lmTopP ?? 0.9, // 32: LM Top-P
|
|
||||||
params.lmNegativePrompt || 'NO USER INPUT', // 33: LM Negative Prompt
|
|
||||||
useCot ? (params.useCotMetas ?? true) : false, // 34: CoT Metas
|
|
||||||
useCot ? (params.useCotCaption ?? true) : false, // 35: CaptionRewrite
|
|
||||||
useCot ? (params.useCotLanguage ?? true) : false, // 36: CoT Language
|
|
||||||
params.isFormatCaption ?? false, // 37: Is Format Caption State
|
|
||||||
params.constrainedDecodingDebug ?? false, // 38: Constrained Decoding Debug
|
|
||||||
params.allowLmBatch ?? true, // 39: ParallelThinking
|
|
||||||
params.getScores ?? false, // 40: Auto Score
|
|
||||||
params.getLrc ?? false, // 41: Auto LRC (timestamped lyrics)
|
|
||||||
params.scoreScale ?? 0.5, // 42: Quality Score Sensitivity (0.01-1.0)
|
|
||||||
params.lmBatchChunkSize ?? 8, // 43: LM Batch Chunk Size
|
|
||||||
params.trackName || null, // 44: Track Name
|
|
||||||
params.completeTrackClasses || [], // 45: Track Names
|
|
||||||
true, // 46: Enable Normalization (ACE-Step v1.5, default true)
|
|
||||||
-1.0, // 47: Normalization DB (ACE-Step v1.5, default -1.0)
|
|
||||||
0.0, // 48: Latent Shift (ACE-Step v1.5, default 0.0)
|
|
||||||
1.0, // 49: Latent Rescale (ACE-Step v1.5, default 1.0)
|
|
||||||
params.autogen ?? false, // 50: AutoGen`;
|
|
||||||
|
|
||||||
const newArgs = ` params.instruction || 'Fill the audio semantic mask based on the given conditions:', // 17: Instruction
|
const helperStart = text.indexOf('// Gradio generation: map params');
|
||||||
params.audioCoverStrength ?? 1.0, // 18: LM Codes Strength
|
const helperEnd = text.indexOf('/**\n * Download a Gradio audio result file', helperStart);
|
||||||
0.0, // 19: Cover Strength
|
if (helperStart < 0 || helperEnd < 0) throw new Error('legacy Gradio helper anchors missing');
|
||||||
false, // 20: no_fsq
|
const helperCommentStart = text.lastIndexOf('// ---------------------------------------------------------------------------', helperStart);
|
||||||
params.useAdg ?? false, // 21: Use ADG
|
const namedHelpers = `// ---------------------------------------------------------------------------
|
||||||
params.cfgIntervalStart ?? 0.0, // 22: CFG Interval Start
|
// Named REST generation through ACE-Step's official /release_task endpoint
|
||||||
params.cfgIntervalEnd ?? 1.0, // 23: CFG Interval End
|
// ---------------------------------------------------------------------------
|
||||||
params.shift ?? 1.0, // 24: Shift
|
|
||||||
params.inferMethod || 'ode', // 25: Inference Method
|
|
||||||
'euler', // 26: Sampler Mode
|
|
||||||
0.0, // 27: Velocity Norm Threshold
|
|
||||||
0.0, // 28: Velocity EMA Factor
|
|
||||||
false, // 29: Enable DCW
|
|
||||||
'double', // 30: DCW Mode
|
|
||||||
0.02, // 31: DCW Scaler
|
|
||||||
0.06, // 32: DCW High Scaler
|
|
||||||
'haar', // 33: DCW Wavelet
|
|
||||||
params.customTimesteps || '', // 34: Custom Timesteps
|
|
||||||
params.audioFormat || 'flac', // 35: Audio Format
|
|
||||||
'320k', // 36: MP3 Bitrate
|
|
||||||
48000, // 37: MP3 Sample Rate
|
|
||||||
params.lmTemperature ?? 0.85, // 38: LM Temperature
|
|
||||||
isThinking, // 39: Think
|
|
||||||
params.lmCfgScale ?? 2.0, // 40: LM CFG Scale
|
|
||||||
params.lmTopK ?? 0, // 41: LM Top-K
|
|
||||||
params.lmTopP ?? 0.9, // 42: LM Top-P
|
|
||||||
params.lmNegativePrompt || 'NO USER INPUT', // 43: LM Negative Prompt
|
|
||||||
useCot ? (params.useCotMetas ?? true) : false, // 44: CoT Metas
|
|
||||||
useCot ? (params.useCotCaption ?? true) : false, // 45: CaptionRewrite
|
|
||||||
useCot ? (params.useCotLanguage ?? true) : false, // 46: CoT Language
|
|
||||||
params.constrainedDecodingDebug ?? false, // 47: Constrained Decoding Debug
|
|
||||||
params.allowLmBatch ?? true, // 48: ParallelThinking
|
|
||||||
params.getScores ?? false, // 49: Auto Score
|
|
||||||
params.getLrc ?? false, // 50: Auto LRC
|
|
||||||
params.scoreScale ?? 0.5, // 51: Quality Score Sensitivity
|
|
||||||
params.lmBatchChunkSize ?? 8, // 52: LM Batch Chunk Size
|
|
||||||
params.trackName || null, // 53: Track Name
|
|
||||||
params.completeTrackClasses || [], // 54: Track Names
|
|
||||||
true, // 55: Enable Normalization
|
|
||||||
-1.0, // 56: Target Peak (dB)
|
|
||||||
0.0, // 57: Fade In
|
|
||||||
0.0, // 58: Fade Out
|
|
||||||
0.0, // 59: Latent Shift
|
|
||||||
1.0, // 60: Latent Rescale
|
|
||||||
'balanced', // 61: Repaint Mode
|
|
||||||
0.5, // 62: Repaint Strength
|
|
||||||
0.0, // 63: variance
|
|
||||||
'', // 64: repaint seed
|
|
||||||
false, // 65: Edit
|
|
||||||
null, // 66: source caption
|
|
||||||
null, // 67: source lyrics
|
|
||||||
0.0, // 68: n_min
|
|
||||||
1.0, // 69: n_max
|
|
||||||
1, // 70: n_avg
|
|
||||||
params.autogen ?? false, // 71: AutoGen`;
|
|
||||||
|
|
||||||
return replaceOnce(text, oldArgs, newArgs, 'Athena Gradio argument schema');
|
function resolveAudioPath(audioUrl: string): string {
|
||||||
|
if (audioUrl.startsWith('/audio/')) {
|
||||||
|
return path.join(AUDIO_DIR, audioUrl.replace('/audio/', ''));
|
||||||
|
}
|
||||||
|
if (audioUrl.startsWith('http')) {
|
||||||
|
try {
|
||||||
|
const parsed = new URL(audioUrl);
|
||||||
|
if (parsed.pathname.startsWith('/audio/')) {
|
||||||
|
return path.join(AUDIO_DIR, parsed.pathname.replace('/audio/', ''));
|
||||||
|
}
|
||||||
|
} catch { /* fall through */ }
|
||||||
|
}
|
||||||
|
return audioUrl;
|
||||||
|
}
|
||||||
|
|
||||||
|
function resolveWorkerAudioPath(audioUrl: string | undefined): string | undefined {
|
||||||
|
if (!audioUrl) return undefined;
|
||||||
|
const localPath = resolveAudioPath(audioUrl);
|
||||||
|
if (!existsSync(localPath)) {
|
||||||
|
throw new Error(\`Uploaded audio is missing: \${localPath}\`);
|
||||||
|
}
|
||||||
|
const relativePath = path.relative(AUDIO_DIR, localPath);
|
||||||
|
if (relativePath.startsWith('..') || path.isAbsolute(relativePath)) {
|
||||||
|
throw new Error('Audio path is outside the shared Community UI storage');
|
||||||
|
}
|
||||||
|
return path.posix.join('/data/community-audio', relativePath.split(path.sep).join('/'));
|
||||||
|
}
|
||||||
|
|
||||||
|
function buildReleaseTaskPayload(params: GenerationParams): Record<string, unknown> {
|
||||||
|
const caption = params.style || 'pop music';
|
||||||
|
const prompt = params.customMode ? caption : (params.songDescription || caption);
|
||||||
|
const thinking = params.thinking ?? true;
|
||||||
|
const enhance = params.enhance ?? false;
|
||||||
|
const taskType = params.taskType === 'audio2audio' ? 'cover' : (params.taskType || 'text2music');
|
||||||
|
|
||||||
|
return {
|
||||||
|
prompt,
|
||||||
|
lyrics: params.instrumental ? '[Instrumental]' : (params.lyrics || ''),
|
||||||
|
instrumental: params.instrumental,
|
||||||
|
vocal_language: params.vocalLanguage || 'en',
|
||||||
|
bpm: params.bpm && params.bpm > 0 ? params.bpm : 0,
|
||||||
|
key_scale: params.keyScale || '',
|
||||||
|
time_signature: params.timeSignature || '',
|
||||||
|
audio_duration: params.duration && params.duration > 0 ? params.duration : -1,
|
||||||
|
inference_steps: params.inferenceSteps ?? 50,
|
||||||
|
guidance_scale: params.guidanceScale ?? 7.0,
|
||||||
|
shift: params.shift ?? 1.0,
|
||||||
|
infer_method: params.inferMethod || 'ode',
|
||||||
|
batch_size: Math.min(Math.max(params.batchSize ?? 1, 1), 16),
|
||||||
|
use_random_seed: params.randomSeed !== false,
|
||||||
|
seed: params.seed ?? -1,
|
||||||
|
thinking,
|
||||||
|
use_format: enhance,
|
||||||
|
lm_temperature: params.lmTemperature ?? 0.85,
|
||||||
|
lm_cfg_scale: params.lmCfgScale ?? 2.0,
|
||||||
|
lm_top_k: params.lmTopK ?? 0,
|
||||||
|
lm_top_p: params.lmTopP ?? 0.9,
|
||||||
|
lm_negative_prompt: params.lmNegativePrompt || 'NO USER INPUT',
|
||||||
|
use_cot_metas: thinking ? (params.useCotMetas ?? true) : false,
|
||||||
|
use_cot_caption: thinking ? (params.useCotCaption ?? true) : false,
|
||||||
|
use_cot_language: thinking ? (params.useCotLanguage ?? true) : false,
|
||||||
|
allow_lm_batch: params.allowLmBatch ?? true,
|
||||||
|
constrained_decoding_debug: params.constrainedDecodingDebug ?? false,
|
||||||
|
lm_batch_chunk_size: params.lmBatchChunkSize ?? 8,
|
||||||
|
task_type: taskType,
|
||||||
|
instruction: params.instruction || 'Fill the audio semantic mask based on the given conditions:',
|
||||||
|
reference_audio_path: resolveWorkerAudioPath(params.referenceAudioUrl),
|
||||||
|
src_audio_path: resolveWorkerAudioPath(params.sourceAudioUrl),
|
||||||
|
audio_codes: params.audioCodes || '',
|
||||||
|
repainting_start: params.repaintingStart ?? 0.0,
|
||||||
|
repainting_end: params.repaintingEnd ?? -1,
|
||||||
|
audio_cover_strength: params.audioCoverStrength ?? 1.0,
|
||||||
|
use_adg: params.useAdg ?? false,
|
||||||
|
cfg_interval_start: params.cfgIntervalStart ?? 0.0,
|
||||||
|
cfg_interval_end: params.cfgIntervalEnd ?? 1.0,
|
||||||
|
audio_format: params.audioFormat || 'flac',
|
||||||
|
mp3_bitrate: '320k',
|
||||||
|
mp3_sample_rate: 48000,
|
||||||
|
};
|
||||||
|
}
|
||||||
|
|
||||||
|
`;
|
||||||
|
text = text.slice(0, helperCommentStart) + namedHelpers + text.slice(helperEnd);
|
||||||
|
|
||||||
|
const processStart = text.indexOf('// processGeneration — Gradio primary');
|
||||||
|
const processEnd = text.indexOf('function isAudioFile', processStart);
|
||||||
|
if (processStart < 0 || processEnd < 0) throw new Error('legacy generation anchors missing');
|
||||||
|
const processCommentStart = text.lastIndexOf('// ---------------------------------------------------------------------------', processStart);
|
||||||
|
const namedProcess = `// ---------------------------------------------------------------------------
|
||||||
|
// processGeneration — official named REST API only
|
||||||
|
// ---------------------------------------------------------------------------
|
||||||
|
|
||||||
|
async function processGeneration(
|
||||||
|
jobId: string,
|
||||||
|
params: GenerationParams,
|
||||||
|
job: ActiveJob,
|
||||||
|
): Promise<void> {
|
||||||
|
job.status = 'running';
|
||||||
|
job.stage = 'Preparing named ACE-Step request...';
|
||||||
|
|
||||||
|
if ((params.taskType === 'cover' || params.taskType === 'audio2audio') && !params.sourceAudioUrl && !params.audioCodes) {
|
||||||
|
job.status = 'failed';
|
||||||
|
job.error = \`task_type='\${params.taskType}' requires source audio or audio codes\`;
|
||||||
|
return;
|
||||||
|
}
|
||||||
|
|
||||||
|
try {
|
||||||
|
if (params.ditModel) {
|
||||||
|
job.stage = \`Loading model \${params.ditModel}...\`;
|
||||||
|
await switchModelIfNeeded(params.ditModel);
|
||||||
|
}
|
||||||
|
|
||||||
|
const payload = buildReleaseTaskPayload(params);
|
||||||
|
console.log(\`Job \${jobId}: POST /release_task with named parameters\`, payload);
|
||||||
|
job.stage = 'Generating music via named ACE-Step API...';
|
||||||
|
|
||||||
|
const releaseResponse = await fetch(\`\${ACESTEP_API}/release_task\`, {
|
||||||
|
method: 'POST',
|
||||||
|
headers: { 'Content-Type': 'application/json' },
|
||||||
|
body: JSON.stringify(payload),
|
||||||
|
});
|
||||||
|
const releaseText = await releaseResponse.text();
|
||||||
|
if (!releaseResponse.ok) {
|
||||||
|
throw new Error(\`ACE-Step /release_task failed (\${releaseResponse.status}): \${releaseText}\`);
|
||||||
|
}
|
||||||
|
const release = JSON.parse(releaseText) as any;
|
||||||
|
if (release.code !== 200 || !release.data?.task_id) {
|
||||||
|
throw new Error(release.error || 'ACE-Step returned no task_id');
|
||||||
|
}
|
||||||
|
|
||||||
|
const taskId = String(release.data.task_id);
|
||||||
|
job.taskId = taskId;
|
||||||
|
const queryResponse = await fetch(\`\${ACESTEP_API}/query_result\`, {
|
||||||
|
method: 'POST',
|
||||||
|
headers: { 'Content-Type': 'application/json' },
|
||||||
|
body: JSON.stringify({ task_id_list: [taskId] }),
|
||||||
|
});
|
||||||
|
const queryText = await queryResponse.text();
|
||||||
|
if (!queryResponse.ok) {
|
||||||
|
throw new Error(\`ACE-Step /query_result failed (\${queryResponse.status}): \${queryText}\`);
|
||||||
|
}
|
||||||
|
const query = JSON.parse(queryText) as any;
|
||||||
|
const taskResult = query.data?.[0];
|
||||||
|
const audioItems = taskResult?.result ? JSON.parse(taskResult.result) : [];
|
||||||
|
if (!Array.isArray(audioItems) || audioItems.length === 0) {
|
||||||
|
throw new Error('ACE-Step completed without downloadable audio results');
|
||||||
|
}
|
||||||
|
|
||||||
|
const audioUrls: string[] = [];
|
||||||
|
let actualDuration = 0;
|
||||||
|
for (const item of audioItems) {
|
||||||
|
if (!item?.url) continue;
|
||||||
|
const remoteUrl = new URL(item.url, ACESTEP_API).toString();
|
||||||
|
const remoteName = String(item.file || item.url);
|
||||||
|
const ext = path.extname(remoteName) || \`.\${params.audioFormat || 'flac'}\`;
|
||||||
|
const filename = \`\${jobId}_\${audioUrls.length}\${ext}\`;
|
||||||
|
const destPath = path.join(AUDIO_DIR, filename);
|
||||||
|
await downloadGradioAudioFile({ url: remoteUrl, orig_name: remoteName }, destPath);
|
||||||
|
if (audioUrls.length === 0) actualDuration = getAudioDuration(destPath);
|
||||||
|
audioUrls.push(\`/audio/\${filename}\`);
|
||||||
|
}
|
||||||
|
if (audioUrls.length === 0) throw new Error('ACE-Step returned no supported audio files');
|
||||||
|
|
||||||
|
const first = audioItems[0] || {};
|
||||||
|
job.status = 'succeeded';
|
||||||
|
job.result = {
|
||||||
|
audioUrls,
|
||||||
|
duration: actualDuration || Number(first.duration) || params.duration || 0,
|
||||||
|
bpm: Number(first.bpm) || params.bpm,
|
||||||
|
keyScale: first.keyscale || params.keyScale,
|
||||||
|
timeSignature: first.timesignature || params.timeSignature,
|
||||||
|
status: 'succeeded',
|
||||||
|
};
|
||||||
|
job.rawResponse = { release, query, transmittedParameters: payload };
|
||||||
|
console.log(\`Job \${jobId}: Completed via named REST API with \${audioUrls.length} audio files\`);
|
||||||
|
} catch (error) {
|
||||||
|
job.status = 'failed';
|
||||||
|
job.error = error instanceof Error ? error.message : String(error);
|
||||||
|
console.error(\`Job \${jobId}: Named REST generation failed\`, error);
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
`;
|
||||||
|
text = text.slice(0, processCommentStart) + namedProcess + text.slice(processEnd);
|
||||||
|
return text;
|
||||||
});
|
});
|
||||||
|
|
||||||
patch('server/src/services/storage/local.ts', (text) => {
|
patch('server/src/services/storage/local.ts', (text) => {
|
||||||
@@ -159,6 +263,17 @@ patch('server/src/routes/referenceTrack.ts', (text) => {
|
|||||||
});
|
});
|
||||||
|
|
||||||
patch('server/src/routes/generate.ts', (text) => {
|
patch('server/src/routes/generate.ts', (text) => {
|
||||||
|
text = replaceOnce(
|
||||||
|
text,
|
||||||
|
" thinking?: boolean;\n audioFormat?: 'mp3' | 'flac';",
|
||||||
|
" thinking?: boolean;\n enhance?: boolean;\n audioFormat?: 'mp3' | 'flac';",
|
||||||
|
'server enhance type',
|
||||||
|
);
|
||||||
|
const enhanceAnchor = ' thinking,\n audioFormat,';
|
||||||
|
if (text.split(enhanceAnchor).length - 1 !== 2) {
|
||||||
|
throw new Error('expected enhance anchor in destructuring and forwarding');
|
||||||
|
}
|
||||||
|
text = text.replaceAll(enhanceAnchor, ' thinking,\n enhance,\n audioFormat,');
|
||||||
text = replaceOnce(
|
text = replaceOnce(
|
||||||
text,
|
text,
|
||||||
" const ALL_DIT_MODELS = [\n 'acestep-v15-turbo',",
|
" const ALL_DIT_MODELS = [\n 'acestep-v15-turbo',",
|
||||||
@@ -185,3 +300,108 @@ patch('server/src/routes/generate.ts', (text) => {
|
|||||||
`;
|
`;
|
||||||
return text.slice(0, start) + limits + text.slice(end);
|
return text.slice(0, start) + limits + text.slice(end);
|
||||||
});
|
});
|
||||||
|
|
||||||
|
patch('services/api.ts', (text) => replaceOnce(
|
||||||
|
text,
|
||||||
|
" thinking?: boolean;\n audioFormat?: 'mp3' | 'flac';",
|
||||||
|
" thinking?: boolean;\n enhance?: boolean;\n audioFormat?: 'mp3' | 'flac';",
|
||||||
|
'client enhance type',
|
||||||
|
));
|
||||||
|
|
||||||
|
patch('App.tsx', (text) => {
|
||||||
|
text = replaceOnce(
|
||||||
|
text,
|
||||||
|
' thinking: params.thinking,\n audioFormat: params.audioFormat,',
|
||||||
|
' thinking: params.thinking,\n enhance: params.enhance,\n audioFormat: params.audioFormat,',
|
||||||
|
'client enhance forwarding',
|
||||||
|
);
|
||||||
|
return replaceOnce(
|
||||||
|
text,
|
||||||
|
' title: params.title,\n instrumental: params.instrumental,',
|
||||||
|
' title: params.title,\n ditModel: params.ditModel,\n instrumental: params.instrumental,',
|
||||||
|
'client model forwarding',
|
||||||
|
);
|
||||||
|
});
|
||||||
|
|
||||||
|
patch('components/CreatePanel.tsx', (text) => {
|
||||||
|
text = replaceOnce(text, 'useState(9.0);', 'useState(7.0);', 'guidance default');
|
||||||
|
text = replaceOnce(
|
||||||
|
text,
|
||||||
|
'useState(false); // Default false for GPU compatibility',
|
||||||
|
'useState(true); // Athena default: use the 1.7B planner for coherent structure',
|
||||||
|
'thinking default',
|
||||||
|
);
|
||||||
|
text = replaceOnce(text, "useState<'mp3' | 'flac'>('mp3');", "useState<'mp3' | 'flac'>('flac');", 'lossless default');
|
||||||
|
text = replaceOnce(text, 'useState(12);', 'useState(50);', 'XL-SFT steps default');
|
||||||
|
text = replaceOnce(text, "localStorage.getItem('ace-lmModel') || 'acestep-5Hz-lm-0.6B'", "localStorage.getItem('ace-lmModel') || 'acestep-5Hz-lm-1.7B'", 'planner model default');
|
||||||
|
text = replaceOnce(text, 'useState(3.0);', 'useState(1.0);', 'shift default');
|
||||||
|
|
||||||
|
text = replaceOnce(
|
||||||
|
text,
|
||||||
|
' // Bulk generation: loop bulkCount times\n for (let i = 0; i < bulkCount; i++) {',
|
||||||
|
` const requestedText = customMode ? styleWithGender : songDescription;
|
||||||
|
const asksForVocals = /\\b(vocals?|singer|singing|male voice|female voice|gesang|stimme|sänger(?:in)?|singt)\\b/i.test(
|
||||||
|
\`\${requestedText || ''}\\n\${lyrics}\`,
|
||||||
|
);
|
||||||
|
if (!instrumental && asksForVocals && !lyrics.trim()) {
|
||||||
|
window.alert('Widerspruch: Der Auftrag verlangt Gesang, aber das Liedtextfeld ist leer. Bitte Text eintragen oder „Instrumental“ wählen.');
|
||||||
|
return;
|
||||||
|
}
|
||||||
|
if ((taskType === 'cover' || taskType === 'audio2audio') && !sourceAudioUrl.trim() && !audioCodes.trim()) {
|
||||||
|
window.alert('Für einen Cover-Auftrag fehlt das Quellaudio. Bitte unter „Quellaudio / Cover“ eine Datei auswählen.');
|
||||||
|
return;
|
||||||
|
}
|
||||||
|
|
||||||
|
const taskLabel = taskType === 'cover' || taskType === 'audio2audio' ? 'Cover / Audio-zu-Audio' : taskType;
|
||||||
|
const summary = [
|
||||||
|
'Folgende Parameter werden tatsächlich an Athena übertragen:',
|
||||||
|
'',
|
||||||
|
\`Aufgabe: \${taskLabel}\`,
|
||||||
|
\`Modell: \${selectedModel}\`,
|
||||||
|
\`Dauer: \${duration > 0 ? \`\${duration} Sekunden\` : 'automatisch'}\`,
|
||||||
|
\`Tempo: \${bpm > 0 ? \`\${bpm} BPM\` : 'automatisch'}\`,
|
||||||
|
\`Tonart: \${keyScale || 'automatisch'}\`,
|
||||||
|
\`Taktart: \${timeSignature || 'automatisch'}\`,
|
||||||
|
\`Thinking/Planung: \${thinking ? 'AN' : 'AUS'}\`,
|
||||||
|
\`AI Enhance: \${enhance ? 'AN' : 'AUS'}\`,
|
||||||
|
\`XL-SFT: \${inferenceSteps} Schritte, Guidance \${guidanceScale}, Shift \${shift}\`,
|
||||||
|
\`Gesang: \${instrumental ? 'nein (Instrumental)' : 'ja'}\`,
|
||||||
|
\`Referenzaudio (nur Klang/Produktion): \${referenceAudioUrl ? 'vorhanden' : 'keines'}\`,
|
||||||
|
\`Quellaudio (Melodie/Rhythmus/Akkorde): \${sourceAudioUrl ? 'vorhanden' : 'keines'}\`,
|
||||||
|
\`Ausgabe: \${audioFormat.toUpperCase()}, \${batchSize} Variation(en), \${bulkCount} Auftrag/Aufträge\`,
|
||||||
|
'',
|
||||||
|
'Auftrag jetzt starten?',
|
||||||
|
].join('\\n');
|
||||||
|
if (!window.confirm(summary)) return;
|
||||||
|
|
||||||
|
// Bulk generation: loop bulkCount times
|
||||||
|
for (let i = 0; i < bulkCount; i++) {`,
|
||||||
|
'validation and transmitted parameter summary',
|
||||||
|
);
|
||||||
|
|
||||||
|
text = replaceOnce(
|
||||||
|
text,
|
||||||
|
" {t('reference')}\n </button>",
|
||||||
|
" Referenzaudio\n </button>",
|
||||||
|
'reference tab label',
|
||||||
|
);
|
||||||
|
text = replaceOnce(
|
||||||
|
text,
|
||||||
|
" {t('cover')}\n </button>",
|
||||||
|
" Quellaudio / Cover\n </button>",
|
||||||
|
'source tab label',
|
||||||
|
);
|
||||||
|
text = replaceOnce(
|
||||||
|
text,
|
||||||
|
' {/* Audio Content */}\n <div className="p-3 space-y-2">',
|
||||||
|
` {/* Audio Content */}
|
||||||
|
<div className="p-3 space-y-2">
|
||||||
|
<p className="text-[11px] leading-relaxed text-zinc-500 dark:text-zinc-400">
|
||||||
|
{audioTab === 'reference'
|
||||||
|
? 'Referenzaudio beeinflusst nur Klang, Instrumentierung und Produktion – nicht die Melodie.'
|
||||||
|
: 'Quellaudio / Cover erhält Melodie, Rhythmus und Akkorde des hochgeladenen Titels.'}
|
||||||
|
</p>`,
|
||||||
|
'audio semantics explanation',
|
||||||
|
);
|
||||||
|
return text;
|
||||||
|
});
|
||||||
|
|||||||
@@ -1,6 +1,8 @@
|
|||||||
services:
|
services:
|
||||||
music-worker:
|
music-worker:
|
||||||
image: ghcr.io/ace-step/ace-step-1.5:latest@sha256:95652cd780c78a1b1a7f6f0335530430f0ae53d96c7c12d59f9f39fa23d38567
|
build:
|
||||||
|
context: ./worker
|
||||||
|
image: mike-ai/ace-step-1.5:named-api-v1
|
||||||
container_name: mike-ai-music-acestep-test
|
container_name: mike-ai-music-acestep-test
|
||||||
labels:
|
labels:
|
||||||
com.mike-ai.music-worker: "acestep"
|
com.mike-ai.music-worker: "acestep"
|
||||||
@@ -34,6 +36,8 @@ services:
|
|||||||
- ${ACESTEP_HF_CACHE_DIR:-/data/models/acestep/hf-cache}:/root/.cache/huggingface
|
- ${ACESTEP_HF_CACHE_DIR:-/data/models/acestep/hf-cache}:/root/.cache/huggingface
|
||||||
- ${ACESTEP_OUTPUT_DIR:-/data/music/acestep}:/app/gradio_outputs
|
- ${ACESTEP_OUTPUT_DIR:-/data/music/acestep}:/app/gradio_outputs
|
||||||
- ${ACESTEP_OUTPUT_DIR:-/data/music/acestep}:/app/output
|
- ${ACESTEP_OUTPUT_DIR:-/data/music/acestep}:/app/output
|
||||||
|
# Uploaded Community-UI audio is shared read-only with the named REST API.
|
||||||
|
- ${ACESTEP_UI_DATA_DIR:-/data/music/ace-step-ui}/audio:/data/community-audio:ro
|
||||||
# Version-pinned UI defaults for XL-SFT quality and lossless output.
|
# Version-pinned UI defaults for XL-SFT quality and lossless output.
|
||||||
- ./overrides/model_config.py:/app/acestep/ui/gradio/events/generation/model_config.py:ro
|
- ./overrides/model_config.py:/app/acestep/ui/gradio/events/generation/model_config.py:ro
|
||||||
- ./overrides/generation_advanced_output_controls.py:/app/acestep/ui/gradio/interfaces/generation_advanced_output_controls.py:ro
|
- ./overrides/generation_advanced_output_controls.py:/app/acestep/ui/gradio/interfaces/generation_advanced_output_controls.py:ro
|
||||||
|
|||||||
@@ -0,0 +1,6 @@
|
|||||||
|
FROM ghcr.io/ace-step/ace-step-1.5:latest@sha256:95652cd780c78a1b1a7f6f0335530430f0ae53d96c7c12d59f9f39fa23d38567
|
||||||
|
|
||||||
|
COPY patch-api-routes.py /tmp/patch-api-routes.py
|
||||||
|
RUN /usr/bin/python3 /tmp/patch-api-routes.py \
|
||||||
|
/app/acestep/ui/gradio/api/api_routes.py \
|
||||||
|
&& rm /tmp/patch-api-routes.py
|
||||||
@@ -0,0 +1,137 @@
|
|||||||
|
"""Extend ACE-Step's official /release_task route with named generation inputs.
|
||||||
|
|
||||||
|
The base image already provides the route. This build-time patch only exposes
|
||||||
|
the parameters supported by its installed GenerationParams/GenerationConfig
|
||||||
|
dataclasses, so the separate Community UI never has to depend on Gradio's
|
||||||
|
positional component order.
|
||||||
|
"""
|
||||||
|
|
||||||
|
from pathlib import Path
|
||||||
|
import sys
|
||||||
|
|
||||||
|
|
||||||
|
target = Path(sys.argv[1])
|
||||||
|
source = target.read_text(encoding="utf-8")
|
||||||
|
|
||||||
|
|
||||||
|
def replace_once(old: str, new: str, label: str) -> None:
|
||||||
|
global source
|
||||||
|
count = source.count(old)
|
||||||
|
if count != 1:
|
||||||
|
raise RuntimeError(f"{label}: expected one anchor, found {count}")
|
||||||
|
source = source.replace(old, new, 1)
|
||||||
|
|
||||||
|
|
||||||
|
old_params = ''' # Build generation params with alias support
|
||||||
|
params = GenerationParams(
|
||||||
|
task_type=get_param("task_type", default="text2music"),
|
||||||
|
caption=caption,
|
||||||
|
lyrics=lyrics,
|
||||||
|
bpm=sample_bpm or get_param("bpm"),
|
||||||
|
keyscale=sample_keyscale or get_param("key_scale", "keyscale", "key", default=""),
|
||||||
|
timesignature=sample_timesignature or get_param("time_signature", "timesignature", default=""),
|
||||||
|
duration=sample_duration or get_param("audio_duration", "duration", default=-1),
|
||||||
|
vocal_language=sample_language,
|
||||||
|
inference_steps=get_param("inference_steps", default=8),
|
||||||
|
guidance_scale=float(get_param("guidance_scale", default=7.0) or 7.0),
|
||||||
|
seed=int(get_param("seed", default=-1) or -1),
|
||||||
|
thinking=to_bool(get_param("thinking"), False),
|
||||||
|
lm_temperature=lm_temperature,
|
||||||
|
lm_cfg_scale=float(get_param("lm_cfg_scale", default=2.0) or 2.0),
|
||||||
|
lm_negative_prompt=get_param("lm_negative_prompt", default="NO USER INPUT") or "NO USER INPUT",
|
||||||
|
repaint_latent_crossfade_frames=int(
|
||||||
|
get_param("repaint_latent_crossfade_frames", default=10) or 10,
|
||||||
|
),
|
||||||
|
repaint_wav_crossfade_sec=float(
|
||||||
|
get_param("repaint_wav_crossfade_sec", default=0.0) or 0.0,
|
||||||
|
),
|
||||||
|
repaint_mode=get_param("repaint_mode", default="balanced") or "balanced",
|
||||||
|
repaint_strength=float(
|
||||||
|
get_param("repaint_strength", default=0.5) or 0.5,
|
||||||
|
),
|
||||||
|
)
|
||||||
|
'''
|
||||||
|
|
||||||
|
new_params = ''' # Build generation params with alias support. Keep this
|
||||||
|
# mapping explicit: every public API field below is named and independent
|
||||||
|
# from the order of components in the Gradio interface.
|
||||||
|
raw_bpm = sample_bpm or get_param("bpm")
|
||||||
|
params = GenerationParams(
|
||||||
|
task_type=get_param("task_type", default="text2music") or "text2music",
|
||||||
|
instruction=get_param("instruction", default="Fill the audio semantic mask based on the given conditions:") or "Fill the audio semantic mask based on the given conditions:",
|
||||||
|
reference_audio=get_param("reference_audio_path", "reference_audio"),
|
||||||
|
src_audio=get_param("src_audio_path", "src_audio", "source_audio"),
|
||||||
|
audio_codes=get_param("audio_codes", default="") or "",
|
||||||
|
caption=caption,
|
||||||
|
lyrics=lyrics,
|
||||||
|
instrumental=to_bool(get_param("instrumental"), False),
|
||||||
|
bpm=int(float(raw_bpm)) if raw_bpm not in (None, "", 0, "0") else None,
|
||||||
|
keyscale=sample_keyscale or get_param("key_scale", "keyscale", "key", default=""),
|
||||||
|
timesignature=sample_timesignature or get_param("time_signature", "timesignature", default=""),
|
||||||
|
duration=float(sample_duration or get_param("audio_duration", "duration", default=-1) or -1),
|
||||||
|
vocal_language=sample_language,
|
||||||
|
inference_steps=int(get_param("inference_steps", default=50) or 50),
|
||||||
|
guidance_scale=float(get_param("guidance_scale", default=7.0) or 7.0),
|
||||||
|
seed=int(get_param("seed", default=-1) or -1),
|
||||||
|
use_adg=to_bool(get_param("use_adg"), False),
|
||||||
|
cfg_interval_start=float(get_param("cfg_interval_start", default=0.0) or 0.0),
|
||||||
|
cfg_interval_end=float(get_param("cfg_interval_end", default=1.0) or 1.0),
|
||||||
|
shift=float(get_param("shift", default=1.0) or 1.0),
|
||||||
|
infer_method=get_param("infer_method", default="ode") or "ode",
|
||||||
|
sampler_mode=get_param("sampler_mode", default="euler") or "euler",
|
||||||
|
repainting_start=float(get_param("repainting_start", default=0.0) or 0.0),
|
||||||
|
repainting_end=float(get_param("repainting_end", default=-1.0) or -1.0),
|
||||||
|
chunk_mask_mode=get_param("chunk_mask_mode", default="auto") or "auto",
|
||||||
|
audio_cover_strength=float(get_param("audio_cover_strength", default=1.0) or 1.0),
|
||||||
|
cover_noise_strength=float(get_param("cover_noise_strength", default=0.0) or 0.0),
|
||||||
|
thinking=to_bool(get_param("thinking"), True),
|
||||||
|
lm_temperature=lm_temperature,
|
||||||
|
lm_cfg_scale=float(get_param("lm_cfg_scale", default=2.0) or 2.0),
|
||||||
|
lm_top_k=int(get_param("lm_top_k", default=0) or 0),
|
||||||
|
lm_top_p=float(get_param("lm_top_p", default=0.9) or 0.9),
|
||||||
|
lm_negative_prompt=get_param("lm_negative_prompt", default="NO USER INPUT") or "NO USER INPUT",
|
||||||
|
use_cot_metas=to_bool(get_param("use_cot_metas"), True),
|
||||||
|
use_cot_caption=to_bool(get_param("use_cot_caption"), True),
|
||||||
|
use_cot_lyrics=to_bool(get_param("use_cot_lyrics"), False),
|
||||||
|
use_cot_language=to_bool(get_param("use_cot_language"), True),
|
||||||
|
use_constrained_decoding=to_bool(get_param("use_constrained_decoding"), True),
|
||||||
|
enable_normalization=to_bool(get_param("enable_normalization"), True),
|
||||||
|
normalization_db=float(get_param("normalization_db", default=-1.0) or -1.0),
|
||||||
|
fade_in_duration=float(get_param("fade_in_duration", default=0.0) or 0.0),
|
||||||
|
fade_out_duration=float(get_param("fade_out_duration", default=0.0) or 0.0),
|
||||||
|
latent_shift=float(get_param("latent_shift", default=0.0) or 0.0),
|
||||||
|
latent_rescale=float(get_param("latent_rescale", default=1.0) or 1.0),
|
||||||
|
repaint_latent_crossfade_frames=int(get_param("repaint_latent_crossfade_frames", default=10) or 10),
|
||||||
|
repaint_wav_crossfade_sec=float(get_param("repaint_wav_crossfade_sec", default=0.0) or 0.0),
|
||||||
|
repaint_mode=get_param("repaint_mode", default="balanced") or "balanced",
|
||||||
|
repaint_strength=float(get_param("repaint_strength", default=0.5) or 0.5),
|
||||||
|
)
|
||||||
|
'''
|
||||||
|
|
||||||
|
replace_once(old_params, new_params, "GenerationParams mapping")
|
||||||
|
|
||||||
|
old_config = ''' config = GenerationConfig(
|
||||||
|
batch_size=get_param("batch_size", default=2),
|
||||||
|
use_random_seed=use_random_seed,
|
||||||
|
seeds=resolved_seeds,
|
||||||
|
audio_format=get_param("audio_format", default="flac"),
|
||||||
|
mp3_bitrate=get_param("mp3_bitrate", default="128k"),
|
||||||
|
mp3_sample_rate=get_param("mp3_sample_rate", default=48000),
|
||||||
|
)
|
||||||
|
'''
|
||||||
|
|
||||||
|
new_config = ''' config = GenerationConfig(
|
||||||
|
batch_size=int(get_param("batch_size", default=1) or 1),
|
||||||
|
allow_lm_batch=to_bool(get_param("allow_lm_batch"), True),
|
||||||
|
use_random_seed=to_bool(use_random_seed, True),
|
||||||
|
seeds=resolved_seeds,
|
||||||
|
lm_batch_chunk_size=int(get_param("lm_batch_chunk_size", default=8) or 8),
|
||||||
|
constrained_decoding_debug=to_bool(get_param("constrained_decoding_debug"), False),
|
||||||
|
audio_format=get_param("audio_format", default="flac") or "flac",
|
||||||
|
mp3_bitrate=get_param("mp3_bitrate", default="320k") or "320k",
|
||||||
|
mp3_sample_rate=int(get_param("mp3_sample_rate", default=48000) or 48000),
|
||||||
|
)
|
||||||
|
'''
|
||||||
|
|
||||||
|
replace_once(old_config, new_config, "GenerationConfig mapping")
|
||||||
|
target.write_text(source, encoding="utf-8")
|
||||||
Reference in New Issue
Block a user