@qvac/sdk 0.15.0 → 0.17.0

package/dist/server/bare/plugins/sdcpp-generation/plugin.d.ts

@@ -1,5 +1,5 @@
-import { type CreateModelParams, type PluginModelResult, type ResolveContext, type ResolveResult, type SdcppConfig } from '../../../../schemas/index';
-type DiffusionArtifactKey = 'clipLModelPath' | 'clipGModelPath' | 'clipVisionModelPath' | 't5XxlModelPath' | 'llmModelPath' | 'vaeModelPath' | 'highNoiseDiffusionModelPath' | 'esrganModelPath';
+import { type CreateModelParams, type PluginModelResult, type ResolveContext, type ResolveResult, type SdcppConfig } from '../../../../schemas/index.js';
+type DiffusionArtifactKey = 'clipLModelPath' | 'clipGModelPath' | 'clipVisionModelPath' | 't5XxlModelPath' | 'llmModelPath' | 'vaeModelPath' | 'highNoiseDiffusionModelPath' | 'uncondModelPath' | 'audioVaeModelPath' | 'embeddingsConnectorsModelPath' | 'esrganModelPath';
 /**
  * Stable-diffusion.cpp plugin for image diffusion, upscaling, and Wan video.
  *
@@ -23,6 +23,10 @@ export declare const diffusionPlugin: {
             gpu: "gpu";
             cpu: "cpu";
         }>>;
+        'main-gpu': import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodNumber, import("zod").ZodEnum<{
+            integrated: "integrated";
+            dedicated: "dedicated";
+        }>]>>;
         prediction: import("zod").ZodOptional<import("zod").ZodEnum<{
             auto: "auto";
             eps: "eps";
@@ -83,6 +87,7 @@ export declare const diffusionPlugin: {
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
                 embeddings: "embeddings";
                 diffusion: "diffusion";
+                audiogen: "audiogen";
                 tts: "tts";
                 ocr: "ocr";
                 vla: "vla";
@@ -96,6 +101,7 @@ export declare const diffusionPlugin: {
                 "parakeet-transcription": "parakeet-transcription";
                 "ggml-ocr": "ggml-ocr";
                 "sdcpp-generation": "sdcpp-generation";
+                "audiogen-ggml": "audiogen-ggml";
                 "ggml-vla": "ggml-vla";
                 "ggml-classification": "ggml-classification";
                 llm: "llm";
@@ -120,6 +126,7 @@ export declare const diffusionPlugin: {
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
                 embeddings: "embeddings";
                 diffusion: "diffusion";
+                audiogen: "audiogen";
                 tts: "tts";
                 ocr: "ocr";
                 vla: "vla";
@@ -133,6 +140,7 @@ export declare const diffusionPlugin: {
                 "parakeet-transcription": "parakeet-transcription";
                 "ggml-ocr": "ggml-ocr";
                 "sdcpp-generation": "sdcpp-generation";
+                "audiogen-ggml": "audiogen-ggml";
                 "ggml-vla": "ggml-vla";
                 "ggml-classification": "ggml-classification";
                 llm: "llm";
@@ -157,6 +165,7 @@ export declare const diffusionPlugin: {
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
                 embeddings: "embeddings";
                 diffusion: "diffusion";
+                audiogen: "audiogen";
                 tts: "tts";
                 ocr: "ocr";
                 vla: "vla";
@@ -170,6 +179,7 @@ export declare const diffusionPlugin: {
                 "parakeet-transcription": "parakeet-transcription";
                 "ggml-ocr": "ggml-ocr";
                 "sdcpp-generation": "sdcpp-generation";
+                "audiogen-ggml": "audiogen-ggml";
                 "ggml-vla": "ggml-vla";
                 "ggml-classification": "ggml-classification";
                 llm: "llm";
@@ -194,6 +204,7 @@ export declare const diffusionPlugin: {
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
                 embeddings: "embeddings";
                 diffusion: "diffusion";
+                audiogen: "audiogen";
                 tts: "tts";
                 ocr: "ocr";
                 vla: "vla";
@@ -207,6 +218,7 @@ export declare const diffusionPlugin: {
                 "parakeet-transcription": "parakeet-transcription";
                 "ggml-ocr": "ggml-ocr";
                 "sdcpp-generation": "sdcpp-generation";
+                "audiogen-ggml": "audiogen-ggml";
                 "ggml-vla": "ggml-vla";
                 "ggml-classification": "ggml-classification";
                 llm: "llm";
@@ -231,6 +243,7 @@ export declare const diffusionPlugin: {
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
                 embeddings: "embeddings";
                 diffusion: "diffusion";
+                audiogen: "audiogen";
                 tts: "tts";
                 ocr: "ocr";
                 vla: "vla";
@@ -244,6 +257,7 @@ export declare const diffusionPlugin: {
                 "parakeet-transcription": "parakeet-transcription";
                 "ggml-ocr": "ggml-ocr";
                 "sdcpp-generation": "sdcpp-generation";
+                "audiogen-ggml": "audiogen-ggml";
                 "ggml-vla": "ggml-vla";
                 "ggml-classification": "ggml-classification";
                 llm: "llm";
@@ -268,6 +282,7 @@ export declare const diffusionPlugin: {
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
                 embeddings: "embeddings";
                 diffusion: "diffusion";
+                audiogen: "audiogen";
                 tts: "tts";
                 ocr: "ocr";
                 vla: "vla";
@@ -281,6 +296,46 @@ export declare const diffusionPlugin: {
                 "parakeet-transcription": "parakeet-transcription";
                 "ggml-ocr": "ggml-ocr";
                 "sdcpp-generation": "sdcpp-generation";
+                "audiogen-ggml": "audiogen-ggml";
+                "ggml-vla": "ggml-vla";
+                "ggml-classification": "ggml-classification";
+                llm: "llm";
+                whisper: "whisper";
+                bci: "bci";
+                nmt: "nmt";
+                parakeet: "parakeet";
+                classification: "classification";
+            }>, import("zod").ZodLiteral<"vad">]>>;
+        }, import("zod/v4/core").$strip>]>>;
+        uncondModelSrc: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodObject<{
+            src: import("zod").ZodString;
+            name: import("zod").ZodOptional<import("zod").ZodString>;
+            modelId: import("zod").ZodOptional<import("zod").ZodString>;
+            registryPath: import("zod").ZodOptional<import("zod").ZodString>;
+            registrySource: import("zod").ZodOptional<import("zod").ZodString>;
+            blobCoreKey: import("zod").ZodOptional<import("zod").ZodString>;
+            blobIndex: import("zod").ZodOptional<import("zod").ZodNumber>;
+            engine: import("zod").ZodOptional<import("zod").ZodString>;
+            expectedSize: import("zod").ZodOptional<import("zod").ZodNumber>;
+            sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
+            addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
+                embeddings: "embeddings";
+                diffusion: "diffusion";
+                audiogen: "audiogen";
+                tts: "tts";
+                ocr: "ocr";
+                vla: "vla";
+                "llamacpp-completion": "llamacpp-completion";
+                "whispercpp-transcription": "whispercpp-transcription";
+                "bci-whispercpp-transcription": "bci-whispercpp-transcription";
+                "llamacpp-embedding": "llamacpp-embedding";
+                "nmtcpp-translation": "nmtcpp-translation";
+                "onnx-tts": "onnx-tts";
+                "tts-ggml": "tts-ggml";
+                "parakeet-transcription": "parakeet-transcription";
+                "ggml-ocr": "ggml-ocr";
+                "sdcpp-generation": "sdcpp-generation";
+                "audiogen-ggml": "audiogen-ggml";
                 "ggml-vla": "ggml-vla";
                 "ggml-classification": "ggml-classification";
                 llm: "llm";
@@ -305,6 +360,85 @@ export declare const diffusionPlugin: {
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
                 embeddings: "embeddings";
                 diffusion: "diffusion";
+                audiogen: "audiogen";
+                tts: "tts";
+                ocr: "ocr";
+                vla: "vla";
+                "llamacpp-completion": "llamacpp-completion";
+                "whispercpp-transcription": "whispercpp-transcription";
+                "bci-whispercpp-transcription": "bci-whispercpp-transcription";
+                "llamacpp-embedding": "llamacpp-embedding";
+                "nmtcpp-translation": "nmtcpp-translation";
+                "onnx-tts": "onnx-tts";
+                "tts-ggml": "tts-ggml";
+                "parakeet-transcription": "parakeet-transcription";
+                "ggml-ocr": "ggml-ocr";
+                "sdcpp-generation": "sdcpp-generation";
+                "audiogen-ggml": "audiogen-ggml";
+                "ggml-vla": "ggml-vla";
+                "ggml-classification": "ggml-classification";
+                llm: "llm";
+                whisper: "whisper";
+                bci: "bci";
+                nmt: "nmt";
+                parakeet: "parakeet";
+                classification: "classification";
+            }>, import("zod").ZodLiteral<"vad">]>>;
+        }, import("zod/v4/core").$strip>]>>;
+        audioVaeModelSrc: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodObject<{
+            src: import("zod").ZodString;
+            name: import("zod").ZodOptional<import("zod").ZodString>;
+            modelId: import("zod").ZodOptional<import("zod").ZodString>;
+            registryPath: import("zod").ZodOptional<import("zod").ZodString>;
+            registrySource: import("zod").ZodOptional<import("zod").ZodString>;
+            blobCoreKey: import("zod").ZodOptional<import("zod").ZodString>;
+            blobIndex: import("zod").ZodOptional<import("zod").ZodNumber>;
+            engine: import("zod").ZodOptional<import("zod").ZodString>;
+            expectedSize: import("zod").ZodOptional<import("zod").ZodNumber>;
+            sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
+            addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
+                embeddings: "embeddings";
+                diffusion: "diffusion";
+                audiogen: "audiogen";
+                tts: "tts";
+                ocr: "ocr";
+                vla: "vla";
+                "llamacpp-completion": "llamacpp-completion";
+                "whispercpp-transcription": "whispercpp-transcription";
+                "bci-whispercpp-transcription": "bci-whispercpp-transcription";
+                "llamacpp-embedding": "llamacpp-embedding";
+                "nmtcpp-translation": "nmtcpp-translation";
+                "onnx-tts": "onnx-tts";
+                "tts-ggml": "tts-ggml";
+                "parakeet-transcription": "parakeet-transcription";
+                "ggml-ocr": "ggml-ocr";
+                "sdcpp-generation": "sdcpp-generation";
+                "audiogen-ggml": "audiogen-ggml";
+                "ggml-vla": "ggml-vla";
+                "ggml-classification": "ggml-classification";
+                llm: "llm";
+                whisper: "whisper";
+                bci: "bci";
+                nmt: "nmt";
+                parakeet: "parakeet";
+                classification: "classification";
+            }>, import("zod").ZodLiteral<"vad">]>>;
+        }, import("zod/v4/core").$strip>]>>;
+        embeddingsConnectorsModelSrc: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodObject<{
+            src: import("zod").ZodString;
+            name: import("zod").ZodOptional<import("zod").ZodString>;
+            modelId: import("zod").ZodOptional<import("zod").ZodString>;
+            registryPath: import("zod").ZodOptional<import("zod").ZodString>;
+            registrySource: import("zod").ZodOptional<import("zod").ZodString>;
+            blobCoreKey: import("zod").ZodOptional<import("zod").ZodString>;
+            blobIndex: import("zod").ZodOptional<import("zod").ZodNumber>;
+            engine: import("zod").ZodOptional<import("zod").ZodString>;
+            expectedSize: import("zod").ZodOptional<import("zod").ZodNumber>;
+            sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
+            addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
+                embeddings: "embeddings";
+                diffusion: "diffusion";
+                audiogen: "audiogen";
                 tts: "tts";
                 ocr: "ocr";
                 vla: "vla";
@@ -318,6 +452,7 @@ export declare const diffusionPlugin: {
                 "parakeet-transcription": "parakeet-transcription";
                 "ggml-ocr": "ggml-ocr";
                 "sdcpp-generation": "sdcpp-generation";
+                "audiogen-ggml": "audiogen-ggml";
                 "ggml-vla": "ggml-vla";
                 "ggml-classification": "ggml-classification";
                 llm: "llm";
@@ -344,6 +479,7 @@ export declare const diffusionPlugin: {
                 addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
                     embeddings: "embeddings";
                     diffusion: "diffusion";
+                    audiogen: "audiogen";
                     tts: "tts";
                     ocr: "ocr";
                     vla: "vla";
@@ -357,6 +493,7 @@ export declare const diffusionPlugin: {
                     "parakeet-transcription": "parakeet-transcription";
                     "ggml-ocr": "ggml-ocr";
                     "sdcpp-generation": "sdcpp-generation";
+                    "audiogen-ggml": "audiogen-ggml";
                     "ggml-vla": "ggml-vla";
                     "ggml-classification": "ggml-classification";
                     llm: "llm";
@@ -376,7 +513,7 @@ export declare const diffusionPlugin: {
     resolveConfig(cfg: SdcppConfig, ctx: ResolveContext): Promise<ResolveResult<SdcppConfig, DiffusionArtifactKey>>;
     createModel(params: CreateModelParams): PluginModelResult;
     handlers: {
-        diffusionStream: import("../../../../schemas/index").PluginHandlerDefinition<import("zod").ZodObject<{
+        diffusionStream: import("../../../../schemas/index.js").PluginHandlerDefinition<import("zod").ZodObject<{
             modelId: import("zod").ZodString;
             prompt: import("zod").ZodString;
             negative_prompt: import("zod").ZodOptional<import("zod").ZodString>;
@@ -439,6 +576,11 @@ export declare const diffusionPlugin: {
             stats: import("zod").ZodOptional<import("zod").ZodObject<{
                 modelLoadMs: import("zod").ZodOptional<import("zod").ZodNumber>;
                 generationMs: import("zod").ZodOptional<import("zod").ZodNumber>;
+                conditionerMs: import("zod").ZodOptional<import("zod").ZodNumber>;
+                denoiseMs: import("zod").ZodOptional<import("zod").ZodNumber>;
+                vaeMs: import("zod").ZodOptional<import("zod").ZodNumber>;
+                postProcessMs: import("zod").ZodOptional<import("zod").ZodNumber>;
+                stepsPerSecond: import("zod").ZodOptional<import("zod").ZodNumber>;
                 totalGenerationMs: import("zod").ZodOptional<import("zod").ZodNumber>;
                 totalWallMs: import("zod").ZodOptional<import("zod").ZodNumber>;
                 totalSteps: import("zod").ZodOptional<import("zod").ZodNumber>;
@@ -450,7 +592,7 @@ export declare const diffusionPlugin: {
                 seed: import("zod").ZodOptional<import("zod").ZodNumber>;
             }, import("zod/v4/core").$strip>>;
         }, import("zod/v4/core").$strip>>;
-        videoStream: import("../../../../schemas/index").PluginHandlerDefinition<import("zod").ZodObject<{
+        videoStream: import("../../../../schemas/index.js").PluginHandlerDefinition<import("zod").ZodObject<{
             modelId: import("zod").ZodString;
             requestId: import("zod").ZodOptional<import("zod").ZodString>;
             prompt: import("zod").ZodString;
@@ -530,6 +672,7 @@ export declare const diffusionPlugin: {
             vae_tiling: import("zod").ZodOptional<import("zod").ZodBoolean>;
             vae_tile_size: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodNumber, import("zod").ZodString]>>;
             vae_tile_overlap: import("zod").ZodOptional<import("zod").ZodNumber>;
+            temporal_tiling: import("zod").ZodOptional<import("zod").ZodBoolean>;
             cache_mode: import("zod").ZodOptional<import("zod").ZodEnum<{
                 disabled: "disabled";
                 easycache: "easycache";
@@ -560,6 +703,11 @@ export declare const diffusionPlugin: {
                 height: import("zod").ZodOptional<import("zod").ZodNumber>;
                 modelLoadMs: import("zod").ZodOptional<import("zod").ZodNumber>;
                 generationMs: import("zod").ZodOptional<import("zod").ZodNumber>;
+                conditionerMs: import("zod").ZodOptional<import("zod").ZodNumber>;
+                denoiseMs: import("zod").ZodOptional<import("zod").ZodNumber>;
+                vaeMs: import("zod").ZodOptional<import("zod").ZodNumber>;
+                postProcessMs: import("zod").ZodOptional<import("zod").ZodNumber>;
+                stepsPerSecond: import("zod").ZodOptional<import("zod").ZodNumber>;
                 totalGenerationMs: import("zod").ZodOptional<import("zod").ZodNumber>;
                 totalWallMs: import("zod").ZodOptional<import("zod").ZodNumber>;
                 totalSteps: import("zod").ZodOptional<import("zod").ZodNumber>;
@@ -570,9 +718,11 @@ export declare const diffusionPlugin: {
                 totalVideoFrames: import("zod").ZodOptional<import("zod").ZodNumber>;
                 videoFrames: import("zod").ZodOptional<import("zod").ZodNumber>;
                 fps: import("zod").ZodOptional<import("zod").ZodNumber>;
+                hasAudio: import("zod").ZodOptional<import("zod").ZodBoolean>;
+                audioSampleRate: import("zod").ZodOptional<import("zod").ZodNumber>;
             }, import("zod/v4/core").$strip>>;
         }, import("zod/v4/core").$strip>>;
-        upscaleStream: import("../../../../schemas/index").PluginHandlerDefinition<import("zod").ZodObject<{
+        upscaleStream: import("../../../../schemas/index.js").PluginHandlerDefinition<import("zod").ZodObject<{
             modelId: import("zod").ZodString;
             image: import("zod").ZodString;
             repeats: import("zod").ZodOptional<import("zod").ZodNumber>;

package/dist/server/bare/plugins/sdcpp-generation/plugin.js

@@ -1,10 +1,12 @@
 import ImgStableDiffusion, { EsrganUpscaler, VideoStableDiffusion } from '@qvac/diffusion-cpp';
-import addonLogging from '@qvac/diffusion-cpp/addonLogging.js';
+import addonLogging from '@qvac/diffusion-cpp/addonLogging';
 import { definePlugin, defineHandler, sdcppConfigSchema, diffusionRequestSchema, diffusionStreamResponseSchema, videoRequestSchema, videoStreamResponseSchema, upscaleRequestSchema, upscaleStreamResponseSchema, ModelType, ADDON_DIFFUSION } from '../../../../schemas/index.js';
-import { createStreamLogger, registerAddonLogger } from '../../../../logging/index.js';
+import { createStreamLogger, registerAddonLogger, getServerLogger } from '../../../../logging/index.js';
 import { ModelLoadFailedError } from '../../../../utils/errors-server.js';
+import { isMobile } from '../../../../server/bare/registry/runtime-context-registry.js';
+import { stripMultiGpuKeys } from '../../../../server/utils/multi-gpu-mobile.js';
 import { diffusion } from './ops/diffusion.js';
-import { video } from './ops/video.js';
+import { markLtxVideoModel, markMoeCapableVideoModel, video } from './ops/video.js';
 import { upscale } from './ops/upscale.js';
 // Single source of truth for `SdcppConfig.upscaler.*` → addon-config key
 // mapping. Used by both the diffusion-mode (post-generation upscaler) and the
@@ -52,13 +54,48 @@ export const diffusionPlugin = definePlugin({
     addonPackage: ADDON_DIFFUSION,
     loadConfigSchema: sdcppConfigSchema,
     async resolveConfig(cfg, ctx) {
+        if (cfg.uncondModelSrc && cfg.mode !== 'diffusion') {
+            throw new ModelLoadFailedError('modelConfig.uncondModelSrc is Ideogram 4 diffusion only. ' +
+                "Use mode: 'diffusion' or remove uncondModelSrc.");
+        }
+        if (cfg.uncondModelSrc && (!cfg.llmModelSrc || !cfg.vaeModelSrc)) {
+            throw new ModelLoadFailedError('modelConfig.uncondModelSrc selects the Ideogram 4 layout and requires ' +
+                'modelConfig.llmModelSrc (Qwen3-VL) and modelConfig.vaeModelSrc.');
+        }
+        // A high-noise expert only has somewhere to go in the Wan 2.2 A14B video
+        // layout: every other mode builds a model with no second-expert slot, and
+        // embeddingsConnectorsModelSrc selects LTX-2 instead. Check it ahead of the
+        // upscale early-return so a wrong layout fails loudly rather than
+        // downloading a multi-gigabyte expert and dropping it in createModel.
+        const isWan22MoeLayout = cfg.mode === 'video' && !cfg.embeddingsConnectorsModelSrc;
+        if (cfg.highNoiseDiffusionModelSrc && !isWan22MoeLayout) {
+            throw new ModelLoadFailedError('modelConfig.highNoiseDiffusionModelSrc selects the Wan 2.2 A14B ' +
+                "mixture-of-experts video layout. It requires mode: 'video' and " +
+                'cannot be combined with modelConfig.embeddingsConnectorsModelSrc ' +
+                '(which selects the LTX-2 layout).');
+        }
         // Standalone-upscaler mode never references auxiliary models: the primary
         // modelSrc IS the ESRGAN file. Skip resolution to avoid downloading
         // unused encoders/VAEs and to keep load fast.
         if (cfg.mode === 'upscale') {
             return { config: cfg };
         }
-        const { clipLModelSrc, clipGModelSrc, clipVisionModelSrc, t5XxlModelSrc, llmModelSrc, vaeModelSrc, highNoiseDiffusionModelSrc, upscaler, ...rest } = cfg;
+        const { clipLModelSrc, clipGModelSrc, clipVisionModelSrc, t5XxlModelSrc, llmModelSrc, vaeModelSrc, highNoiseDiffusionModelSrc, uncondModelSrc, audioVaeModelSrc, embeddingsConnectorsModelSrc, upscaler, ...rest } = cfg;
+        // audioVae / embeddingsConnectors are LTX-2 video companions. Reject them
+        // up front — before any companion download — when the selected layout
+        // can't consume them, instead of silently resolving the file and then
+        // dropping it in createModel. Both fields are new, so this cannot break
+        // existing callers.
+        if (embeddingsConnectorsModelSrc && cfg.mode !== 'video') {
+            throw new ModelLoadFailedError('modelConfig.embeddingsConnectorsModelSrc selects the LTX-2 video ' +
+                "layout and is only valid with mode: 'video'.");
+        }
+        if (audioVaeModelSrc && !(cfg.mode === 'video' && embeddingsConnectorsModelSrc)) {
+            throw new ModelLoadFailedError('modelConfig.audioVaeModelSrc is LTX-2 video only. It requires ' +
+                "mode: 'video' together with modelConfig.embeddingsConnectorsModelSrc " +
+                '(which selects the LTX-2 layout). Add embeddingsConnectorsModelSrc ' +
+                'or remove audioVaeModelSrc.');
+        }
         // Video jobs do not apply ESRGAN so we drop the whole `upscaler` object.
         const effectiveUpscaler = cfg.mode === 'video' ? undefined : upscaler;
         const { model_src: esrganModelSrc, ...upscalerRuntime } = effectiveUpscaler ?? {};
@@ -74,6 +111,9 @@ export const diffusionPlugin = definePlugin({
             llmModelSrc,
             vaeModelSrc,
             highNoiseDiffusionModelSrc,
+            uncondModelSrc,
+            audioVaeModelSrc,
+            embeddingsConnectorsModelSrc,
             esrganModelSrc
         };
         const hasSources = Object.values(sources).some(Boolean);
@@ -81,7 +121,7 @@ export const diffusionPlugin = definePlugin({
             return { config: runtimeConfig };
         }
         const resolve = ctx.resolveModelPath;
-        const [clipLModelPath, clipGModelPath, clipVisionModelPath, t5XxlModelPath, llmModelPath, vaeModelPath, highNoiseDiffusionModelPath, esrganModelPath] = await Promise.all([
+        const [clipLModelPath, clipGModelPath, clipVisionModelPath, t5XxlModelPath, llmModelPath, vaeModelPath, highNoiseDiffusionModelPath, uncondModelPath, audioVaeModelPath, embeddingsConnectorsModelPath, esrganModelPath] = await Promise.all([
             clipLModelSrc ? resolve(clipLModelSrc) : undefined,
             clipGModelSrc ? resolve(clipGModelSrc) : undefined,
             clipVisionModelSrc ? resolve(clipVisionModelSrc) : undefined,
@@ -89,6 +129,9 @@ export const diffusionPlugin = definePlugin({
             llmModelSrc ? resolve(llmModelSrc) : undefined,
             vaeModelSrc ? resolve(vaeModelSrc) : undefined,
             highNoiseDiffusionModelSrc ? resolve(highNoiseDiffusionModelSrc) : undefined,
+            uncondModelSrc ? resolve(uncondModelSrc) : undefined,
+            audioVaeModelSrc ? resolve(audioVaeModelSrc) : undefined,
+            embeddingsConnectorsModelSrc ? resolve(embeddingsConnectorsModelSrc) : undefined,
             esrganModelSrc ? resolve(esrganModelSrc) : undefined
         ]);
         return {
@@ -101,6 +144,9 @@ export const diffusionPlugin = definePlugin({
                 ...(llmModelPath && { llmModelPath }),
                 ...(vaeModelPath && { vaeModelPath }),
                 ...(highNoiseDiffusionModelPath && { highNoiseDiffusionModelPath }),
+                ...(uncondModelPath && { uncondModelPath }),
+                ...(audioVaeModelPath && { audioVaeModelPath }),
+                ...(embeddingsConnectorsModelPath && { embeddingsConnectorsModelPath }),
                 ...(esrganModelPath && { esrganModelPath })
             }
         };
@@ -108,6 +154,13 @@ export const diffusionPlugin = definePlugin({
     createModel(params) {
         const { modelId, modelPath, modelConfig, artifacts } = params;
         const config = (modelConfig ?? {});
+        // no multi-gpu on mobile
+        if (isMobile()) {
+            const stripped = stripMultiGpuKeys(config);
+            if (stripped.length > 0) {
+                getServerLogger().warn(`[${ModelType.sdcppGeneration}:${modelId}] Multi-GPU parameters (${stripped.join(', ')}) are not supported on mobile (single-GPU device) — removing from config; model will load with single-GPU defaults`);
+            }
+        }
         // In diffusion mode the ESRGAN file (when post-generation upscale is
         // wanted) must come from upscaler.model_src — the primary modelPath is
         // the main diffusion checkpoint. Reject early with a clear error
@@ -133,28 +186,54 @@ export const diffusionPlugin = definePlugin({
             return { model };
         }
         if (config.mode === 'video') {
-            if (!artifacts?.['t5XxlModelPath']) {
-                throw new ModelLoadFailedError('modelConfig.t5XxlModelSrc is required in video mode. ' +
-                    'Provide the Wan text encoder model before loading the video pipeline.');
-            }
+            // Layout is selected the same way the addon self-detects it: the
+            // presence of the LTX-2 text-embedding connectors switches from the
+            // Wan layout (UMT5 via t5Xxl) to the LTX-2 layout (Gemma via llm +
+            // video VAE + connectors, optional audio VAE). Mirrors
+            // `SdModel::isLtxModel_ = !embeddingsConnectorsPath.empty()`.
+            const embeddingsConnectorsModelPath = artifacts?.['embeddingsConnectorsModelPath'];
             if (!artifacts?.['vaeModelPath']) {
                 throw new ModelLoadFailedError('modelConfig.vaeModelSrc is required in video mode. ' +
-                    'Provide the Wan VAE model before loading the video pipeline.');
+                    'Provide the Wan or LTX-2 video VAE model before loading the video pipeline.');
+            }
+            const vaeModelPath = artifacts['vaeModelPath'];
+            let files;
+            if (embeddingsConnectorsModelPath) {
+                if (!artifacts['llmModelPath']) {
+                    throw new ModelLoadFailedError('modelConfig.llmModelSrc is required for LTX-2 video. ' +
+                        'Provide the Gemma text encoder model before loading the LTX-2 pipeline.');
+                }
+                files = {
+                    model: modelPath,
+                    vae: vaeModelPath,
+                    llm: artifacts['llmModelPath'],
+                    embeddingsConnectors: embeddingsConnectorsModelPath,
+                    ...(artifacts['audioVaeModelPath'] && {
+                        audioVae: artifacts['audioVaeModelPath']
+                    }),
+                    ...(artifacts['esrganModelPath'] && { esrgan: artifacts['esrganModelPath'] })
+                };
+            }
+            else {
+                if (!artifacts['t5XxlModelPath']) {
+                    throw new ModelLoadFailedError('modelConfig.t5XxlModelSrc is required in video mode. ' +
+                        'Provide the Wan text encoder model before loading the video pipeline.');
+                }
+                files = {
+                    model: modelPath,
+                    vae: vaeModelPath,
+                    t5Xxl: artifacts['t5XxlModelPath'],
+                    ...(artifacts['highNoiseDiffusionModelPath'] && {
+                        highNoiseDiffusionModel: artifacts['highNoiseDiffusionModelPath']
+                    }),
+                    ...(artifacts['clipVisionModelPath'] && {
+                        clipVision: artifacts['clipVisionModelPath']
+                    }),
+                    ...(artifacts['esrganModelPath'] && { esrgan: artifacts['esrganModelPath'] })
+                };
             }
-            const files = {
-                model: modelPath,
-                t5Xxl: artifacts['t5XxlModelPath'],
-                vae: artifacts['vaeModelPath'],
-                ...(artifacts?.['highNoiseDiffusionModelPath'] && {
-                    highNoiseDiffusionModel: artifacts['highNoiseDiffusionModelPath']
-                }),
-                ...(artifacts?.['clipVisionModelPath'] && {
-                    clipVision: artifacts['clipVisionModelPath']
-                }),
-                ...(artifacts?.['esrganModelPath'] && { esrgan: artifacts['esrganModelPath'] })
-            };
             /* eslint-disable @typescript-eslint/no-unused-vars */
-            const { clipLModelSrc, clipGModelSrc, clipVisionModelSrc, t5XxlModelSrc, llmModelSrc, vaeModelSrc, highNoiseDiffusionModelSrc, upscaler, mode, ...rest } = config;
+            const { clipLModelSrc, clipGModelSrc, clipVisionModelSrc, t5XxlModelSrc, llmModelSrc, vaeModelSrc, highNoiseDiffusionModelSrc, uncondModelSrc, audioVaeModelSrc, embeddingsConnectorsModelSrc, upscaler, mode, ...rest } = config;
             /* eslint-enable @typescript-eslint/no-unused-vars */
             const model = new VideoStableDiffusion({
                 files,
@@ -162,6 +241,10 @@ export const diffusionPlugin = definePlugin({
                 logger,
                 opts: { stats: true }
             });
+            if (embeddingsConnectorsModelPath)
+                markLtxVideoModel(model);
+            if (files.highNoiseDiffusionModel)
+                markMoeCapableVideoModel(model);
             return { model };
         }
         const files = {
@@ -171,6 +254,9 @@ export const diffusionPlugin = definePlugin({
             ...(artifacts?.['t5XxlModelPath'] && { t5Xxl: artifacts['t5XxlModelPath'] }),
             ...(artifacts?.['llmModelPath'] && { llm: artifacts['llmModelPath'] }),
             ...(artifacts?.['vaeModelPath'] && { vae: artifacts['vaeModelPath'] }),
+            ...(artifacts?.['uncondModelPath'] && {
+                uncondModel: artifacts['uncondModelPath']
+            }),
             ...(artifacts?.['esrganModelPath'] && { esrgan: artifacts['esrganModelPath'] })
         };
         // `mode` is consumed by this plugin to select the model class above; the

package/dist/server/bare/plugins/tts-ggml/ops/parler-options.d.ts

@@ -0,0 +1,7 @@
+import type { TextToSpeechStreamRequest, TtsRequest } from '../../../../../schemas/index.js';
+export type ParlerJobOptions = Pick<TtsRequest, 'description' | 'voiceDescription' | 'voice' | 'emotion' | 'pitch' | 'pace' | 'expressivity' | 'noise' | 'reverb' | 'quality'>;
+type ParlerJobOptionsRequest = TtsRequest | TextToSpeechStreamRequest;
+export declare function getParlerJobOptions(request: ParlerJobOptionsRequest): ParlerJobOptions;
+export declare function assertParlerJobOptionsSupported(model: unknown, options: ParlerJobOptions, handlerName: 'textToSpeech' | 'textToSpeechStream'): void;
+export {};
+//# sourceMappingURL=parler-options.d.ts.map

package/dist/server/bare/plugins/tts-ggml/ops/parler-options.js

@@ -0,0 +1,30 @@
+import { PluginRequestValidationFailedError } from '../../../../../utils/errors-server.js';
+function hasEngineType(model) {
+    return (typeof model === 'object' &&
+        model !== null &&
+        'getEngineType' in model &&
+        typeof model.getEngineType === 'function');
+}
+export function getParlerJobOptions(request) {
+    return {
+        ...(request.description !== undefined ? { description: request.description } : {}),
+        ...(request.voiceDescription !== undefined
+            ? { voiceDescription: request.voiceDescription }
+            : {}),
+        ...(request.voice !== undefined ? { voice: request.voice } : {}),
+        ...(request.emotion !== undefined ? { emotion: request.emotion } : {}),
+        ...(request.pitch !== undefined ? { pitch: request.pitch } : {}),
+        ...(request.pace !== undefined ? { pace: request.pace } : {}),
+        ...(request.expressivity !== undefined ? { expressivity: request.expressivity } : {}),
+        ...(request.noise !== undefined ? { noise: request.noise } : {}),
+        ...(request.reverb !== undefined ? { reverb: request.reverb } : {}),
+        ...(request.quality !== undefined ? { quality: request.quality } : {})
+    };
+}
+export function assertParlerJobOptionsSupported(model, options, handlerName) {
+    if (Object.keys(options).length === 0)
+        return;
+    if (hasEngineType(model) && model.getEngineType() === 'parler')
+        return;
+    throw new PluginRequestValidationFailedError(handlerName, 'description, voiceDescription, voice, emotion, pitch, pace, expressivity, noise, reverb, and quality are only supported by Parler TTS models');
+}

package/dist/server/bare/plugins/tts-ggml/ops/text-to-speech-stream.d.ts

@@ -1,5 +1,5 @@
-import { type TextToSpeechStreamRequest, type TtsStats } from '../../../../../schemas/index';
-import { type TtsOpYield } from '../../../utils/tts-stats';
+import { type TextToSpeechStreamRequest, type TtsStats } from '../../../../../schemas/index.js';
+import { type TtsOpYield } from '../../../../../server/bare/utils/tts-stats.js';
 export declare function textToSpeechStream(params: TextToSpeechStreamRequest, inputStream: AsyncIterable<Buffer>): AsyncGenerator<TtsOpYield, {
     modelExecutionMs: number;
     stats?: TtsStats;

package/dist/server/bare/plugins/tts-ggml/ops/text-to-speech-stream.js

@@ -1,9 +1,10 @@
-import { getModel } from '../../../registry/model-registry.js';
+import { getModel } from '../../../../../server/bare/registry/model-registry.js';
 import { textToSpeechStreamRequestSchema } from '../../../../../schemas/index.js';
 import { nowMs } from '../../../../../profiling/index.js';
 import { buildStreamResult, hasDefinedValues } from '../../../../../profiling/model-execution.js';
 import { TextToSpeechStreamFailedError } from '../../../../../utils/errors-server.js';
-import { collectTtsStats } from '../../../utils/tts-stats.js';
+import { collectTtsStats } from '../../../../../server/bare/utils/tts-stats.js';
+import { assertParlerJobOptionsSupported, getParlerJobOptions } from '../../../../../server/bare/plugins/tts-ggml/ops/parler-options.js';
 function hasRunStreaming(model) {
     return (typeof model === 'object' &&
         model !== null &&
@@ -53,7 +54,7 @@ async function* buffersToUtf8Fragments(inputStream) {
         }
     }
 }
-function buildRunStreamingOptions(request) {
+function buildRunStreamingOptions(request, parlerJobOptions) {
     const o = {};
     if (request.accumulateSentences !== undefined) {
         o['accumulateSentences'] = request.accumulateSentences;
@@ -67,17 +68,19 @@ function buildRunStreamingOptions(request) {
     if (request.flushAfterMs !== undefined) {
         o['flushAfterMs'] = request.flushAfterMs;
     }
-    return o;
+    return { ...o, ...parlerJobOptions };
 }
 export async function* textToSpeechStream(params, inputStream) {
     const request = textToSpeechStreamRequestSchema.parse(params);
     const model = getModel(request.modelId);
+    const parlerJobOptions = getParlerJobOptions(request);
+    assertParlerJobOptionsSupported(model, parlerJobOptions, 'textToSpeechStream');
     const modelStart = nowMs();
     if (!hasRunStreaming(model)) {
         throw new TextToSpeechStreamFailedError('textToSpeechStream requires a TTS model with runStreaming');
     }
     const textSource = buffersToUtf8Fragments(inputStream);
-    const streamOpts = buildRunStreamingOptions(request);
+    const streamOpts = buildRunStreamingOptions(request, parlerJobOptions);
     const response = await model.runStreaming(textSource, Object.keys(streamOpts).length > 0 ? streamOpts : undefined);
     for await (const data of response.iterate()) {
         if (data.outputArray == null) {

package/dist/server/bare/plugins/tts-ggml/ops/text-to-speech.d.ts

@@ -1,5 +1,5 @@
-import { type TtsRequest, type TtsStats } from '../../../../../schemas/index';
-import { type TtsOpYield } from '../../../utils/tts-stats';
+import { type TtsRequest, type TtsStats } from '../../../../../schemas/index.js';
+import { type TtsOpYield } from '../../../../../server/bare/utils/tts-stats.js';
 export declare function textToSpeech(params: TtsRequest): AsyncGenerator<TtsOpYield, {
     modelExecutionMs: number;
     stats?: TtsStats;