@picsart/ai-sdk 6.6.0 → 6.8.0

package/index.d.ts

@@ -191,20 +191,47 @@ type ModelInputById = {
         audioUrl: string;
         voiceId?: string;
         removeBackgroundNoise?: boolean;
+        stability?: number;
+        similarityBoost?: number;
+        styleExaggeration?: number;
+        speed?: number;
     };
     "eleven-multilingual-v2": {
         prompt: string;
         voiceId?: string;
+        stability?: number;
+        similarityBoost?: number;
+        styleExaggeration?: number;
+        speed?: number;
+        useSpeakerBoost?: boolean;
     };
     "eleven-sts-v2": {
         audioUrl: string;
         voiceId?: string;
         removeBackgroundNoise?: boolean;
+        stability?: number;
+        similarityBoost?: number;
+        styleExaggeration?: number;
+        speed?: number;
+    };
+    "eleven-text-to-dialogue": {
+        dialogue: Array<{
+            voiceId: string;
+            text: string;
+        }>;
+        stability?: 0 | 0.5 | 1;
+        language?: string;
+        seed?: number;
     };
     "eleven-v3": {
         language?: string;
         prompt: string;
         voiceId?: string;
+        stability?: number;
+        similarityBoost?: number;
+        styleExaggeration?: number;
+        speed?: number;
+        useSpeakerBoost?: boolean;
     };
     "eleven-voice-create": {
         prompt: string;
@@ -2042,6 +2069,21 @@ interface FileDescriptor {
      * worker stays the authoritative gate. Omit for no client-side cap.
      */
     maxDurationSec?: number;
+    /**
+     * Min intrinsic duration (seconds) accepted for a video/audio file. Unlike
+     * every other bound here this one has no remedy: a clip that is too short
+     * cannot be lengthened, so the client's only honest move is to refuse the
+     * attachment and say why. Enforced client-side at upload by measuring the
+     * media before it is sent; the backend worker stays the authoritative gate.
+     * Omit for no client-side floor.
+     */
+    minDurationSec?: number;
+    /**
+     * Max intrinsic frame rate (fps) accepted for a video file. Enforced
+     * client-side at upload by measuring the media before it is sent; the backend
+     * worker stays the authoritative gate. Omit for no client-side cap.
+     */
+    maxFrameRate?: number;
     /**
      * Min intrinsic pixel count (width × height) accepted for an image/video file.
      * Enforced client-side at upload by measuring the media before it is sent; the
@@ -2057,6 +2099,41 @@ interface FileDescriptor {
      * authoritative gate. Omit for no client-side floor.
      */
     minSidePixels?: number;
+    /**
+     * Max intrinsic pixel count (width × height) accepted for an image/video
+     * file. The ceiling matching `minPixels`, and a separate rule from
+     * `maxSidePixels`: a 12000×2000 frame passes a 36 MP area cap and still
+     * breaks a 6000 px per-side one. Enforced client-side at upload by measuring
+     * the media before it is sent; the backend worker stays the authoritative
+     * gate. Omit for no client-side ceiling.
+     */
+    maxPixels?: number;
+    /**
+     * Max intrinsic long-side length (pixels) accepted for an image/video file:
+     * `max(width, height)` must not exceed this. The ceiling matching
+     * `minSidePixels`, so the pair expresses the shape vendors usually publish
+     * ("width and height must each be between 300 and 6000 px"). Distinct from
+     * `maxShortSidePixels`, which bounds the OTHER side and exists for upscaler
+     * sources. Enforced client-side at upload by measuring the media before it is
+     * sent; the backend worker stays the authoritative gate. Omit for no
+     * client-side ceiling.
+     */
+    maxSidePixels?: number;
+    /**
+     * Min intrinsic aspect ratio (width / height) accepted for an image/video
+     * file. Vendors publish this as the bound a panorama or a tall crop breaks,
+     * and it is not derivable from the pixel and side bounds: a 6000x1200 frame
+     * satisfies a [300, 6000] per-side rule and a 36 MP area rule and is still
+     * refused at 5.0. Enforced client-side at upload by measuring the media
+     * before it is sent; the backend worker stays the authoritative gate. Omit
+     * for no client-side floor.
+     */
+    minAspectRatio?: number;
+    /**
+     * Max intrinsic aspect ratio (width / height) accepted for an image/video
+     * file. The ceiling matching `minAspectRatio`.
+     */
+    maxAspectRatio?: number;
     /**
      * Max intrinsic short-side length (pixels) accepted for an image/video file —
      * `min(width, height)` must not exceed this. Used by upscalers whose source
@@ -2927,6 +3004,7 @@ declare const Models: {
     readonly ElevenMultilingualStsV2: "eleven-multilingual-sts-v2";
     readonly ElevenMultilingualV2: "eleven-multilingual-v2";
     readonly ElevenStsV2: "eleven-sts-v2";
+    readonly ElevenTextToDialogue: "eleven-text-to-dialogue";
     readonly ElevenV3: "eleven-v3";
     readonly ElevenVoiceCreate: "eleven-voice-create";
     readonly ElevenVoiceDesignV2: "eleven-voice-design-v2";

package/index.js

@@ -927,8 +927,14 @@ var p = {
           accept,
           ...opts?.array ? { array: opts.array } : {},
           ...opts?.maxDurationSec != null ? { maxDurationSec: opts.maxDurationSec } : {},
+          ...opts?.minDurationSec != null ? { minDurationSec: opts.minDurationSec } : {},
+          ...opts?.maxFrameRate != null ? { maxFrameRate: opts.maxFrameRate } : {},
           ...opts?.minPixels != null ? { minPixels: opts.minPixels } : {},
+          ...opts?.maxPixels != null ? { maxPixels: opts.maxPixels } : {},
           ...opts?.minSidePixels != null ? { minSidePixels: opts.minSidePixels } : {},
+          ...opts?.maxSidePixels != null ? { maxSidePixels: opts.maxSidePixels } : {},
+          ...opts?.minAspectRatio != null ? { minAspectRatio: opts.minAspectRatio } : {},
+          ...opts?.maxAspectRatio != null ? { maxAspectRatio: opts.maxAspectRatio } : {},
           ...opts?.maxShortSidePixels != null ? { maxShortSidePixels: opts.maxShortSidePixels } : {},
           ...opts?.maxBytes != null ? { maxBytes: opts.maxBytes } : {}
         }
@@ -1162,50 +1168,37 @@ var params = {
   // `category` defaults to the most common role for the slot (overridable per call):
   //   asset    → start/end frame, sync audio (direct inputs to the output)
   //   reference → ref images/videos/audios (guidance signals)
-  /** Array of image inputs (writes to `imageUrls`). `bounds` carries the
-   *  client-side dimension floors enforced at upload: `minPixels` for a vendor
-   *  rule stated as a total pixel count, `minSidePixels` for one stated per
-   *  side (width and height each), which is what most vendors publish. */
-  imageInput: (max = 1, label = "Start Image", required = false, category = "reference", bounds) => p.file("imageUrls", "image", {
-    array: { max },
-    label,
-    required,
-    category,
-    ...bounds?.minPixels != null ? { minPixels: bounds.minPixels } : {},
-    ...bounds?.minSidePixels != null ? { minSidePixels: bounds.minSidePixels } : {}
-  }),
+  /** Array of image inputs (writes to `imageUrls`). */
+  imageInput: (max = 1, label = "Start Image", required = false, category = "reference", bounds) => p.file("imageUrls", "image", { array: { max }, label, required, category, ...bounds }),
   /** Single source-video slot (v2v / video edit). Writes to `videoUrl`.
-   *  `maxDurationSec` caps the source clip length, `maxShortSidePixels` caps
-   *  the shorter side (upscaler sources) and `maxBytes` caps the file size,
-   *  all enforced client-side at upload. */
-  videoInput: (label = "Source Video", category = "reference", required = true, maxDurationSec, maxShortSidePixels, maxBytes) => p.file("videoUrl", "video", {
+   *  The four positional limits predate `bounds` and are kept so the dozens of
+   *  existing call sites do not churn; new limits go in `bounds`. */
+  videoInput: (label = "Source Video", category = "reference", required = true, maxDurationSec, maxShortSidePixels, maxBytes, bounds) => p.file("videoUrl", "video", {
     label,
     required,
     category,
-    ...maxDurationSec != null ? { maxDurationSec } : {},
-    ...maxShortSidePixels != null ? { maxShortSidePixels } : {},
-    ...maxBytes != null ? { maxBytes } : {}
+    maxDurationSec,
+    maxShortSidePixels,
+    maxBytes,
+    ...bounds
   }),
   /** Single driving / sync-audio slot. Writes to `audioUrl`. */
   audioInput: (label = "Audio Track", required = false, category = "asset") => p.file("audioUrl", "audio", { label, required, category }),
-  /** Array of reference videos (writes to `videoUrls`). Backend enforces
-   *  per-model total-duration caps (e.g. ≤ 15s for seedance). `bounds` carries
-   *  the client-side checks run at upload: `minPixels` for a total-pixel floor,
-   *  `minSidePixels` for a per-side one, `maxBytes` for each clip's file size. */
-  videoInputs: (max = 3, label = "Reference Videos", required = false, bounds) => p.file("videoUrls", "video", {
-    array: { max },
-    label,
-    required,
-    category: "reference",
-    ...bounds?.minPixels != null ? { minPixels: bounds.minPixels } : {},
-    ...bounds?.minSidePixels != null ? { minSidePixels: bounds.minSidePixels } : {},
-    ...bounds?.maxBytes != null ? { maxBytes: bounds.maxBytes } : {}
-  }),
-  /** Array of reference audios (writes to `audioUrls`). Backend enforces
-   *  per-model total-duration caps. */
-  audioInputs: (max = 3, label = "Reference Audios", required = false) => p.file("audioUrls", "audio", { array: { max }, label, required, category: "reference" }),
-  startFrame: (label = "Start Frame", required = false) => p.file("startFrame", "image", { label, required, category: "asset" }),
-  endFrame: (label = "End Frame") => p.file("endFrame", "image", { label, category: "asset" })
+  /** Array of reference videos (writes to `videoUrls`). `bounds` is checked per
+   *  clip. The backend also caps the COMBINED length of the array per model,
+   *  which no descriptor field expresses yet; declaring the per-clip bounds
+   *  still turns the common single-clip case into a trim prompt in the composer
+   *  instead of a vendor `400` after the user pressed Generate. */
+  videoInputs: (max = 3, label = "Reference Videos", required = false, bounds) => p.file("videoUrls", "video", { array: { max }, label, required, category: "reference", ...bounds }),
+  /** Array of reference audios (writes to `audioUrls`). `bounds` is checked per
+   *  clip; as with `videoInputs` the backend also caps the combined length. */
+  audioInputs: (max = 3, label = "Reference Audios", required = false, bounds) => p.file("audioUrls", "audio", { array: { max }, label, required, category: "reference", ...bounds }),
+  /** First-frame slot (writes to `startFrame`). Vendors routinely publish a
+   *  DIFFERENT pixel floor for a frame than for a reference image, because a
+   *  frame becomes the output's first rendered frame. */
+  startFrame: (label = "Start Frame", required = false, bounds) => p.file("startFrame", "image", { label, required, category: "asset", ...bounds }),
+  /** Last-frame slot (writes to `endFrame`). Same `bounds` as `startFrame`. */
+  endFrame: (label = "End Frame", bounds) => p.file("endFrame", "image", { label, category: "asset", ...bounds })
 };
 var paramPresets = {
   videoStartEndWithAudio
@@ -3673,8 +3666,71 @@ registerPayloads(MODELS11, {
 // src/vendors/catalog/seedance.ts
 var SEEDANCE_FRAME_REF_REASON = "Start/End frames cannot be combined with reference images, videos, or audios";
 var SEEDANCE_MIN_SIDE_PIXELS = 300;
+var SEEDANCE_MAX_SIDE_PIXELS = 6e3;
+var SEEDANCE_MIN_ASPECT_RATIO = 0.39;
+var SEEDANCE_MAX_ASPECT_RATIO = 2.5;
+var SEEDANCE_MAX_IMAGE_PIXELS = 36e6;
 var SEEDANCE_VIDEO_MIN_PIXELS = 407696;
-var SEEDANCE_25_MAX_VIDEO_BYTES = 209715200;
+var SEEDANCE_VIDEO_MAX_PIXELS = 8295044;
+var SEEDANCE_MIN_MEDIA_SEC = 1.8;
+var SEEDANCE_MAX_FRAME_RATE = 60;
+var SEEDANCE_MAX_VIDEO_BYTES = 209715200;
+var SEEDANCE_FRAME_MIN_PIXELS = 9e4;
+var SEEDANCE_MAX_IMAGE_BYTES = 31457280;
+var SEEDANCE_MAX_AUDIO_BYTES = 15728640;
+var SEEDANCE_25_MAX_MEDIA_SEC = 30;
+var SEEDANCE_20_MAX_MEDIA_SEC = 15;
+var SEEDANCE_IMAGE_BOUNDS = {
+  minAspectRatio: SEEDANCE_MIN_ASPECT_RATIO,
+  maxAspectRatio: SEEDANCE_MAX_ASPECT_RATIO,
+  minSidePixels: SEEDANCE_MIN_SIDE_PIXELS,
+  maxSidePixels: SEEDANCE_MAX_SIDE_PIXELS,
+  maxPixels: SEEDANCE_MAX_IMAGE_PIXELS,
+  maxBytes: SEEDANCE_MAX_IMAGE_BYTES
+};
+var SEEDANCE_FRAME_BOUNDS = {
+  minAspectRatio: SEEDANCE_MIN_ASPECT_RATIO,
+  maxAspectRatio: SEEDANCE_MAX_ASPECT_RATIO,
+  minPixels: SEEDANCE_FRAME_MIN_PIXELS,
+  maxPixels: SEEDANCE_MAX_IMAGE_PIXELS,
+  minSidePixels: SEEDANCE_MIN_SIDE_PIXELS,
+  maxSidePixels: SEEDANCE_MAX_SIDE_PIXELS,
+  maxBytes: SEEDANCE_MAX_IMAGE_BYTES
+};
+var SEEDANCE_25_VIDEO_BOUNDS = {
+  minAspectRatio: SEEDANCE_MIN_ASPECT_RATIO,
+  maxAspectRatio: SEEDANCE_MAX_ASPECT_RATIO,
+  minPixels: SEEDANCE_VIDEO_MIN_PIXELS,
+  maxPixels: SEEDANCE_VIDEO_MAX_PIXELS,
+  minSidePixels: SEEDANCE_MIN_SIDE_PIXELS,
+  maxSidePixels: SEEDANCE_MAX_SIDE_PIXELS,
+  maxBytes: SEEDANCE_MAX_VIDEO_BYTES,
+  minDurationSec: SEEDANCE_MIN_MEDIA_SEC,
+  maxDurationSec: SEEDANCE_25_MAX_MEDIA_SEC,
+  maxFrameRate: SEEDANCE_MAX_FRAME_RATE
+};
+var SEEDANCE_20_VIDEO_BOUNDS = {
+  minAspectRatio: SEEDANCE_MIN_ASPECT_RATIO,
+  maxAspectRatio: SEEDANCE_MAX_ASPECT_RATIO,
+  minPixels: SEEDANCE_VIDEO_MIN_PIXELS,
+  maxPixels: SEEDANCE_VIDEO_MAX_PIXELS,
+  minSidePixels: SEEDANCE_MIN_SIDE_PIXELS,
+  maxSidePixels: SEEDANCE_MAX_SIDE_PIXELS,
+  maxBytes: SEEDANCE_MAX_VIDEO_BYTES,
+  minDurationSec: SEEDANCE_MIN_MEDIA_SEC,
+  maxDurationSec: SEEDANCE_20_MAX_MEDIA_SEC,
+  maxFrameRate: SEEDANCE_MAX_FRAME_RATE
+};
+var SEEDANCE_25_AUDIO_BOUNDS = {
+  minDurationSec: SEEDANCE_MIN_MEDIA_SEC,
+  maxDurationSec: SEEDANCE_25_MAX_MEDIA_SEC,
+  maxBytes: SEEDANCE_MAX_AUDIO_BYTES
+};
+var SEEDANCE_20_AUDIO_BOUNDS = {
+  minDurationSec: SEEDANCE_MIN_MEDIA_SEC,
+  maxDurationSec: SEEDANCE_20_MAX_MEDIA_SEC,
+  maxBytes: SEEDANCE_MAX_AUDIO_BYTES
+};
 var seedance20Constraints = [
   {
     when: {
@@ -3944,15 +4000,11 @@ var { MODELS: MODELS12 } = defineModels("seedance", [
       ...p.enum("outputFormat", SEEDANCE_25_FORMATS, "mp4", { label: "Format" }),
       ...p.enum("colorDepth", SEEDANCE_25_COLOR_DEPTHS, "10bit", { label: "Color Depth" }),
       // 2.5 lifts the reference caps to 30 images / 10 videos / 10 audios.
-      ...params.imageInput(30, "Reference Images", false, "reference", { minSidePixels: SEEDANCE_MIN_SIDE_PIXELS }),
-      ...params.videoInputs(10, "Reference Videos", false, {
-        minPixels: SEEDANCE_VIDEO_MIN_PIXELS,
-        minSidePixels: SEEDANCE_MIN_SIDE_PIXELS,
-        maxBytes: SEEDANCE_25_MAX_VIDEO_BYTES
-      }),
-      ...params.audioInputs(10, "Reference Audios"),
-      ...params.startFrame(),
-      ...params.endFrame()
+      ...params.imageInput(30, "Reference Images", false, "reference", SEEDANCE_IMAGE_BOUNDS),
+      ...params.videoInputs(10, "Reference Videos", false, SEEDANCE_25_VIDEO_BOUNDS),
+      ...params.audioInputs(10, "Reference Audios", false, SEEDANCE_25_AUDIO_BOUNDS),
+      ...params.startFrame("Start Frame", false, SEEDANCE_FRAME_BOUNDS),
+      ...params.endFrame("End Frame", SEEDANCE_FRAME_BOUNDS)
     }
   },
   {
@@ -3983,15 +4035,11 @@ var { MODELS: MODELS12 } = defineModels("seedance", [
       ...p.enum("outputFormat", SEEDANCE_25_FORMATS, "mp4", { label: "Format" }),
       ...p.enum("colorDepth", SEEDANCE_25_COLOR_DEPTHS, "10bit", { label: "Color Depth" }),
       // 2.5 lifts the reference caps to 30 images / 10 videos / 10 audios.
-      ...params.imageInput(30, "Reference Images", false, "reference", { minSidePixels: SEEDANCE_MIN_SIDE_PIXELS }),
-      ...params.videoInputs(10, "Reference Videos", false, {
-        minPixels: SEEDANCE_VIDEO_MIN_PIXELS,
-        minSidePixels: SEEDANCE_MIN_SIDE_PIXELS,
-        maxBytes: SEEDANCE_25_MAX_VIDEO_BYTES
-      }),
-      ...params.audioInputs(10, "Reference Audios"),
-      ...params.startFrame(),
-      ...params.endFrame()
+      ...params.imageInput(30, "Reference Images", false, "reference", SEEDANCE_IMAGE_BOUNDS),
+      ...params.videoInputs(10, "Reference Videos", false, SEEDANCE_25_VIDEO_BOUNDS),
+      ...params.audioInputs(10, "Reference Audios", false, SEEDANCE_25_AUDIO_BOUNDS),
+      ...params.startFrame("Start Frame", false, SEEDANCE_FRAME_BOUNDS),
+      ...params.endFrame("End Frame", SEEDANCE_FRAME_BOUNDS)
     }
   },
   {
@@ -4018,8 +4066,10 @@ var { MODELS: MODELS12 } = defineModels("seedance", [
       ...params.returnLastFrame(),
       ...p.enum("outputFormat", SEEDANCE_25_FORMATS, "mp4", { label: "Format" }),
       ...p.enum("colorDepth", SEEDANCE_25_COLOR_DEPTHS, "10bit", { label: "Color Depth" }),
-      ...params.videoInput("Source Video", "reference", true, void 0, void 0, SEEDANCE_25_MAX_VIDEO_BYTES),
-      ...params.imageInput(30, "Reference Images")
+      // Every limit rides in `bounds`; the positional maxDuration / maxShortSide
+      // / maxBytes args predate it and are skipped rather than duplicated.
+      ...params.videoInput("Source Video", "reference", true, void 0, void 0, void 0, SEEDANCE_25_VIDEO_BOUNDS),
+      ...params.imageInput(30, "Reference Images", false, "reference", SEEDANCE_IMAGE_BOUNDS)
     }
   },
   {
@@ -4047,8 +4097,10 @@ var { MODELS: MODELS12 } = defineModels("seedance", [
       ...params.returnLastFrame(),
       ...p.enum("outputFormat", SEEDANCE_25_FORMATS, "mp4", { label: "Format" }),
       ...p.enum("colorDepth", SEEDANCE_25_COLOR_DEPTHS, "10bit", { label: "Color Depth" }),
-      ...params.videoInput("Source Video", "reference", true, void 0, void 0, SEEDANCE_25_MAX_VIDEO_BYTES),
-      ...params.imageInput(30, "Reference Images")
+      // Every limit rides in `bounds`; the positional maxDuration / maxShortSide
+      // / maxBytes args predate it and are skipped rather than duplicated.
+      ...params.videoInput("Source Video", "reference", true, void 0, void 0, void 0, SEEDANCE_25_VIDEO_BOUNDS),
+      ...params.imageInput(30, "Reference Images", false, "reference", SEEDANCE_IMAGE_BOUNDS)
     }
   },
   {
@@ -4075,7 +4127,7 @@ var { MODELS: MODELS12 } = defineModels("seedance", [
       ...params.generateAudio(),
       ...p.enum("outputFormat", SEEDANCE_25_FORMATS, "mp4", { label: "Format" }),
       ...p.enum("colorDepth", SEEDANCE_25_COLOR_DEPTHS, "10bit", { label: "Color Depth" }),
-      ...params.videoInputs(10, "Source Videos", true, { maxBytes: SEEDANCE_25_MAX_VIDEO_BYTES })
+      ...params.videoInputs(10, "Source Videos", true, SEEDANCE_25_VIDEO_BOUNDS)
     }
   },
   {
@@ -4103,7 +4155,7 @@ var { MODELS: MODELS12 } = defineModels("seedance", [
       ...params.generateAudio(),
       ...p.enum("outputFormat", SEEDANCE_25_FORMATS, "mp4", { label: "Format" }),
       ...p.enum("colorDepth", SEEDANCE_25_COLOR_DEPTHS, "10bit", { label: "Color Depth" }),
-      ...params.videoInputs(10, "Source Videos", true, { maxBytes: SEEDANCE_25_MAX_VIDEO_BYTES })
+      ...params.videoInputs(10, "Source Videos", true, SEEDANCE_25_VIDEO_BOUNDS)
     }
   },
   {
@@ -4129,14 +4181,11 @@ var { MODELS: MODELS12 } = defineModels("seedance", [
       ...params.returnLastFrame(),
       // Reference roles map directly to backend `reference_*` content entries.
       // start/end frame stay on their own named slots.
-      ...params.imageInput(9, "Reference Images", false, "reference", { minSidePixels: SEEDANCE_MIN_SIDE_PIXELS }),
-      ...params.videoInputs(3, "Reference Videos", false, {
-        minPixels: SEEDANCE_VIDEO_MIN_PIXELS,
-        minSidePixels: SEEDANCE_MIN_SIDE_PIXELS
-      }),
-      ...params.audioInputs(3, "Reference Audios"),
-      ...params.startFrame(),
-      ...params.endFrame()
+      ...params.imageInput(9, "Reference Images", false, "reference", SEEDANCE_IMAGE_BOUNDS),
+      ...params.videoInputs(3, "Reference Videos", false, SEEDANCE_20_VIDEO_BOUNDS),
+      ...params.audioInputs(3, "Reference Audios", false, SEEDANCE_20_AUDIO_BOUNDS),
+      ...params.startFrame("Start Frame", false, SEEDANCE_FRAME_BOUNDS),
+      ...params.endFrame("End Frame", SEEDANCE_FRAME_BOUNDS)
     }
   },
   {
@@ -4166,14 +4215,11 @@ var { MODELS: MODELS12 } = defineModels("seedance", [
       ...params.returnLastFrame(),
       // Reference roles map directly to backend `reference_*` content entries.
       // start/end frame stay on their own named slots.
-      ...params.imageInput(9, "Reference Images", false, "reference", { minSidePixels: SEEDANCE_MIN_SIDE_PIXELS }),
-      ...params.videoInputs(3, "Reference Videos", false, {
-        minPixels: SEEDANCE_VIDEO_MIN_PIXELS,
-        minSidePixels: SEEDANCE_MIN_SIDE_PIXELS
-      }),
-      ...params.audioInputs(3, "Reference Audios"),
-      ...params.startFrame(),
-      ...params.endFrame()
+      ...params.imageInput(9, "Reference Images", false, "reference", SEEDANCE_IMAGE_BOUNDS),
+      ...params.videoInputs(3, "Reference Videos", false, SEEDANCE_20_VIDEO_BOUNDS),
+      ...params.audioInputs(3, "Reference Audios", false, SEEDANCE_20_AUDIO_BOUNDS),
+      ...params.startFrame("Start Frame", false, SEEDANCE_FRAME_BOUNDS),
+      ...params.endFrame("End Frame", SEEDANCE_FRAME_BOUNDS)
     }
   },
   {
@@ -4199,14 +4245,11 @@ var { MODELS: MODELS12 } = defineModels("seedance", [
       ...params.returnLastFrame(),
       // Reference roles map directly to backend `reference_*` content entries.
       // start/end frame stay on their own named slots.
-      ...params.imageInput(9, "Reference Images", false, "reference", { minSidePixels: SEEDANCE_MIN_SIDE_PIXELS }),
-      ...params.videoInputs(3, "Reference Videos", false, {
-        minPixels: SEEDANCE_VIDEO_MIN_PIXELS,
-        minSidePixels: SEEDANCE_MIN_SIDE_PIXELS
-      }),
-      ...params.audioInputs(3, "Reference Audios"),
-      ...params.startFrame(),
-      ...params.endFrame()
+      ...params.imageInput(9, "Reference Images", false, "reference", SEEDANCE_IMAGE_BOUNDS),
+      ...params.videoInputs(3, "Reference Videos", false, SEEDANCE_20_VIDEO_BOUNDS),
+      ...params.audioInputs(3, "Reference Audios", false, SEEDANCE_20_AUDIO_BOUNDS),
+      ...params.startFrame("Start Frame", false, SEEDANCE_FRAME_BOUNDS),
+      ...params.endFrame("End Frame", SEEDANCE_FRAME_BOUNDS)
     }
   },
   {
@@ -4232,14 +4275,11 @@ var { MODELS: MODELS12 } = defineModels("seedance", [
       ...params.returnLastFrame(),
       // Reference roles map directly to backend `reference_*` content entries.
       // start/end frame stay on their own named slots.
-      ...params.imageInput(9, "Reference Images", false, "reference", { minSidePixels: SEEDANCE_MIN_SIDE_PIXELS }),
-      ...params.videoInputs(3, "Reference Videos", false, {
-        minPixels: SEEDANCE_VIDEO_MIN_PIXELS,
-        minSidePixels: SEEDANCE_MIN_SIDE_PIXELS
-      }),
-      ...params.audioInputs(3, "Reference Audios"),
-      ...params.startFrame(),
-      ...params.endFrame()
+      ...params.imageInput(9, "Reference Images", false, "reference", SEEDANCE_IMAGE_BOUNDS),
+      ...params.videoInputs(3, "Reference Videos", false, SEEDANCE_20_VIDEO_BOUNDS),
+      ...params.audioInputs(3, "Reference Audios", false, SEEDANCE_20_AUDIO_BOUNDS),
+      ...params.startFrame("Start Frame", false, SEEDANCE_FRAME_BOUNDS),
+      ...params.endFrame("End Frame", SEEDANCE_FRAME_BOUNDS)
     }
   },
   {
@@ -4262,8 +4302,10 @@ var { MODELS: MODELS12 } = defineModels("seedance", [
       ...params.duration(SEEDANCE_V2_DURATIONS, 5),
       ...params.generateAudio(),
       ...params.returnLastFrame(),
-      ...params.videoInput("Source Video"),
-      ...params.imageInput(9, "Reference Images")
+      // Every limit rides in `bounds`; the positional maxDuration / maxShortSide
+      // / maxBytes args predate it and are skipped rather than duplicated.
+      ...params.videoInput("Source Video", "reference", true, void 0, void 0, void 0, SEEDANCE_20_VIDEO_BOUNDS),
+      ...params.imageInput(9, "Reference Images", false, "reference", SEEDANCE_IMAGE_BOUNDS)
     }
   },
   {
@@ -4287,8 +4329,10 @@ var { MODELS: MODELS12 } = defineModels("seedance", [
       ...params.duration(SEEDANCE_V2_DURATIONS, 5),
       ...params.generateAudio(),
       ...params.returnLastFrame(),
-      ...params.videoInput("Source Video"),
-      ...params.imageInput(9, "Reference Images")
+      // Every limit rides in `bounds`; the positional maxDuration / maxShortSide
+      // / maxBytes args predate it and are skipped rather than duplicated.
+      ...params.videoInput("Source Video", "reference", true, void 0, void 0, void 0, SEEDANCE_20_VIDEO_BOUNDS),
+      ...params.imageInput(9, "Reference Images", false, "reference", SEEDANCE_IMAGE_BOUNDS)
     }
   },
   {
@@ -4311,8 +4355,10 @@ var { MODELS: MODELS12 } = defineModels("seedance", [
       ...params.duration(SEEDANCE_V2_DURATIONS, 5),
       ...params.generateAudio(),
       ...params.returnLastFrame(),
-      ...params.videoInput("Source Video"),
-      ...params.imageInput(9, "Reference Images")
+      // Every limit rides in `bounds`; the positional maxDuration / maxShortSide
+      // / maxBytes args predate it and are skipped rather than duplicated.
+      ...params.videoInput("Source Video", "reference", true, void 0, void 0, void 0, SEEDANCE_20_VIDEO_BOUNDS),
+      ...params.imageInput(9, "Reference Images", false, "reference", SEEDANCE_IMAGE_BOUNDS)
     }
   },
   {
@@ -4335,8 +4381,10 @@ var { MODELS: MODELS12 } = defineModels("seedance", [
       ...params.duration(SEEDANCE_V2_DURATIONS, 5),
       ...params.generateAudio(),
       ...params.returnLastFrame(),
-      ...params.videoInput("Source Video"),
-      ...params.imageInput(9, "Reference Images")
+      // Every limit rides in `bounds`; the positional maxDuration / maxShortSide
+      // / maxBytes args predate it and are skipped rather than duplicated.
+      ...params.videoInput("Source Video", "reference", true, void 0, void 0, void 0, SEEDANCE_20_VIDEO_BOUNDS),
+      ...params.imageInput(9, "Reference Images", false, "reference", SEEDANCE_IMAGE_BOUNDS)
     }
   },
   {
@@ -4358,7 +4406,7 @@ var { MODELS: MODELS12 } = defineModels("seedance", [
       ...params.resolution(["480p", "720p", "1080p", "4k"], "720p"),
       ...params.duration(SEEDANCE_V2_DURATIONS, 15),
       ...params.generateAudio(),
-      ...params.videoInputs(3, "Source Videos", true)
+      ...params.videoInputs(3, "Source Videos", true, SEEDANCE_20_VIDEO_BOUNDS)
     }
   },
   {
@@ -4381,7 +4429,7 @@ var { MODELS: MODELS12 } = defineModels("seedance", [
       ...params.resolution(["480p", "720p", "1080p", "4k"], "720p"),
       ...params.duration(SEEDANCE_V2_DURATIONS, 15),
       ...params.generateAudio(),
-      ...params.videoInputs(3, "Source Videos", true)
+      ...params.videoInputs(3, "Source Videos", true, SEEDANCE_20_VIDEO_BOUNDS)
     }
   },
   {
@@ -4403,7 +4451,7 @@ var { MODELS: MODELS12 } = defineModels("seedance", [
       ...params.resolution(["480p", "720p"], "720p"),
       ...params.duration(SEEDANCE_V2_DURATIONS, 15),
       ...params.generateAudio(),
-      ...params.videoInputs(3, "Source Videos", true)
+      ...params.videoInputs(3, "Source Videos", true, SEEDANCE_20_VIDEO_BOUNDS)
     }
   },
   {
@@ -4425,7 +4473,7 @@ var { MODELS: MODELS12 } = defineModels("seedance", [
       ...params.resolution(["480p", "720p"], "720p"),
       ...params.duration(SEEDANCE_V2_DURATIONS, 15),
       ...params.generateAudio(),
-      ...params.videoInputs(3, "Source Videos", true)
+      ...params.videoInputs(3, "Source Videos", true, SEEDANCE_20_VIDEO_BOUNDS)
     }
   },
   {
@@ -6296,23 +6344,12 @@ registerEditPayloads(MODELS22, {
 });
 
 // src/vendors/catalog/elevenlabs.ts
-var buildElevenLabsTTSPayload = (modelId) => (ctx) => ({
-  text: ctx.prompt,
-  voice_id: ctx.voiceId ?? DEFAULT_VOICE_ID,
-  model_id: modelId,
-  ...ctx.language ? { language_code: ctx.language } : {}
-});
+var VOICE_CATALOG = { workflow: "elevenlabs/v1/catalog/voices" };
 var buildElevenLabsSFXPayload = (modelId) => (ctx) => ({
   text: ctx.prompt,
   duration_seconds: ctx.duration ?? 5,
   model_id: modelId
 });
-var buildElevenLabsSTSPayload = (modelId) => (ctx) => ({
-  audio_url: ctx.audioUrl,
-  voice_id: ctx.voiceId ?? DEFAULT_VOICE_ID,
-  model_id: modelId,
-  remove_background_noise: ctx.removeBackgroundNoise ?? false
-});
 var buildElevenLabsAudioIsolationPayload = (ctx) => ({
   audio_url: ctx.audioUrl
 });
@@ -6335,13 +6372,35 @@ var buildElevenLabsVoicePreviewsPayload = (ctx) => ({
   voice_description: ctx.prompt,
   auto_generate_text: true
 });
+var voiceSettingsParams = (withSpeakerBoost) => ({
+  stability: {
+    label: "Stability",
+    descriptor: { kind: "range", min: 0, max: 1, step: 0.05 }
+  },
+  similarityBoost: {
+    label: "Similarity",
+    descriptor: { kind: "range", min: 0, max: 1, step: 0.05 }
+  },
+  // `style` is taken on GenerationContext by the image models' style *name*,
+  // so the vendor's 0–1 style exaggeration gets a key of its own.
+  styleExaggeration: {
+    label: "Style Exaggeration",
+    descriptor: { kind: "range", min: 0, max: 1, step: 0.05 }
+  },
+  speed: {
+    label: "Speed",
+    descriptor: { kind: "range", min: 0.1, max: 5, step: 0.05 }
+  },
+  ...withSpeakerBoost ? { useSpeakerBoost: { label: "Speaker Boost", descriptor: { kind: "boolean", default: true } } } : {}
+});
 var ttsParamConfig = (promptMaxLength, withLanguage) => ({
   // language_code is honoured by eleven_v3 only — the vendor documents it as
   // "not supported for multilingual_v2 models" (silently ignored there).
   // No accent param anywhere: no builder ever read it.
   ...withLanguage ? params.language(false) : {},
   ...params.prompt({ maxLength: promptMaxLength }),
-  ...params.voiceId([], DEFAULT_VOICE_ID, { catalog: { workflow: "elevenlabs/v1/catalog/voices" } })
+  ...params.voiceId([], DEFAULT_VOICE_ID, { catalog: VOICE_CATALOG }),
+  ...voiceSettingsParams(true)
 });
 var { MODELS: MODELS23 } = defineModels("elevenlabs", [
   // ── TTS ───────────────────────────────────────────────────────────
@@ -6351,7 +6410,6 @@ var { MODELS: MODELS23 } = defineModels("elevenlabs", [
     modelId: "eleven_v3",
     addedAt: "2026-02-06",
     workflow: "elevenlabs/v1/text-to-speech",
-    buildPayload: buildElevenLabsTTSPayload("eleven_v3"),
     estimatedTime: 11,
     mode: "audio",
     inputType: "tts",
@@ -6366,7 +6424,6 @@ var { MODELS: MODELS23 } = defineModels("elevenlabs", [
     modelId: "eleven_multilingual_v2",
     addedAt: "2026-02-06",
     workflow: "elevenlabs/v1/text-to-speech",
-    buildPayload: buildElevenLabsTTSPayload("eleven_multilingual_v2"),
     estimatedTime: 9,
     mode: "audio",
     inputType: "tts",
@@ -6375,6 +6432,61 @@ var { MODELS: MODELS23 } = defineModels("elevenlabs", [
     features: [feat("Stable", "characteristic"), feat("Professional", "characteristic")],
     paramConfig: ttsParamConfig(1e4, false)
   },
+  // ── Dialogue ──────────────────────────────────────────────────────
+  {
+    // `modelId` is the pricing-catalog key, not the vendor id: the worker
+    // registers the operation as `eleven-text-to-dialogue` while the vendor
+    // model behind it is plain `eleven_v3`.
+    id: "eleven-text-to-dialogue",
+    name: "Eleven Dialogue v3",
+    modelId: "eleven-text-to-dialogue",
+    addedAt: "2026-09-22",
+    workflow: "elevenlabs/v1/text-to-dialogue",
+    estimatedTime: 12,
+    mode: "audio",
+    inputType: "tts",
+    description: "Generate a multi-speaker conversation \u2014 a voice per line \u2014 in one take.",
+    features: [feat("Multi-Speaker", "characteristic"), feat("Creative Control", "characteristic")],
+    paramConfig: {
+      // One entry per spoken line, in order. The vendor publishes no cap on the
+      // number of lines, so none is invented here.
+      //
+      // The 5,000-character limit is the vendor's hard cap and applies to the
+      // dialogue AS A WHOLE, which a per-field maxLength cannot express — a
+      // single line simply cannot exceed it either. Probed against the live
+      // `elevenlabs/v1/text-to-dialogue` worker: 10,000 characters are refused
+      // with "Request text length (10000) exceeds the maximum text length of
+      // 5000 characters", while 2,600 generate fine. The docs' "keep it at or
+      // below 2,000" is guidance, not a limit. Note the far end is slow: a
+      // 5,000-character dialogue outruns the worker's 120s vendor timeout.
+      dialogue: {
+        label: "Dialogue",
+        required: true,
+        descriptor: {
+          kind: "object",
+          array: { min: 1 },
+          fields: {
+            // A plain id, not a catalog-bound picker: catalog loading resolves
+            // top-level params only. Every ElevenLabs model shares one voices
+            // task, so a caller lists them from any TTS model — e.g.
+            // `ai.catalogs.voices('eleven-v3')` — and uses the ids here.
+            voiceId: { label: "Voice ID", kind: "text", placeholder: DEFAULT_VOICE_ID },
+            text: { label: "Line", kind: "text", maxLength: 5e3 }
+          }
+        }
+      },
+      // The vendor takes any 0–1 double and defaults to 0.5; eleven_v3 reads the
+      // three presets below, so the picker offers those instead of a slider
+      // whose in-between values the engine rounds anyway.
+      ...p.enum("stability", [
+        { id: 0, label: "Creative" },
+        { id: 0.5, label: "Natural" },
+        { id: 1, label: "Robust" }
+      ], 0.5, { label: "Stability" }),
+      ...params.language(false),
+      ...params.seed(4294967295)
+    }
+  },
   // ── Sound Effects ─────────────────────────────────────────────────
   {
     id: "elevenlabs-sfx",
@@ -6421,7 +6533,6 @@ var { MODELS: MODELS23 } = defineModels("elevenlabs", [
     modelId: "eleven_english_sts_v2",
     addedAt: "2026-02-15",
     workflow: "elevenlabs/v1/speech-to-speech",
-    buildPayload: buildElevenLabsSTSPayload("eleven_english_sts_v2"),
     estimatedTime: 15,
     mode: "audio",
     inputType: "sts",
@@ -6429,8 +6540,9 @@ var { MODELS: MODELS23 } = defineModels("elevenlabs", [
     features: [feat("Voice Changer", "characteristic"), feat("Emotion Preserved", "characteristic")],
     paramConfig: {
       ...params.audioInput("Speech Audio", true),
-      ...params.voiceId([], DEFAULT_VOICE_ID, { catalog: { workflow: "elevenlabs/v1/catalog/voices" } }),
-      ...p.boolean("removeBackgroundNoise", false, "Remove Background Noise")
+      ...params.voiceId([], DEFAULT_VOICE_ID, { catalog: VOICE_CATALOG }),
+      ...p.boolean("removeBackgroundNoise", false, "Remove Background Noise"),
+      ...voiceSettingsParams(false)
     }
   },
   {
@@ -6439,7 +6551,6 @@ var { MODELS: MODELS23 } = defineModels("elevenlabs", [
     modelId: "eleven_multilingual_sts_v2",
     addedAt: "2026-02-15",
     workflow: "elevenlabs/v1/speech-to-speech",
-    buildPayload: buildElevenLabsSTSPayload("eleven_multilingual_sts_v2"),
     estimatedTime: 15,
     mode: "audio",
     inputType: "sts",
@@ -6447,8 +6558,9 @@ var { MODELS: MODELS23 } = defineModels("elevenlabs", [
     features: [feat("Voice Changer", "characteristic"), feat("Multilingual", "characteristic"), feat("29 Languages", "characteristic")],
     paramConfig: {
       ...params.audioInput("Speech Audio", true),
-      ...params.voiceId([], DEFAULT_VOICE_ID, { catalog: { workflow: "elevenlabs/v1/catalog/voices" } }),
-      ...p.boolean("removeBackgroundNoise", false, "Remove Background Noise")
+      ...params.voiceId([], DEFAULT_VOICE_ID, { catalog: VOICE_CATALOG }),
+      ...p.boolean("removeBackgroundNoise", false, "Remove Background Noise"),
+      ...voiceSettingsParams(false)
     }
   },
   // ── Audio Processing ────────────────────────────────────────────
@@ -6563,8 +6675,47 @@ var buildElevenLabsMusicPayload = (input) => ({
   model_id: "music_v2",
   force_instrumental: input.isInstrumental ?? false
 });
+var voiceSettings = (input) => {
+  const settings = {
+    ...input.stability != null ? { stability: input.stability } : {},
+    ...input.similarityBoost != null ? { similarity_boost: input.similarityBoost } : {},
+    ...input.styleExaggeration != null ? { style: input.styleExaggeration } : {},
+    ...input.speed != null ? { speed: input.speed } : {},
+    ...input.useSpeakerBoost != null ? { use_speaker_boost: input.useSpeakerBoost } : {}
+  };
+  return Object.keys(settings).length > 0 ? { voice_settings: settings } : {};
+};
+var buildElevenLabsTTSPayload = (modelId) => (input) => ({
+  text: input.prompt,
+  voice_id: input.voiceId ?? DEFAULT_VOICE_ID,
+  model_id: modelId,
+  ...input.language ? { language_code: input.language } : {},
+  ...voiceSettings(input)
+});
+var buildElevenLabsSTSPayload = (modelId) => (input) => ({
+  audio_url: input.audioUrl,
+  voice_id: input.voiceId ?? DEFAULT_VOICE_ID,
+  model_id: modelId,
+  remove_background_noise: input.removeBackgroundNoise ?? false,
+  ...voiceSettings(input)
+});
+var buildElevenLabsDialoguePayload = (input) => ({
+  conversation: input.dialogue.map((line) => ({
+    voice_id: line.voiceId,
+    text: line.text
+  })),
+  model_id: "eleven_v3",
+  ...input.stability != null ? { settings: { stability: input.stability } } : {},
+  ...input.language ? { language_code: input.language } : {},
+  ...input.seed != null ? { seed: input.seed } : {}
+});
 registerPayloads(MODELS23, {
-  "elevenlabs-music-v2": buildElevenLabsMusicPayload
+  "elevenlabs-music-v2": buildElevenLabsMusicPayload,
+  "eleven-v3": buildElevenLabsTTSPayload("eleven_v3"),
+  "eleven-multilingual-v2": buildElevenLabsTTSPayload("eleven_multilingual_v2"),
+  "eleven-sts-v2": buildElevenLabsSTSPayload("eleven_english_sts_v2"),
+  "eleven-multilingual-sts-v2": buildElevenLabsSTSPayload("eleven_multilingual_sts_v2"),
+  "eleven-text-to-dialogue": buildElevenLabsDialoguePayload
 });
 
 // src/vendors/catalog/heygen.ts
@@ -12059,6 +12210,7 @@ var ElevenDubbing = "eleven-dubbing";
 var ElevenMultilingualStsV2 = "eleven-multilingual-sts-v2";
 var ElevenMultilingualV2 = "eleven-multilingual-v2";
 var ElevenStsV2 = "eleven-sts-v2";
+var ElevenTextToDialogue = "eleven-text-to-dialogue";
 var ElevenV3 = "eleven-v3";
 var ElevenVoiceCreate = "eleven-voice-create";
 var ElevenVoiceDesignV2 = "eleven-voice-design-v2";
@@ -12299,6 +12451,7 @@ var Models = {
   ElevenMultilingualStsV2,
   ElevenMultilingualV2,
   ElevenStsV2,
+  ElevenTextToDialogue,
   ElevenV3,
   ElevenVoiceCreate,
   ElevenVoiceDesignV2,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@picsart/ai-sdk",
-  "version": "6.6.0",
+  "version": "6.8.0",
   "type": "module",
   "description": "Type-safe SDK for 100+ AI models — image, video, audio, and text generation with Picsart",
   "license": "MIT",