@picsart/ai-sdk 6.43.0 → 6.45.0

package/index.d.ts

@@ -359,6 +359,15 @@ type ModelInputById = {
         imageUrls?: string[];
         safetyTolerance?: number;
     };
+    "flux-3-image-klein": {
+        prompt: string;
+        aspectRatio?: "auto" | "21:9" | "2:1" | "16:9" | "3:2" | "7:5" | "4:3" | "5:4" | "1:1" | "4:5" | "3:4" | "5:7" | "2:3" | "9:16" | "9:21" | "1:2";
+        resolution?: "768sq" | "1k" | "1.5k" | "2k" | "4k";
+        count?: 1 | 2 | 4 | 6 | 8 | 10;
+        imageUrls?: string[];
+        steps?: number;
+        safetyTolerance?: number;
+    };
     "flux-3-video": {
         prompt: string;
         aspectRatio?: "auto" | "21:9" | "2:1" | "16:9" | "4:3" | "1:1" | "3:4" | "9:16";
@@ -1025,6 +1034,29 @@ type ModelInputById = {
         resolution?: "720p" | "1080p";
         startFrame?: string;
     };
+    "kling-v4": {
+        prompt: string;
+        aspectRatio?: "16:9" | "9:16" | "1:1" | "21:9";
+        duration?: number;
+        resolution?: "720p" | "1080p";
+        startFrame?: string;
+        imageUrls?: string[];
+        endFrame?: string;
+    };
+    "kling-v4-omni": {
+        prompt: string;
+        aspectRatio?: "16:9" | "9:16" | "1:1" | "21:9" | "adaptive";
+        duration?: number;
+        resolution?: "720p" | "1080p";
+        startFrame?: string;
+        endFrame?: string;
+        imageUrls?: string[];
+        videoUrl?: string;
+        referType?: "feature" | "base";
+        elementList?: Array<{
+            element_id: string;
+        }>;
+    };
     "kling-video-effects": {
         templateId?: string;
         imageUrls: [string, ...string[]];
@@ -3563,6 +3595,7 @@ declare const Models: {
     readonly Flux2Max: "flux-2-max";
     readonly Flux2Pro: "flux-2-pro";
     readonly Flux3Image: "flux-3-image";
+    readonly Flux3ImageKlein: "flux-3-image-klein";
     readonly Flux3Video: "flux-3-video";
     readonly FluxKontextMax: "flux-kontext-max";
     readonly FluxKontextPro: "flux-kontext-pro";
@@ -3651,6 +3684,8 @@ declare const Models: {
     readonly KlingV3: "kling-v3";
     readonly KlingV3Omni: "kling-v3-omni";
     readonly KlingV3Turbo: "kling-v3-turbo";
+    readonly KlingV4: "kling-v4";
+    readonly KlingV4Omni: "kling-v4-omni";
     readonly KlingVideoEffects: "kling-video-effects";
     readonly KlingVideoO1: "kling-video-o1";
     readonly Ltx23A2v: "ltx-2.3-a2v";

package/index.js

@@ -1425,10 +1425,25 @@ var klingOmniAdvancedParams = {
     }
   }
 };
+var klingV4OmniElementList = {
+  elementList: {
+    label: "Element References",
+    descriptor: {
+      kind: "object",
+      array: { max: 7 },
+      fields: {
+        element_id: { kind: "text" }
+      }
+    }
+  }
+};
 
 // src/vendors/catalog/kling/index.ts
 var V3_DURATION = { min: 3, max: 15 };
 var V26_DURATIONS = [5, 10];
+var V4_DURATION = { min: 3, max: 30 };
+var V4_PROMPT_MAX = 26e3;
+var V4_MAX_MIDDLE_FRAMES = 8;
 var KLING_IMAGE_AR = ["16:9", "9:16", "1:1", "21:9", "4:3", "3:2", "2:3", "3:4"];
 var klingV3ProVideoBase = {
   workflow: "kling-text-to-video",
@@ -1470,6 +1485,27 @@ var klingV3TurboVideoBase = {
     ...params.startFrame("Start Frame")
   }
 };
+var klingV4VideoBase = {
+  workflow: "kling-text-to-video",
+  editWorkflow: "kling-image-to-video",
+  mode: "video",
+  inputType: "t2v",
+  features: [
+    feat("Image Input", "input"),
+    feat("Start/Middle/End Frames", "frame"),
+    feat("1080p", "resolution"),
+    feat("30 sec", "duration")
+  ],
+  paramConfig: {
+    ...params.prompt({ maxLength: V4_PROMPT_MAX }),
+    ...params.aspectRatio(["16:9", "9:16", "1:1", "21:9"]),
+    ...params.durationRange(V4_DURATION.min, V4_DURATION.max, 5),
+    ...params.resolution(["720p", "1080p"], "720p"),
+    ...params.startFrame("Start Frame"),
+    ...params.imageInput(V4_MAX_MIDDLE_FRAMES, "Middle Frames", false, "asset"),
+    ...params.endFrame("End Frame")
+  }
+};
 var klingV26VideoBase = {
   ...klingV3ProVideoBase,
   features: [
@@ -1523,6 +1559,28 @@ var { MODELS } = defineModels("kling", [
     badge: ["new"],
     description: "Faster V3 variant \u2014 long-form video up to 15s with native audio, start/end frame control, and 720p/1080p output."
   },
+  // ── Video: Kling 4.0 (resolution-tiered T2V + keyframe I2V) ───────
+  {
+    ...klingV4VideoBase,
+    id: "kling-v4",
+    name: "Kling 4.0",
+    modelId: "kling-v4",
+    addedAt: "2026-10-16",
+    // Observed on the vendor (2026-10-16): ~3-5 min for 5s, ~7 min for 30s.
+    estimatedTime: 240,
+    editEstimatedTime: 240,
+    badge: ["new"],
+    description: "Kling 4.0 \u2014 video up to 30s at 720p or 1080p, with start, middle and end keyframes.",
+    constraints: [
+      // Worker: image-to-video needs the start frame (`image`); middle and
+      // end frames are added around it. (Raw API callers may instead pass the
+      // start frame as imageUrls[0] — the builder handles that.)
+      { when: { startFrame: { exists: false } }, then: {
+        imageUrls: { disabled: true, reason: "Middle frames require a start frame." },
+        endFrame: { disabled: true, reason: "End frame requires a start frame." }
+      } }
+    ]
+  },
   // ── Video: Kling V2.6 (consolidated: std/pro via renderingSpeed) ──
   {
     ...klingV26VideoBase,
@@ -1585,6 +1643,37 @@ var { MODELS } = defineModels("kling", [
       } }
     ]
   },
+  // ── Video: Kling 4.0 Omni (dedicated /omni-video/kling-4.0 tasks API) ──
+  {
+    id: "kling-v4-omni",
+    name: "Kling 4.0 Omni",
+    modelId: "kling-v4-omni",
+    addedAt: "2026-10-16",
+    workflow: "kling-omni-video",
+    // Observed on the vendor (2026-10-16): ~2.5-5.5 min for 5s.
+    estimatedTime: 240,
+    mode: "video",
+    inputType: "t2v",
+    badge: ["new"],
+    description: "Kling 4.0 Omni \u2014 up to 30s at 720p or 1080p from text, frames, reference images, a reference or base video, and elements.",
+    features: [feat("Image + Video Input", "input"), feat("1080p", "resolution"), feat("30 sec", "duration")],
+    // Same slot layout as kling-v3-omni (payloads.ts folds them into
+    // image_list / video_list), minus what the 4.0 API dropped: no audio,
+    // mode tiers, multi-shot or keep_original_sound.
+    paramConfig: {
+      ...params.prompt({ maxLength: V4_PROMPT_MAX }),
+      ...params.aspectRatio(["16:9", "9:16", "1:1", "21:9", "adaptive"]),
+      ...params.durationRange(V4_DURATION.min, V4_DURATION.max, 5),
+      ...params.resolution(["720p", "1080p"], "720p"),
+      ...params.startFrame("First Frame"),
+      ...params.endFrame("End Frame"),
+      // Vendor cap: 10 images in total — 2 frame slots + 8 reference images.
+      ...params.imageInput(8, "Reference Images"),
+      ...params.videoInput("Reference Video", "reference", false),
+      ...klingOmniReferType,
+      ...klingV4OmniElementList
+    }
+  },
   {
     id: "kling-video-o1",
     name: "Kling Video O1",
@@ -1990,6 +2079,23 @@ var buildKlingV3TurboPayload = (input) => ({
   resolution: input.resolution ?? "720p",
   ...input.startFrame ? { image: input.startFrame } : {}
 });
+var buildKlingV4Payload = (input) => {
+  const startFrame = input.startFrame || input.imageUrls?.[0];
+  const middleFrames = (input.startFrame ? input.imageUrls : input.imageUrls?.slice(1)) ?? [];
+  if (input.endFrame && !startFrame) {
+    throw new ApiError("Kling 4.0: an end frame requires a start frame.", { status: 400, code: "validation_error" });
+  }
+  return {
+    prompt: input.prompt,
+    aspect_ratio: input.aspectRatio ?? "16:9",
+    duration: String(input.duration ?? 5),
+    model_name: "kling-v4",
+    resolution: input.resolution ?? "720p",
+    ...startFrame ? { image: startFrame } : {},
+    ...middleFrames.length > 0 ? { image_middle: middleFrames } : {},
+    ...input.endFrame ? { image_tail: input.endFrame } : {}
+  };
+};
 var buildKlingV26Payload = (input) => {
   const hasEndFrame = !!(input.startFrame && input.endFrame);
   const hasSound = (input.generateAudio ?? true) && !hasEndFrame;
@@ -2041,6 +2147,24 @@ var buildOmniV3 = (input) => {
     ...hasSound ? { sound: "on" } : {}
   };
 };
+var buildOmniV4 = (input) => {
+  const imageList = [
+    ...input.startFrame ? [{ image_url: input.startFrame, type: "first_frame" }] : [],
+    ...input.endFrame ? [{ image_url: input.endFrame, type: "end_frame" }] : [],
+    ...(input.imageUrls ?? []).map((image_url) => ({ image_url }))
+  ];
+  const videoList = input.videoUrl ? [{ video_url: input.videoUrl, refer_type: input.referType ?? "feature" }] : [];
+  return {
+    prompt: input.prompt,
+    model_name: "kling-v4-omni",
+    aspect_ratio: input.aspectRatio ?? "16:9",
+    duration: String(input.duration ?? 5),
+    resolution: input.resolution ?? "720p",
+    ...imageList.length ? { image_list: imageList } : {},
+    ...videoList.length ? { video_list: videoList } : {},
+    ...stringElementList(input.elementList)
+  };
+};
 var buildVideoO1 = (input) => {
   const hasSound = input.generateAudio ?? true;
   return {
@@ -2160,9 +2284,11 @@ registerPayloads(MODELS, {
   // Combined-entry video — primary slot
   "kling-v3": klingV3Builder,
   "kling-v3-turbo": buildKlingV3TurboPayload,
+  "kling-v4": buildKlingV4Payload,
   "kling-v2-6": buildKlingV26Payload,
   // Omni video
   "kling-v3-omni": buildOmniV3,
+  "kling-v4-omni": buildOmniV4,
   "kling-video-o1": buildVideoO1,
   // Motion control
   "kling-motion-control-v3": buildMotionControl("kling-v3"),
@@ -2187,6 +2313,7 @@ registerPayloads(MODELS, {
 registerEditPayloads(MODELS, {
   "kling-v3": klingV3Builder,
   "kling-v3-turbo": buildKlingV3TurboPayload,
+  "kling-v4": buildKlingV4Payload,
   "kling-v2-6": buildKlingV26Payload
 });
 
@@ -6529,6 +6656,49 @@ var { MODELS: MODELS21 } = defineModels("flux", [
       ...p.range("safetyTolerance", 0, 4, 2, { label: "Safety Tolerance" })
     }
   },
+  {
+    // Distilled "Klein" tier of FLUX 3 Image. Same contract as flux-3-image —
+    // a pure pass-through, since every Command field already carries the SDK's
+    // unified name (prompt, imageUrls, aspectRatio, resolution, count,
+    // safetyTolerance) — with two deltas: a `steps` knob (1–16 denoising steps
+    // for the distilled model; fewer is faster and it does not affect cost) and
+    // a wider safetyTolerance range (0–7 vs the 0–4 of flux-3-image). As with
+    // flux-3-image, supplying imageUrls switches the vendor into editing mode
+    // (hence inputType 't2i'), `version` pins the endpoint contract and is not
+    // surfaced, and `options` is dropped.
+    id: "flux-3-image-klein",
+    name: "Flux 3 Image Klein",
+    workflow: "bfl/v1/flux-3-image-klein",
+    mode: "image",
+    inputType: "t2i",
+    addedAt: "2026-10-16",
+    badge: ["fast"],
+    // Klein is the distilled/low-step tier, so it generates faster than
+    // flux-3-image. Reuse that model's per-tier map as a safe upper bound —
+    // over-estimating only finishes the bar early; tighten once stage
+    // measurements land.
+    estimatedTime: { "768sq": 110, "1k": 150, "1.5k": 195, "2k": 240, "4k": 360 },
+    description: "Distilled, faster FLUX 3 Image \u2014 generate and edit with up to 10 references, with a steps control to trade speed for detail, natively up to 4K.",
+    features: [
+      feat("Multi-Image Input", "input"),
+      feat("4K", "resolution")
+    ],
+    paramConfig: {
+      ...params.prompt(),
+      ...params.aspectRatio(flux3ImageAspectRatios, "auto"),
+      ...params.resolution(flux3ImageResolutions, "1k"),
+      ...params.count(),
+      // Up to 10 references; the first one drives the output ratio while
+      // aspectRatio is 'auto'.
+      ...params.imageInput(10, "Reference Images", false, "reference"),
+      // Denoising steps for the distilled Klein model: 1 (fastest) … 16.
+      // Vendor default is 16; does not affect cost (priced per resolution tier).
+      ...p.range("steps", 1, 16, 16, { label: "Steps" }),
+      // Moderation level: 0 (strict) … 7 (permissive). Wider than the 0–4 of
+      // flux-3-image.
+      ...p.range("safetyTolerance", 0, 7, 2, { label: "Safety Tolerance" })
+    }
+  },
   {
     // Single workflow `flux/v1/video` derives its mode from the input it's
     // given: none → t2v, keyframe images → i2v, a start video → v2v. No
@@ -6571,13 +6741,13 @@ var { MODELS: MODELS21 } = defineModels("flux", [
     // payload builder needed. webhookUrl is polling plumbing, not surfaced
     // (same as flux-3-video).
     id: "flux-video-upscale",
-    name: "Flux Video Upscale",
+    name: "Flux 3 Video Upscale",
     workflow: "flux/v1/video-upscale",
     mode: "video",
     inputType: "v2v",
     addedAt: "2026-08-28",
     estimatedTime: 180,
-    description: "Upscale videos toward 4K (1.5x\u20133x) in precise (source-faithful) or creative (detail-enhancing) mode. Source clips up to 20 seconds and 2K.",
+    description: "FLUX 3 powered video upscale toward 4K (1.5x\u20133x) in precise (source-faithful) or creative (detail-enhancing) mode. Source clips up to 20 seconds and 2K.",
     features: [
       feat("Upscale", "quality"),
       feat("Video Required", "input"),
@@ -15089,6 +15259,7 @@ var Flux2Flex = "flux-2-flex";
 var Flux2Max = "flux-2-max";
 var Flux2Pro = "flux-2-pro";
 var Flux3Image = "flux-3-image";
+var Flux3ImageKlein = "flux-3-image-klein";
 var Flux3Video = "flux-3-video";
 var FluxKontextMax = "flux-kontext-max";
 var FluxKontextPro = "flux-kontext-pro";
@@ -15177,6 +15348,8 @@ var KlingV2a = "kling-v2a";
 var KlingV3 = "kling-v3";
 var KlingV3Omni = "kling-v3-omni";
 var KlingV3Turbo = "kling-v3-turbo";
+var KlingV4 = "kling-v4";
+var KlingV4Omni = "kling-v4-omni";
 var KlingVideoEffects = "kling-video-effects";
 var KlingVideoO1 = "kling-video-o1";
 var Ltx23A2v = "ltx-2.3-a2v";
@@ -15359,6 +15532,7 @@ var Models = {
   Flux2Max,
   Flux2Pro,
   Flux3Image,
+  Flux3ImageKlein,
   Flux3Video,
   FluxKontextMax,
   FluxKontextPro,
@@ -15447,6 +15621,8 @@ var Models = {
   KlingV3,
   KlingV3Omni,
   KlingV3Turbo,
+  KlingV4,
+  KlingV4Omni,
   KlingVideoEffects,
   KlingVideoO1,
   Ltx23A2v,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@picsart/ai-sdk",
-  "version": "6.43.0",
+  "version": "6.45.0",
   "type": "module",
   "description": "Type-safe SDK for 100+ AI models — image, video, audio, and text generation with Picsart",
   "license": "MIT",