@picsart/ai-sdk 5.28.7 → 5.29.0

package/index.js

@@ -2293,6 +2293,7 @@ var FAST_DURATIONS = [6, 8, 10, 12, 14, 16, 18, 20];
 var LTX_RESOLUTIONS = ["1080p", "1440p", "2160p"];
 var LTX_23_AR = ["16:9", "9:16"];
 var LTX_23_FPS = [24, 25, 48, 50];
+var LTX_PROMPT_MAX = 5e3;
 var FAST_LONG = "Videos longer than 10s render at 1080p / 25 fps.";
 var ltxFastLongConstraints = [12, 14, 16, 18, 20].map((d) => ({
   when: { duration: { is: d } },
@@ -2390,7 +2391,7 @@ var { MODELS: MODELS2 } = defineModels("ltx", [
     description: "4K output with audio \u2014 streamlined for fast, production-ready results.",
     features: [feat("Image Input", "input"), feat("Start Frame", "frame"), feat("Audio", "audio"), feat("4K", "resolution"), feat("6/8/10 sec", "duration")],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: LTX_PROMPT_MAX }),
       ...params.duration(PRO_DURATIONS, 6),
       ...params.resolution([...LTX_RESOLUTIONS]),
       ...params.generateAudio(),
@@ -2414,7 +2415,7 @@ var { MODELS: MODELS2 } = defineModels("ltx", [
     description: "Fast with long video support \u2014 up to 20s at 1080p, ideal for drafts and extended scenes.",
     features: [feat("Image Input", "input"), feat("Fast", "duration"), feat("Up to 20s", "duration"), feat("Audio", "audio"), feat("4K", "resolution")],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: LTX_PROMPT_MAX }),
       ...params.duration(FAST_DURATIONS, 6),
       ...params.resolution([...LTX_RESOLUTIONS]),
       ...params.generateAudio(),
@@ -2435,7 +2436,7 @@ var { MODELS: MODELS2 } = defineModels("ltx", [
     description: "Reinterpret existing footage with a new visual direction \u2014 up to 20s segments.",
     features: [feat("Video Input", "input"), feat("Up to 20s", "duration")],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: LTX_PROMPT_MAX }),
       ...params.duration([5, 10, 15, 20], 5),
       ...params.videoInput("Source Video")
     }
@@ -2456,7 +2457,7 @@ var { MODELS: MODELS2 } = defineModels("ltx", [
     description: "4K output with audio and aspect ratio control \u2014 production-ready v2.3.",
     features: [feat("Image Input", "input"), feat("Start Frame", "frame"), feat("Audio", "audio"), feat("4K", "resolution"), feat("16:9 / 9:16", "characteristic"), feat("6/8/10 sec", "duration")],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: LTX_PROMPT_MAX }),
       ...params.duration(PRO_DURATIONS, 6),
       ...params.resolution([...LTX_RESOLUTIONS]),
       ...params.aspectRatio(LTX_23_AR),
@@ -2487,7 +2488,7 @@ var { MODELS: MODELS2 } = defineModels("ltx", [
     description: "Fast 2.3 with long video support \u2014 up to 20s at 1080p with aspect ratio control.",
     features: [feat("Image Input", "input"), feat("Fast", "duration"), feat("Up to 20s", "duration"), feat("Audio", "audio"), feat("4K", "resolution"), feat("16:9 / 9:16", "characteristic")],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: LTX_PROMPT_MAX }),
       ...params.duration(FAST_DURATIONS, 6),
       ...params.resolution([...LTX_RESOLUTIONS]),
       ...params.aspectRatio(LTX_23_AR),
@@ -2517,7 +2518,7 @@ var { MODELS: MODELS2 } = defineModels("ltx", [
     description: "Generate video driven by an audio track \u2014 2-20s, optional image for first frame.",
     features: [feat("Audio Input", "audio"), feat("Image Input", "input"), feat("2\u201320 sec", "duration")],
     paramConfig: {
-      ...params.prompt({ required: false }),
+      ...params.prompt({ required: false, maxLength: LTX_PROMPT_MAX }),
       ...params.audioInput("Audio Track", true),
       ...params.imageInput(1, "First Frame Image", false),
       ...params.aspectRatio(["auto", ...LTX_23_AR]),
@@ -2541,7 +2542,7 @@ var { MODELS: MODELS2 } = defineModels("ltx", [
     description: "Seamlessly extend an existing video forward or backward \u2014 up to 20s.",
     features: [feat("Video Input", "input"), feat("Up to 20s", "duration")],
     paramConfig: {
-      ...params.prompt({ required: false }),
+      ...params.prompt({ required: false, maxLength: LTX_PROMPT_MAX }),
       ...params.durationRange(2, 20, 5, 1),
       ...p.enum("mode", ["end", "start"], "end", { label: "Extend Direction" }),
       ...params.videoInput("Source Video")
@@ -2559,7 +2560,7 @@ var { MODELS: MODELS2 } = defineModels("ltx", [
     description: "Retake video with new direction \u2014 replace audio, video, or both.",
     features: [feat("Video Input", "input"), feat("Up to 20s", "duration")],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: LTX_PROMPT_MAX }),
       ...params.durationRange(2, 20, 5, 1),
       ...p.enum("retakeMode", ["replace_audio_and_video", "replace_audio", "replace_video"], "replace_audio_and_video", { label: "Retake Mode" }),
       // Default-less: when unset the retake starts at 0 (vendor default).
@@ -2741,7 +2742,8 @@ var { MODELS: MODELS6 } = defineModels("bytedance", [
       feat("1080p", "resolution")
     ],
     paramConfig: {
-      ...params.prompt({ required: false }),
+      // BytePlus OmniHuman 1.5 API: the optional prompt is capped at 300 characters.
+      ...params.prompt({ required: false, maxLength: 300 }),
       ...params.imageInput(1, "Portrait Image", true),
       ...params.audioInput("Audio Track", true),
       ...params.resolution([...BYTEDANCE_OMNIHUMAN_RESOLUTION_OPTIONS], "1080p"),
@@ -3224,7 +3226,7 @@ var { MODELS: MODELS10 } = defineModels("wan", [
     description: "Painterly artistic look with audio \u2014 up to 15s at 1080p.",
     features: [feat("Image Input", "input"), feat("Start Frame", "frame"), feat("Audio", "audio"), feat("1080p", "resolution"), feat("5/10/15 sec", "duration")],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: 5e3 }),
       ...params.duration([5, 10, 15], 5),
       ...params.resolution(["480p", "720p", "1080p"], "720p"),
       ...params.aspectRatio(["16:9", "9:16", "1:1", "4:3", "3:4"]),
@@ -3247,7 +3249,7 @@ var { MODELS: MODELS10 } = defineModels("wan", [
     description: "Regenerate video from a reference clip with new stylistic direction.",
     features: [feat("Video Input", "input"), feat("1080p", "resolution"), feat("5/10 sec", "duration")],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: 5e3 }),
       ...params.duration([5, 10], 5),
       ...params.resolution(["720p", "1080p"], "720p"),
       ...params.videoInput("Reference Video")
@@ -3267,7 +3269,8 @@ var { MODELS: MODELS10 } = defineModels("wan", [
     description: "Diverse, stylized images for visual exploration and animation.",
     features: [feat("Multi-Image Input", "input")],
     paramConfig: {
-      ...params.prompt(),
+      // pa-alibaba worker: @MaxLength(1500) on the wan-images prompt (video tasks take 5000).
+      ...params.prompt({ maxLength: 1500 }),
       ...params.count(),
       ...params.negativePrompt()
     }
@@ -3499,8 +3502,9 @@ var buildLumaRay2ReframeVideoPayload = makeReframeVideoPayload("ray-2");
 var buildLumaRayFlash2ReframeVideoPayload = makeReframeVideoPayload("ray-flash-2");
 var LUMA_AR = ["16:9", "9:16", "1:1", "4:3", "3:4", "21:9", "9:21"];
 var LUMA_RESOLUTIONS = ["540p", "720p", "1080p", "4k"];
+var LUMA_PROMPT_MAX = 5e3;
 var lumaParamConfig = {
-  ...params.prompt(),
+  ...params.prompt({ maxLength: LUMA_PROMPT_MAX }),
   ...params.aspectRatio(LUMA_AR),
   ...params.resolution(LUMA_RESOLUTIONS, "720p"),
   ...params.duration([5, 9], 5)
@@ -3534,7 +3538,7 @@ var buildLumaUni1I2IPayload = makeUni1I2IPayload("uni-1");
 var buildLumaUni1MaxT2IPayload = makeUni1T2IPayload("uni-1-max");
 var buildLumaUni1MaxI2IPayload = makeUni1I2IPayload("uni-1-max");
 var lumaUni1ParamConfig = {
-  ...params.prompt({ maxLength: 6e3 }),
+  ...params.prompt({ maxLength: LUMA_PROMPT_MAX }),
   ...params.aspectRatio(LUMA_UNI1_AR, "1:1"),
   ...params.style(LUMA_UNI1_STYLES, "auto"),
   ...params.imageInput(9, "Reference Images", false)
@@ -3645,7 +3649,7 @@ var { MODELS: MODELS11 } = defineModels("luma", [
     description: "Reframe a video to a new aspect ratio using Luma Ray 2.",
     features: [feat("Video Input", "input"), feat("Reframe", "characteristic")],
     paramConfig: {
-      ...params.prompt({ required: false }),
+      ...params.prompt({ required: false, maxLength: LUMA_PROMPT_MAX }),
       ...params.aspectRatio(LUMA_AR, "16:9"),
       ...params.videoInput("Source Video")
     }
@@ -3663,7 +3667,7 @@ var { MODELS: MODELS11 } = defineModels("luma", [
     description: "Reframe a video to a new aspect ratio using Luma Flash 2.",
     features: [feat("Video Input", "input"), feat("Reframe", "characteristic")],
     paramConfig: {
-      ...params.prompt({ required: false }),
+      ...params.prompt({ required: false, maxLength: LUMA_PROMPT_MAX }),
       ...params.aspectRatio(LUMA_AR, "16:9"),
       ...params.videoInput("Source Video")
     }
@@ -3711,7 +3715,7 @@ var { MODELS: MODELS11 } = defineModels("luma", [
     description: "Luma Ray 3.2 \u2014 high-fidelity video generation with start/end frames, HDR, and looping (early access).",
     features: [feat("Image Input", "input"), feat("Start/End Frame", "frame"), feat("HDR", "characteristic"), feat("5/10 sec", "duration")],
     paramConfig: {
-      ...params.prompt({ maxLength: 6e3 }),
+      ...params.prompt({ maxLength: LUMA_PROMPT_MAX }),
       ...params.aspectRatio(RAY32_AR, "16:9"),
       ...params.resolution(RAY32_RESOLUTIONS, "720p"),
       ...params.duration([5, 10], 5),
@@ -3735,7 +3739,7 @@ var { MODELS: MODELS11 } = defineModels("luma", [
     description: "Edit a prior video from a prompt using Luma Ray 3.2 \u2014 preservation-vs-reimagination presets (early access).",
     features: [feat("Video Input", "input"), feat("Edit", "characteristic"), feat("HDR", "characteristic"), feat("5/10 sec", "duration")],
     paramConfig: {
-      ...params.prompt({ maxLength: 6e3 }),
+      ...params.prompt({ maxLength: LUMA_PROMPT_MAX }),
       // Source clip capped at 30s — video_edit rejects longer at ingest (422).
       ...params.videoInput("Source Video", "reference", true, 30),
       ...params.resolution(RAY32_RESOLUTIONS, "720p"),
@@ -3758,7 +3762,7 @@ var { MODELS: MODELS11 } = defineModels("luma", [
     description: "Reframe a video to a new aspect ratio using Luma Ray 3.2 (early access).",
     features: [feat("Video Input", "input"), feat("Reframe", "characteristic")],
     paramConfig: {
-      ...params.prompt({ maxLength: 6e3 }),
+      ...params.prompt({ maxLength: LUMA_PROMPT_MAX }),
       ...params.aspectRatio(RAY32_AR, "16:9"),
       // Source clip capped at 30s — video_reframe rejects longer at ingest (422).
       ...params.videoInput("Source Video", "reference", true, 30),
@@ -4532,6 +4536,7 @@ var buildSora2ExtendPayload = (ctx) => ({
 });
 var SORA_DURATIONS = [4, 8, 12, 16, 20];
 var SORA_AR = ["16:9", "9:16"];
+var SORA_PROMPT_MAX = 5e3;
 var { MODELS: MODELS13 } = defineModels("openai", [
   {
     id: "sora-2-pro",
@@ -4548,7 +4553,7 @@ var { MODELS: MODELS13 } = defineModels("openai", [
     description: "Up to 1080p with strong physical realism and optional reference image.",
     features: [feat("Image Input", "input"), feat("Audio", "audio"), feat("Up to 1080p", "resolution"), feat("4\u201320 sec", "duration")],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: SORA_PROMPT_MAX }),
       ...params.imageInput(1, "Reference Image"),
       ...params.aspectRatio(SORA_AR),
       ...params.resolution(["720p", "1024p", "1080p"]),
@@ -4569,7 +4574,7 @@ var { MODELS: MODELS13 } = defineModels("openai", [
     description: "Naturalistic 720p video with lifelike motion and character detail.",
     features: [feat("Image Input", "input"), feat("Audio", "audio"), feat("720p", "resolution"), feat("4\u201320 sec", "duration")],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: SORA_PROMPT_MAX }),
       ...params.imageInput(1, "Reference Image"),
       ...params.aspectRatio(SORA_AR),
       ...params.duration(SORA_DURATIONS, 4)
@@ -4588,7 +4593,7 @@ var { MODELS: MODELS13 } = defineModels("openai", [
     description: "Seamlessly continue a previously generated Sora video with matching style and pacing.",
     features: [feat("Continue Video", "input"), feat("4\u201320 sec", "duration")],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: SORA_PROMPT_MAX }),
       // video_id is chained from the source Sora asset (declaring the param lets
       // the store seed ctx.videoId); no aspectRatio/size — extend keeps source geometry.
       ...params.videoId([], ""),
@@ -4936,6 +4941,7 @@ var buildGrokTTSPayload = (ctx) => ({
   voice_id: ctx.voiceId ?? DEFAULT_GROK_VOICE_ID
 });
 var GROK_VIDEO_PROMPT_MAX = 4096;
+var GROK_IMAGE_PROMPT_MAX = 8e3;
 var GROK_VIDEO_AR = ["16:9", "9:16", "1:1", "4:3", "3:4", "3:2", "2:3"];
 var GROK_IMAGE_AR = ["1:1", "16:9", "9:16", "4:3", "3:4", "3:2", "2:3", "2:1", "1:2", "19.5:9", "9:19.5", "20:9", "9:20"];
 var GROK_DURATIONS = [3, 5, 6, 8, 10, 12, 15];
@@ -5039,7 +5045,7 @@ var { MODELS: MODELS17 } = defineModels("grok", [
     description: "Rapid image creation with wide aspect-ratio selection and image input.",
     features: [feat("Image Input", "input")],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: GROK_IMAGE_PROMPT_MAX }),
       ...params.aspectRatio(GROK_IMAGE_AR, "1:1"),
       ...params.resolution(GROK_IMAGE_RESOLUTIONS, "1k"),
       ...params.count([1, 2, 4]),
@@ -5061,7 +5067,7 @@ var { MODELS: MODELS17 } = defineModels("grok", [
     description: "Higher-fidelity Grok Imagine variant for production-grade images.",
     features: [feat("Image Input", "input"), feat("2k", "resolution")],
     paramConfig: {
-      ...params.prompt({ maxLength: 8e3 }),
+      ...params.prompt({ maxLength: GROK_IMAGE_PROMPT_MAX }),
       ...params.aspectRatio(GROK_IMAGE_AR, "1:1"),
       ...params.resolution(GROK_IMAGE_RESOLUTIONS, "2k"),
       ...params.count([1, 2, 4]),
@@ -5080,7 +5086,7 @@ var { MODELS: MODELS17 } = defineModels("grok", [
     description: "Latest Grok Imagine generation \u2014 sharper detail with a low/medium quality tier.",
     features: [feat("Image Input", "input"), feat("2k", "resolution")],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: GROK_IMAGE_PROMPT_MAX }),
       ...params.aspectRatio(GROK_IMAGE_AR, "1:1"),
       ...params.resolution(GROK_IMAGE_RESOLUTIONS, "1k"),
       // Vendor-side default is medium; only supported by grok-imagine-image-2.0
@@ -5266,8 +5272,9 @@ var buildVeoPayload = (modelId, opts = {}) => (ctx) => {
     }
   };
 };
+var VEO_PROMPT_MAX = 4e3;
 var veoParamConfig = {
-  ...params.prompt(),
+  ...params.prompt({ maxLength: VEO_PROMPT_MAX }),
   ...params.aspectRatio(["16:9", "9:16"]),
   ...params.duration([4, 6, 8], 8),
   ...params.resolution(["720p", "1080p", "4k"]),
@@ -5290,7 +5297,7 @@ var veoConstraints = [
   } }
 ];
 var veoLiteParamConfig = {
-  ...params.prompt(),
+  ...params.prompt({ maxLength: VEO_PROMPT_MAX }),
   ...params.aspectRatio(["16:9", "9:16"]),
   ...params.duration([4, 6, 8], 8),
   ...params.resolution(["720p", "1080p"]),
@@ -6155,6 +6162,7 @@ var buildGptImage2EditPayload = (ctx) => ({
   ...ctx.background ? { background: ctx.background } : {},
   ...ctx.outputFormat ? { output_format: ctx.outputFormat } : {}
 });
+var GPT_IMAGE_PROMPT_MAX = 32e3;
 var gptImage2Constraints = [
   {
     when: { imageUrls: { exists: true } },
@@ -6188,7 +6196,7 @@ var { MODELS: MODELS23 } = defineModels("openai", [
     description: "Next-gen GPT image model with arbitrary output dimensions and multi-image input.",
     features: [feat("Multi-Image Input", "input"), feat("High Quality", "quality")],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: GPT_IMAGE_PROMPT_MAX }),
       ...params.aspectRatio(["1:1", "3:2", "2:3", "16:9", "9:16", "4:3", "3:4", "auto"], "1:1"),
       ...p.quality(["high", "medium", "low"], "high"),
       // gpt-image-2 supports only an opaque background — the API rejects
@@ -6214,7 +6222,7 @@ var { MODELS: MODELS23 } = defineModels("openai", [
     description: "Strong text-in-image and infographic rendering with multi-image input.",
     features: [feat("Multi-Image Input", "input"), feat("High Quality", "quality")],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: GPT_IMAGE_PROMPT_MAX }),
       ...params.aspectRatio(["1:1", "3:2", "2:3", "16:9", "9:16", "4:3", "3:4"], "1:1"),
       ...p.quality(["high", "medium", "low"], "high"),
       ...p.enum("background", ["opaque", "transparent"], "opaque", { label: "Background" }),
@@ -6241,7 +6249,7 @@ var { MODELS: MODELS23 } = defineModels("openai", [
     description: "Original GPT image model with quality-tiered generation.",
     features: [feat("Multi-Image Input", "input")],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: GPT_IMAGE_PROMPT_MAX }),
       ...params.aspectRatio(["1:1", "3:2", "2:3", "16:9", "9:16", "4:3", "3:4"], "1:1"),
       ...p.quality(["high", "medium", "low"], "high"),
       ...p.enum("background", ["opaque", "transparent"], "opaque", { label: "Background" }),
@@ -6347,7 +6355,8 @@ var { MODELS: MODELS24 } = defineModels("elevenlabs", [
     badge: ["popular"],
     description: "Create custom sound effects from a text description \u2014 up to 30 seconds.",
     features: [feat("Sound Effects", "characteristic")],
-    paramConfig: { ...params.prompt(), ...params.durationRange(0.5, 30, 5, 0.5) }
+    // ElevenLabs sound-generation reference: "The maximum length of the prompt is 450 characters."
+    paramConfig: { ...params.prompt({ maxLength: 450 }), ...params.durationRange(0.5, 30, 5, 0.5) }
   },
   // ── Music ─────────────────────────────────────────────────────────
   {
@@ -6362,7 +6371,8 @@ var { MODELS: MODELS24 } = defineModels("elevenlabs", [
     description: "Generate music with vocals or instrumental from a text prompt.",
     features: [feat("Vocal & Instrumental", "characteristic")],
     paramConfig: {
-      ...params.prompt(),
+      // ElevenLabs music docs state no cap; fal's `elevenlabs/music` schema declares maxLength 4100.
+      ...params.prompt({ maxLength: 4100 }),
       ...params.duration([10, 20, 30, 60, 120, 180, 300, 600], 30),
       ...p.boolean("isInstrumental", false, "Instrumental Only")
     }
@@ -6609,6 +6619,44 @@ var buildMinimaxMusicPayload = (ctx) => ({
     format: ctx.format ?? "mp3"
   }
 });
+var H3_MAX_PROMPT_MAX = 5e4;
+var FRAME_REF_EXCLUSIVE2 = "References and start/end frames cannot be combined.";
+var AUDIO_NEEDS_VISUAL2 = "Audio cannot be the only reference \u2014 add an image or video.";
+var RATIO_FOLLOWS_FRAME = "Aspect ratio follows the frame image.";
+var ADAPTIVE_NEEDS_REFS = "Adaptive aspect ratio requires reference inputs.";
+var H3_MAX_RATIOS = ["21:9", "16:9", "4:3", "1:1", "3:4", "9:16"];
+var REFERENCE_EXCLUDES_FRAMES = {
+  startFrame: { disabled: true, reason: FRAME_REF_EXCLUSIVE2 },
+  endFrame: { disabled: true, reason: FRAME_REF_EXCLUSIVE2 }
+};
+var FRAME_EXCLUDES_REFERENCES = {
+  imageUrls: { disabled: true, reason: FRAME_REF_EXCLUSIVE2 },
+  videoUrls: { disabled: true, reason: FRAME_REF_EXCLUSIVE2 },
+  audioUrls: { disabled: true, reason: FRAME_REF_EXCLUSIVE2 }
+};
+var h3MaxConstraints = [
+  { when: { startFrame: { exists: true } }, then: FRAME_EXCLUDES_REFERENCES },
+  { when: { endFrame: { exists: true } }, then: FRAME_EXCLUDES_REFERENCES },
+  { when: { imageUrls: { exists: true } }, then: REFERENCE_EXCLUDES_FRAMES },
+  { when: { videoUrls: { exists: true } }, then: REFERENCE_EXCLUDES_FRAMES },
+  { when: { audioUrls: { exists: true } }, then: REFERENCE_EXCLUDES_FRAMES },
+  // With a frame the output follows that image, so the ratio is not a choice.
+  // An end frame alone is valid — it generates towards that keyframe.
+  { when: { startFrame: { exists: true } }, then: {
+    aspectRatio: { disabled: true, reason: RATIO_FOLLOWS_FRAME }
+  } },
+  { when: { endFrame: { exists: true } }, then: {
+    aspectRatio: { disabled: true, reason: RATIO_FOLLOWS_FRAME }
+  } },
+  // 'adaptive' means "follow the references" — no meaning without them.
+  { when: { imageUrls: { exists: false }, videoUrls: { exists: false }, audioUrls: { exists: false } }, then: {
+    aspectRatio: { allowed: [...H3_MAX_RATIOS], reason: ADAPTIVE_NEEDS_REFS }
+  } },
+  // Audio carries no visual signal, so it cannot be the only reference.
+  { when: { imageUrls: { exists: false }, videoUrls: { exists: false } }, then: {
+    audioUrls: { disabled: true, reason: AUDIO_NEEDS_VISUAL2 }
+  } }
+];
 var { MODELS: MODELS26 } = defineModels("minimax", [
   {
     id: "minimax-02-hd",
@@ -6679,55 +6727,51 @@ var { MODELS: MODELS26 } = defineModels("minimax", [
     }
   },
   {
-    // Combined T2V/I2V — fal.ai-hosted (pa-fal-ai-pluggable-worker), unlike
-    // the Hailuo/H3 entries in hailuo.ts which route through the minimax
-    // worker. A start frame switches to the image-to-video edit workflow.
+    // One entry for all three H3 Max modes: the inputs pick it — reference
+    // images, videos or audio generate from references; a start and/or end
+    // frame animates those frames; neither generates from the prompt alone.
     id: "minimax-h3-max",
     name: "MiniMax H3 Max",
     modelId: "fal-ai-h3-max",
     addedAt: "2026-08-28",
-    workflow: "minimax/h3-max/text-to-video",
-    editWorkflow: "minimax/h3-max/image-to-video",
+    workflow: "minimax/h3-max/video-generation",
     estimatedTime: 5,
     mode: "video",
     inputType: "t2v",
-    description: "Top-tier MiniMax H3 Max video from text or a start/end frame, with prompt expansion. Up to 15s at 768p.",
+    description: "Top-tier MiniMax H3 Max video from text, a start/end frame, or reference images, videos, and audio \u2014 refer to references in the prompt as Image 1, Video 1, Audio 1, in input order. Up to 15s at 1080p.",
     features: [
-      feat("Image Input", "input"),
       feat("Start/End Frame", "frame"),
-      feat("768p", "resolution"),
+      feat("Multi-Image Input", "input"),
+      feat("Video Input", "input"),
+      feat("Audio Input", "input"),
+      feat("1080p", "resolution"),
       feat("5-15 sec", "duration")
     ],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: H3_MAX_PROMPT_MAX }),
       ...params.startFrame(),
       ...params.endFrame(),
-      // Lowercase on purpose: the worker uppercases for the fal wire ('768P')
-      // and the pricing qualities are lowercase, so this casing serves both.
-      ...params.resolution(["480p", "768p"], "768p"),
+      // Reference slots — reference-to-video route. Clips are 2-15s each
+      // (≤15s combined per modality) and images + videos + audios must add up
+      // to ≤12 files — backend-enforced; only the per-array maxima live here.
+      ...params.imageInput(9, "Reference Images"),
+      ...params.videoInputs(3, "Reference Videos"),
+      ...params.audioInputs(3, "Reference Audios"),
+      // 1080p is a latent refinement of a native 768p generation.
+      ...params.resolution(["480p", "768p", "1080p"], "768p"),
       ...params.durationRange(5, 15, 5),
-      ...params.aspectRatio(["21:9", "16:9", "4:3", "1:1", "3:4", "9:16"], "16:9"),
-      ...p.enum("promptExpansionMode", ["disabled", "balanced", "quality"], "balanced", { label: "Prompt Expansion" }),
+      // 'adaptive' means "follow the references" — constrained below.
+      ...params.aspectRatio(["adaptive", "21:9", "16:9", "4:3", "1:1", "3:4", "9:16"], "16:9"),
+      ...p.enum("promptExpansionMode", ["balanced", "quality"], "balanced", { label: "Prompt Expansion" }),
       // -1 (sentinel) means "pick a random seed"; the builder drops it.
       ...p.range("seed", -1, 2147483647, -1),
       ...p.boolean("enableSafetyChecker", true, "Safety Checker")
     },
-    constraints: [
-      // The T2V wire has no end_image_url — an end frame only reaches the
-      // vendor on the I2V route, which needs a start frame to trigger.
-      { when: { startFrame: { exists: false } }, then: {
-        endFrame: { disabled: true, reason: "An end frame requires a start frame." }
-      } },
-      // I2V derives the ratio from the input image (no aspect_ratio on that wire).
-      { when: { startFrame: { exists: true } }, then: {
-        aspectRatio: { disabled: true, reason: "Aspect ratio follows the start frame image." }
-      } }
-    ]
+    constraints: h3MaxConstraints
   },
   {
-    // Turbo sibling of minimax-h3-max — same fal.ai worker, same wire shape,
-    // tuned for speed (faster-than-realtime generation). The only schema
-    // difference: prompt expansion has no 'disabled' mode on this endpoint.
+    // Turbo sibling of minimax-h3-max — same model family, tuned for speed
+    // (faster-than-realtime generation), text and frames only.
     id: "minimax-h3-max-turbo",
     name: "MiniMax H3 Max Turbo",
     modelId: "fal-ai-h3-max-turbo",
@@ -6737,81 +6781,35 @@ var { MODELS: MODELS26 } = defineModels("minimax", [
     estimatedTime: 5,
     mode: "video",
     inputType: "t2v",
-    description: "Faster-than-realtime MiniMax H3 Max Turbo video from text or a start/end frame, with prompt expansion. Up to 15s at 768p.",
+    description: "Faster-than-realtime MiniMax H3 Max Turbo video from text or a start/end frame, with prompt expansion. Up to 15s at 1080p.",
     features: [
       feat("Fast", "characteristic"),
       feat("Image Input", "input"),
       feat("Start/End Frame", "frame"),
-      feat("768p", "resolution"),
+      feat("1080p", "resolution"),
       feat("5-15 sec", "duration")
     ],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: H3_MAX_PROMPT_MAX }),
       ...params.startFrame(),
       ...params.endFrame(),
-      // Lowercase on purpose: the worker uppercases for the fal wire ('768P')
-      // and the pricing qualities are lowercase, so this casing serves both.
-      ...params.resolution(["480p", "768p"], "768p"),
+      ...params.resolution(["480p", "768p", "1080p"], "768p"),
       ...params.durationRange(5, 15, 5),
       ...params.aspectRatio(["21:9", "16:9", "4:3", "1:1", "3:4", "9:16"], "16:9"),
-      // Unlike minimax-h3-max, the turbo wire has no 'disabled' expansion mode.
       ...p.enum("promptExpansionMode", ["balanced", "quality"], "balanced", { label: "Prompt Expansion" }),
       // -1 (sentinel) means "pick a random seed"; the builder drops it.
       ...p.range("seed", -1, 2147483647, -1),
       ...p.boolean("enableSafetyChecker", true, "Safety Checker")
     },
     constraints: [
-      // The T2V wire has no end_image_url — an end frame only reaches the
-      // vendor on the I2V route, which needs a start frame to trigger.
-      { when: { startFrame: { exists: false } }, then: {
-        endFrame: { disabled: true, reason: "An end frame requires a start frame." }
-      } },
-      // I2V derives the ratio from the input image (no aspect_ratio on that wire).
+      // Either frame routes to image-to-video, which derives the ratio from
+      // the image it was given (no aspect_ratio on that wire). An end frame
+      // alone is valid — fal documents it as end-only keyframe generation.
       { when: { startFrame: { exists: true } }, then: {
-        aspectRatio: { disabled: true, reason: "Aspect ratio follows the start frame image." }
-      } }
-    ]
-  },
-  {
-    // Reference-to-video sibling of minimax-h3-max (same fal.ai worker).
-    // The prompt addresses references by modality and order — Image 1,
-    // Video 1, Audio 1, … Reference clips are 2-15s each (≤15s combined per
-    // modality) and images + videos + audios must add up to ≤12 files —
-    // backend-enforced; paramConfig only carries the per-array maxima.
-    id: "minimax-h3-max-r2v",
-    name: "MiniMax H3 Max Ref-to-Video",
-    modelId: "fal-ai-h3-max",
-    addedAt: "2026-09-01",
-    workflow: "minimax/h3-max/reference-to-video",
-    estimatedTime: 5,
-    mode: "video",
-    inputType: "i2v",
-    description: "MiniMax H3 Max video from reference images, videos, and audio \u2014 refer to them in the prompt as Image 1, Video 1, Audio 1, in input order. Up to 15s at 768p.",
-    features: [
-      feat("Multi-Image Input", "input"),
-      feat("Video Input", "input"),
-      feat("Audio Input", "input"),
-      feat("768p", "resolution"),
-      feat("5-15 sec", "duration")
-    ],
-    paramConfig: {
-      ...params.prompt({ placeholder: "Image 1 is the protagonist. Keep her consistent with the reference while she walks through a sunlit garden..." }),
-      ...params.imageInput(9, "Reference Images"),
-      ...params.videoInputs(3, "Reference Videos"),
-      ...params.audioInputs(3, "Reference Audios"),
-      // Lowercase on purpose — same worker normalization as minimax-h3-max.
-      ...params.resolution(["480p", "768p"], "768p"),
-      ...params.durationRange(5, 15, 5),
-      ...params.aspectRatio(["adaptive", "21:9", "16:9", "4:3", "1:1", "3:4", "9:16"], "adaptive"),
-      // Unlike the T2V/I2V entry, this wire has no 'disabled' expansion mode.
-      ...p.enum("promptExpansionMode", ["balanced", "quality"], "balanced", { label: "Prompt Expansion" }),
-      // -1 (sentinel) means "pick a random seed"; the builder drops it.
-      ...p.range("seed", -1, 2147483647, -1),
-      ...p.boolean("enableSafetyChecker", true, "Safety Checker")
-    },
-    constraints: [
-      { when: { imageUrls: { exists: false }, videoUrls: { exists: false } }, then: {
-        audioUrls: { disabled: true, reason: "Audio cannot be the only reference \u2014 add an image or video." }
+        aspectRatio: { disabled: true, reason: "Aspect ratio follows the frame image." }
+      } },
+      { when: { endFrame: { exists: true } }, then: {
+        aspectRatio: { disabled: true, reason: "Aspect ratio follows the frame image." }
       } }
     ]
   }
@@ -6832,41 +6830,43 @@ var buildMinimaxMusicV3Payload = (input) => ({
     format: input.format ?? "mp3"
   }
 });
-var buildMinimaxH3MaxPayload = (input) => ({
-  prompt: input.prompt,
-  prompt_expansion_mode: input.promptExpansionMode ?? "balanced",
-  duration: input.duration ?? 5,
-  resolution: input.resolution ?? "768p",
-  ...input.startFrame ? {
-    image_url: input.startFrame,
-    ...input.endFrame ? { end_image_url: input.endFrame } : {}
-  } : { aspect_ratio: input.aspectRatio ?? "16:9" },
-  // -1 is the paramConfig sentinel for "random seed" — omit it on the wire.
-  ...input.seed != null && input.seed !== -1 ? { seed: input.seed } : {},
-  enable_safety_checker: input.enableSafetyChecker ?? true
-});
+var buildMinimaxH3MaxPayload = (input) => {
+  const hasReferences = !!(input.imageUrls?.length || input.videoUrls?.length || input.audioUrls?.length);
+  const hasFrame = !!(input.startFrame || input.endFrame);
+  return {
+    prompt: input.prompt,
+    prompt_expansion_mode: input.promptExpansionMode ?? "balanced",
+    duration: input.duration ?? 5,
+    // The vendor enum is uppercase; paramConfig keeps the lowercase form.
+    resolution: (input.resolution ?? "768p").toUpperCase(),
+    // Frames and references pick different modes and are mutually exclusive
+    // (the model's constraints keep them so), hence whichever the caller
+    // filled is what goes out.
+    ...input.startFrame ? { image_url: input.startFrame } : {},
+    ...input.endFrame ? { end_image_url: input.endFrame } : {},
+    ...input.imageUrls?.length ? { reference_image_urls: input.imageUrls } : {},
+    ...input.videoUrls?.length ? { reference_video_urls: input.videoUrls } : {},
+    ...input.audioUrls?.length ? { reference_audio_urls: input.audioUrls } : {},
+    // A frame decides the ratio itself, so the field is dropped there. With
+    // references the default is 'adaptive' (follow them); from the prompt
+    // alone it is 16:9.
+    ...hasFrame ? {} : { aspect_ratio: input.aspectRatio ?? (hasReferences ? "adaptive" : "16:9") },
+    // -1 is the paramConfig sentinel for "random seed" — omit it on the wire.
+    ...input.seed != null && input.seed !== -1 ? { seed: input.seed } : {},
+    enable_safety_checker: input.enableSafetyChecker ?? true
+  };
+};
 var buildMinimaxH3MaxTurboPayload = (input) => ({
   prompt: input.prompt,
   prompt_expansion_mode: input.promptExpansionMode ?? "balanced",
   duration: input.duration ?? 5,
   resolution: input.resolution ?? "768p",
-  ...input.startFrame ? {
-    image_url: input.startFrame,
-    ...input.endFrame ? { end_image_url: input.endFrame } : {}
-  } : { aspect_ratio: input.aspectRatio ?? "16:9" },
-  // -1 is the paramConfig sentinel for "random seed" — omit it on the wire.
-  ...input.seed != null && input.seed !== -1 ? { seed: input.seed } : {},
-  enable_safety_checker: input.enableSafetyChecker ?? true
-});
-var buildMinimaxH3MaxR2VPayload = (input) => ({
-  prompt: input.prompt,
-  prompt_expansion_mode: input.promptExpansionMode ?? "balanced",
-  duration: input.duration ?? 5,
-  resolution: input.resolution ?? "768p",
-  aspect_ratio: input.aspectRatio ?? "adaptive",
-  ...input.imageUrls?.length ? { reference_image_urls: input.imageUrls } : {},
-  ...input.videoUrls?.length ? { reference_video_urls: input.videoUrls } : {},
-  ...input.audioUrls?.length ? { reference_audio_urls: input.audioUrls } : {},
+  // Either frame animates frames, and an end frame alone is valid (it
+  // generates towards that keyframe). The ratio follows the frame when one is
+  // supplied, so it goes out only when neither is set.
+  ...input.startFrame ? { image_url: input.startFrame } : {},
+  ...input.endFrame ? { end_image_url: input.endFrame } : {},
+  ...input.startFrame || input.endFrame ? {} : { aspect_ratio: input.aspectRatio ?? "16:9" },
   // -1 is the paramConfig sentinel for "random seed" — omit it on the wire.
   ...input.seed != null && input.seed !== -1 ? { seed: input.seed } : {},
   enable_safety_checker: input.enableSafetyChecker ?? true
@@ -6874,11 +6874,9 @@ var buildMinimaxH3MaxR2VPayload = (input) => ({
 registerPayloads(MODELS26, {
   "minimax-music-v3": buildMinimaxMusicV3Payload,
   "minimax-h3-max": buildMinimaxH3MaxPayload,
-  "minimax-h3-max-turbo": buildMinimaxH3MaxTurboPayload,
-  "minimax-h3-max-r2v": buildMinimaxH3MaxR2VPayload
+  "minimax-h3-max-turbo": buildMinimaxH3MaxTurboPayload
 });
 registerEditPayloads(MODELS26, {
-  "minimax-h3-max": buildMinimaxH3MaxPayload,
   "minimax-h3-max-turbo": buildMinimaxH3MaxTurboPayload
 });
 
@@ -8231,6 +8229,7 @@ var buildLyria3Payload = (apiModelId) => (ctx) => ({
   ...ctx.imageUrls?.length === 1 ? { image: imagePart(ctx.imageUrls[0]) } : {},
   ...(ctx.imageUrls?.length ?? 0) > 1 ? { images: ctx.imageUrls.slice(0, 10).map(imagePart) } : {}
 });
+var LYRIA_PROMPT_MAX = 5e3;
 var { MODELS: MODELS32 } = defineModels("google", [
   {
     id: "lyria-3-clip",
@@ -8245,7 +8244,7 @@ var { MODELS: MODELS32 } = defineModels("google", [
     description: "Fast music clips from text and image prompts using Google Lyria 3.",
     features: [feat("Image Input", "input"), feat("Vocal & Instrumental", "characteristic")],
     paramConfig: {
-      ...params.prompt(),
+      ...params.prompt({ maxLength: LYRIA_PROMPT_MAX }),
       ...params.imageInput(10, "Mood Images")
     }
   },
@@ -8263,7 +8262,7 @@ var { MODELS: MODELS32 } = defineModels("google", [
     description: "Extended music generation up to 184s with vocals, powered by Google Lyria 3 Pro.",
     features: [feat("Image Input", "input"), feat("Vocal & Instrumental", "characteristic"), feat("Up to 184s", "duration")],
     paramConfig: {
-      ...params.prompt({ placeholder: "Generate voiceover, music and sound effects" }),
+      ...params.prompt({ maxLength: LYRIA_PROMPT_MAX, placeholder: "Generate voiceover, music and sound effects" }),
       ...params.imageInput(10, "Mood Images")
     }
   },
@@ -11444,7 +11443,6 @@ var Lyria35 = "lyria-3.5";
 var Minimax02Hd = "minimax-02-hd";
 var MinimaxH3 = "minimax-h3";
 var MinimaxH3Max = "minimax-h3-max";
-var MinimaxH3MaxR2v = "minimax-h3-max-r2v";
 var MinimaxH3MaxTurbo = "minimax-h3-max-turbo";
 var MinimaxMusicV2 = "minimax-music-v2";
 var MinimaxMusicV3 = "minimax-music-v3";
@@ -11668,7 +11666,6 @@ var Models = {
   Minimax02Hd,
   MinimaxH3,
   MinimaxH3Max,
-  MinimaxH3MaxR2v,
   MinimaxH3MaxTurbo,
   MinimaxMusicV2,
   MinimaxMusicV3,