mulmocast 2.15.0 → 2.17.0

package/lib/types/provider2agent.js

@@ -16,9 +16,10 @@ export const provider2TTSAgent = {
     gemini: {
         agentName: "ttsGeminiAgent",
         hasLimitedConcurrency: false,
-        defaultModel: "gemini-2.5-flash-preview-tts",
+        // The 2.5 preview TTS models shut down on 2026-11-17; Google names gemini-3.8-flash-tts as their successor.
+        defaultModel: "gemini-3.8-flash-tts",
         defaultVoice: "Kore",
-        models: ["gemini-2.5-flash-preview-tts", "gemini-2.5-pro-preview-tts"],
+        models: ["gemini-3.8-flash-tts", "gemini-3.8-flash-lite-tts", "gemini-2.5-flash-preview-tts", "gemini-2.5-pro-preview-tts"],
         keyName: "GEMINI_API_KEY",
     },
     elevenlabs: {
@@ -44,13 +45,18 @@ export const provider2TTSAgent = {
         models: ["mock-model"],
     },
 };
-export const gptImages = ["gpt-image-2", "gpt-image-1.5", "gpt-image-1", "gpt-image-1-mini"];
-const supportedOpenAIImageReplacementHint = "Use 'gpt-image-1' or another supported model.";
+export const gptImages = ["gpt-image-2.5-sunburst", "gpt-image-2.5-flare", "gpt-image-2", "gpt-image-1.5", "gpt-image-1", "gpt-image-1-mini"];
+// The 2.5 models render "auto" (their default) at "low", where gpt-image-1 rendered it at "high".
+export const isGptImage25Model = (model) => model.startsWith("gpt-image-2.5");
+// An unset quality on 2.5 is sent as "high" to keep the look scripts had before the default model changed;
+// an explicit "auto" is the script's choice and is sent as is.
+export const gptImageQuality = (model, quality) => quality ?? (isGptImage25Model(model) ? "high" : undefined);
+const supportedOpenAIImageReplacementHint = "Use 'gpt-image-2.5-sunburst' or another supported model.";
 export const deprecatedOpenAIImageModelHints = {
     "dall-e-2": supportedOpenAIImageReplacementHint,
     "dall-e-3": supportedOpenAIImageReplacementHint,
 };
-const supportedGoogleImageReplacementHint = "Use 'gemini-2.5-flash-image' or 'gemini-3-pro-image-preview' instead.";
+const supportedGoogleImageReplacementHint = "Use 'gemini-3.1-flash-lite-image' or 'gemini-3-pro-image' instead.";
 export const deprecatedGoogleImageModelHints = {
     "imagen-3.0-generate-002": supportedGoogleImageReplacementHint,
     "imagen-4.0-generate-001": supportedGoogleImageReplacementHint,
@@ -59,11 +65,21 @@ export const deprecatedGoogleImageModelHints = {
     "imagen-4.0-generate-preview-06-06": supportedGoogleImageReplacementHint,
     "imagen-4.0-ultra-generate-preview-06-06": supportedGoogleImageReplacementHint,
 };
+// Gemini image models are called through generateContent; anything else is treated as an Imagen model (generateImages).
+export const isGeminiImageModel = (model) => model.startsWith("gemini-");
 // Google image models that on Vertex AI are only published under location "global"
 // (regional endpoints like us-central1 return 404 NOT_FOUND).
 // See https://cloud.google.com/vertex-ai/generative-ai/docs/models/gemini/3-pro-image
 // and https://cloud.google.com/vertex-ai/generative-ai/docs/models/gemini/3-1-flash-image
-export const vertexAIGlobalOnlyImageModels = new Set(["gemini-3-pro-image-preview", "gemini-3.1-flash-image-preview"]);
+export const vertexAIGlobalOnlyImageModels = new Set([
+    "gemini-3.1-flash-lite-image",
+    "gemini-3-pro-image",
+    "gemini-3-pro-image-preview",
+    "gemini-3.1-flash-image-preview",
+]);
+const VERTEX_DEFAULT_IMAGE_LOCATION = "us-central1";
+// A global-only model has no regional endpoint, so an omitted location means "global" for it.
+export const vertexImageLocation = (model, location) => location ?? (vertexAIGlobalOnlyImageModels.has(model) ? "global" : VERTEX_DEFAULT_IMAGE_LOCATION);
 // Per-model reference image limits (image_input array). Only verified entries; unlisted models are not truncated.
 const replicateImageModelParams = {
     "bytedance/seedream-4": { maxReferenceImages: 10 },
@@ -71,7 +87,9 @@ const replicateImageModelParams = {
 export const provider2ImageAgent = {
     openai: {
         agentName: "imageOpenaiAgent",
-        defaultModel: "gpt-image-1",
+        // gpt-image-1 shuts down on 2026-10-23; OpenAI names gpt-image-2.5-sunburst or -flare as its successor.
+        // https://developers.openai.com/api/docs/deprecations
+        defaultModel: "gpt-image-2.5-sunburst",
         models: [...gptImages],
         keyName: "OPENAI_API_KEY",
         baseURLKeyName: "OPENAI_BASE_URL",
@@ -79,8 +97,16 @@ export const provider2ImageAgent = {
     },
     google: {
         agentName: "imageGenAIAgent",
-        defaultModel: "gemini-2.5-flash-image",
-        models: ["gemini-2.5-flash-image", "gemini-3.1-flash-image-preview", "gemini-3-pro-image-preview"],
+        // gemini-2.5-flash-image is deprecated; Google names gemini-3.1-flash-lite-image as its successor.
+        defaultModel: "gemini-3.1-flash-lite-image",
+        models: [
+            "gemini-3.1-flash-lite-image",
+            "gemini-3.1-flash-image",
+            "gemini-3-pro-image",
+            "gemini-2.5-flash-image",
+            "gemini-3.1-flash-image-preview",
+            "gemini-3-pro-image-preview",
+        ],
         keyName: "GEMINI_API_KEY",
     },
     replicate: {
@@ -126,7 +152,6 @@ export const provider2MovieAgent = {
             "bytedance/seedance-1-pro",
             "bytedance/seedance-2.0",
             "bytedance/seedance-2.0-fast",
-            "kwaivgi/kling-v1.6-pro",
             "kwaivgi/kling-v2.1",
             "kwaivgi/kling-v2.1-master",
             "google/veo-2",
@@ -173,20 +198,14 @@ export const provider2MovieAgent = {
                 start_image: "image",
                 last_image: "last_frame_image",
                 audio: { mode: AUDIO_MODE_OPTIONAL, param: "generate_audio" },
-                price_per_sec: 0.29,
+                price_per_sec: 0.18,
             },
             "bytedance/seedance-2.0-fast": {
                 durations: [4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
                 start_image: "image",
                 last_image: "last_frame_image",
                 audio: { mode: AUDIO_MODE_OPTIONAL, param: "generate_audio" },
-                price_per_sec: 0.22,
-            },
-            "kwaivgi/kling-v1.6-pro": {
-                durations: [5, 10],
-                start_image: "start_image",
-                audio: { mode: AUDIO_MODE_NEVER },
-                price_per_sec: 0.095,
+                price_per_sec: 0.15,
             },
             "kwaivgi/kling-v2.1": {
                 durations: [5, 10],
@@ -210,7 +229,7 @@ export const provider2MovieAgent = {
                 durations: [8],
                 start_image: "image",
                 audio: { mode: AUDIO_MODE_OPTIONAL, param: "generate_audio" },
-                price_per_sec: 0.75,
+                price_per_sec: 0.2,
             },
             "google/veo-3.1": {
                 durations: [4, 6, 8],
@@ -218,14 +237,14 @@ export const provider2MovieAgent = {
                 last_image: "last_frame_image",
                 reference_images_param: "reference_images",
                 audio: { mode: AUDIO_MODE_OPTIONAL, param: "generate_audio" },
-                price_per_sec: 0.75,
+                price_per_sec: 0.2,
             },
             "google/veo-3.1-fast": {
                 durations: [4, 6, 8],
                 start_image: "image",
                 last_image: "last_frame_image",
                 audio: { mode: AUDIO_MODE_OPTIONAL, param: "generate_audio" },
-                price_per_sec: 0.4,
+                price_per_sec: 0.1,
             },
             "google/veo-3.1-lite": {
                 durations: [4, 6, 8],
@@ -238,13 +257,13 @@ export const provider2MovieAgent = {
                 durations: [8],
                 start_image: "image",
                 audio: { mode: AUDIO_MODE_OPTIONAL, param: "generate_audio" },
-                price_per_sec: 0.4,
+                price_per_sec: 0.1,
             },
             "minimax/video-01": {
                 durations: [6],
                 start_image: "first_frame_image",
                 audio: { mode: AUDIO_MODE_NEVER },
-                price_per_sec: 0.5,
+                price_per_sec: 0.083,
             },
             "minimax/hailuo-02": {
                 durations: [6], // NOTE: 10 for only 720p
@@ -264,7 +283,7 @@ export const provider2MovieAgent = {
                 start_image: "image",
                 last_image: "last_frame_image",
                 audio: { mode: AUDIO_MODE_OPTIONAL, param: "sound_effect_switch" },
-                price_per_sec: 0.12,
+                price_per_sec: 0.06,
             },
             "wan-video/wan-2.2-i2v-fast": {
                 // No duration input: length is num_frames (default 81) / frames_per_second (default 16) ≈ 5s.
@@ -279,26 +298,26 @@ export const provider2MovieAgent = {
                 durations: [5],
                 start_image: undefined,
                 audio: { mode: AUDIO_MODE_NEVER },
-                price_per_sec: 0.012,
+                price_per_sec: 0.01,
             },
             "xai/grok-imagine-video": {
                 durations: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
                 start_image: "image",
                 audio: { mode: AUDIO_MODE_NEVER },
-                price_per_sec: 0.08,
+                price_per_sec: 0.05,
             },
             "xai/grok-imagine-r2v": {
                 durations: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
                 start_image: undefined,
                 reference_images_param: "reference_images",
                 audio: { mode: AUDIO_MODE_NEVER },
-                price_per_sec: 0.08,
+                price_per_sec: 0.05,
             },
             "runwayml/gen-4.5": {
                 durations: [5, 10],
                 start_image: "image",
                 audio: { mode: AUDIO_MODE_NEVER },
-                price_per_sec: 0.25,
+                price_per_sec: 0.12,
             },
             "kwaivgi/kling-v3-omni-video": {
                 durations: [3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
@@ -306,7 +325,7 @@ export const provider2MovieAgent = {
                 last_image: "end_image",
                 reference_images_param: "reference_images",
                 audio: { mode: AUDIO_MODE_OPTIONAL, param: "generate_audio" },
-                price_per_sec: 0.28, // 'pro' (1080p, default); 'standard' $0.168, '4k' $0.42
+                price_per_sec: 0.224,
             },
             "kwaivgi/kling-v3-video": {
                 durations: [3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
@@ -314,7 +333,7 @@ export const provider2MovieAgent = {
                 last_image: "end_image",
                 reference_images_param: "reference_images",
                 audio: { mode: AUDIO_MODE_OPTIONAL, param: "generate_audio" },
-                price_per_sec: 0.3,
+                price_per_sec: 0.224,
             },
             // TODO: price_per_sec for the models below is a coarse approximation.
             // Actual Replicate pricing varies by resolution / duration / quality and
@@ -324,27 +343,27 @@ export const provider2MovieAgent = {
                 durations: [3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
                 start_image: "image",
                 audio: { mode: AUDIO_MODE_NEVER },
-                price_per_sec: 0.05,
+                price_per_sec: 0.14,
             },
             "minimax/hailuo-2.3": {
                 durations: [6, 10],
                 start_image: "first_frame_image",
                 audio: { mode: AUDIO_MODE_NEVER },
-                price_per_sec: 0.1,
+                price_per_sec: 0.047,
             },
             "minimax/hailuo-2.3-fast": {
                 durations: [6, 10],
                 start_image: "first_frame_image",
                 start_image_required: true,
                 audio: { mode: AUDIO_MODE_NEVER },
-                price_per_sec: 0.06,
+                price_per_sec: 0.032,
             },
             "pixverse/pixverse-v5": {
                 durations: [5, 8],
                 start_image: "image",
                 last_image: "last_frame_image",
                 audio: { mode: AUDIO_MODE_NEVER },
-                price_per_sec: 0.12,
+                price_per_sec: 0.06,
             },
             "prunaai/p-video": {
                 durations: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20],
@@ -364,8 +383,9 @@ export const provider2MovieAgent = {
     },
     google: {
         agentName: "movieGenAIAgent",
-        defaultModel: "veo-2.0-generate-001",
-        models: ["veo-2.0-generate-001", "veo-3.0-generate-001", "veo-3.1-generate-preview", "veo-3.1-lite-generate-preview"],
+        // veo-2.0 and veo-3.0 were shut down on 2026-06-30; the Gemini API now lists only the Veo 3.1 previews.
+        defaultModel: "veo-3.1-generate-preview",
+        models: ["veo-3.1-generate-preview", "veo-3.1-lite-generate-preview", "veo-3.1-generate-001"],
         keyName: "GEMINI_API_KEY",
         modelParams: {
             "veo-3.1-lite-generate-preview": {
@@ -376,7 +396,8 @@ export const provider2MovieAgent = {
                 supportsPersonGeneration: false,
                 audio: { mode: AUDIO_MODE_ALWAYS },
             },
-            "veo-3.1-generate-preview": {
+            // Vertex AI only: the GA name of Veo 3.1 there, where the preview IDs are retired.
+            "veo-3.1-generate-001": {
                 durations: [4, 6, 8],
                 supportsDuration: true,
                 supportsLastFrame: true,
@@ -384,22 +405,14 @@ export const provider2MovieAgent = {
                 supportsPersonGeneration: false,
                 audio: { mode: AUDIO_MODE_ALWAYS },
             },
-            "veo-3.0-generate-001": {
-                durations: [8],
-                supportsDuration: false, // Veo 3.0 always generates 8s
-                supportsLastFrame: false,
-                supportsReferenceImages: false,
+            "veo-3.1-generate-preview": {
+                durations: [4, 6, 8],
+                supportsDuration: true,
+                supportsLastFrame: true,
+                supportsReferenceImages: true,
                 supportsPersonGeneration: false,
                 audio: { mode: AUDIO_MODE_ALWAYS },
             },
-            "veo-2.0-generate-001": {
-                durations: [5, 6, 8],
-                supportsDuration: true,
-                supportsLastFrame: false, // Vertex AI only
-                supportsReferenceImages: false,
-                supportsPersonGeneration: true,
-                audio: { mode: AUDIO_MODE_NEVER },
-            },
         },
     },
     mock: {
@@ -471,12 +484,16 @@ export const provider2LipSyncAgent = {
 export const provider2LLMAgent = {
     openai: {
         agentName: "openAIAgent",
-        defaultModel: "gpt-5",
+        // gpt-5 (gpt-5-2025-08-07) shuts down on 2026-12-11; OpenAI names gpt-5.6-sol as its successor.
+        defaultModel: "gpt-5.6-sol",
         keyName: "OPENAI_API_KEY",
         baseURLKeyName: "OPENAI_BASE_URL",
         apiVersionKeyName: "OPENAI_API_VERSION",
         max_tokens: 8192,
         models: [
+            "gpt-5.6-sol",
+            "gpt-5.6-terra",
+            "gpt-5.6-luna",
             "gpt-5",
             "gpt-5-nano",
             "gpt-5-mini",
@@ -494,26 +511,31 @@ export const provider2LLMAgent = {
     },
     anthropic: {
         agentName: "anthropicAgent",
-        defaultModel: "claude-sonnet-4-5-20250929",
+        // claude-sonnet-4-5 retires on 2026-11-30. Its named successor, claude-sonnet-5-5 (like claude-opus-4-8),
+        // rejects the temperature @graphai/anthropic_agent always sends, so the newest model that answers is used.
+        defaultModel: "claude-sonnet-4-6",
         max_tokens: 8192,
-        models: ["claude-opus-4-1-20250805", "claude-opus-4-20250514", "claude-sonnet-4-20250514", "claude-sonnet-4-5-20250929", "claude-haiku-4-5-20251001"],
+        models: ["claude-sonnet-4-6", "claude-sonnet-4-5-20250929", "claude-haiku-4-5-20251001"],
         keyName: "ANTHROPIC_API_KEY",
         apiKeyNameOverride: "ANTHROPIC_API_TOKEN",
         // GraphAI is currently using ANTHROPIC_API_KEY, but the official name is ANTHROPIC_API_TOKEN.
     },
     gemini: {
         agentName: "geminiAgent",
-        defaultModel: "gemini-2.5-flash",
+        // Google limits the 2.5 models to users who have used them before and recommends 3.8 Flash.
+        defaultModel: "gemini-3.8-flash",
         max_tokens: 8192,
-        models: ["gemini-3.1-pro-preview", "gemini-3-flash-preview", "gemini-2.5-pro", "gemini-2.5-flash", "gemini-2.5-flash-lite"],
+        models: ["gemini-3.8-flash", "gemini-3.1-pro-preview", "gemini-3-flash-preview", "gemini-2.5-pro", "gemini-2.5-flash", "gemini-2.5-flash-lite"],
         keyName: "GEMINI_API_KEY",
     },
     groq: {
         agentName: "groqAgent",
-        defaultModel: "llama-3.1-8b-instant",
+        // The Llama and DeepSeek models were shut down; Groq names openai/gpt-oss-20b as the 8B model's successor.
+        // https://console.groq.com/docs/deprecations
+        defaultModel: "openai/gpt-oss-20b",
         keyName: "GROQ_API_KEY",
         max_tokens: 4096,
-        models: ["llama-3.1-8b-instant", "llama-3.3-70b-versatile", "deepseek-r1-distill-llama-70b", "openai/gpt-oss-120b", "openai/gpt-oss-20b"],
+        models: ["openai/gpt-oss-20b", "openai/gpt-oss-120b"],
     },
     mock: {
         agentName: "mediaMockAgent",
@@ -562,6 +584,14 @@ export const getModelAudio = (provider, model) => {
     const modelParams = provider2MovieAgent[provider]?.modelParams;
     return modelParams?.[model]?.audio;
 };
+const GOOGLE_VERTEX_DEFAULT_MOVIE_MODEL = "veo-3.1-generate-001";
+// The default model is shared by the Gemini API and Vertex AI, but Vertex retired the Veo 3.1 preview IDs.
+export const defaultMovieModel = (provider, movieParams) => provider === "google" && movieParams?.vertexai_project ? GOOGLE_VERTEX_DEFAULT_MOVIE_MODEL : provider2MovieAgent[provider].defaultModel;
+// Veo 3.1 (preview) builds clips longer than one segment by extending an 8-second first segment 8 seconds at a time.
+export const VEO_EXTENSION_MODEL = "veo-3.1-generate-preview";
+export const VEO_SEGMENT_SEC = 8;
+export const veoExtensionCount = (requestedSec) => Math.ceil((requestedSec - VEO_SEGMENT_SEC) / VEO_SEGMENT_SEC);
+export const veoExtendedSeconds = (requestedSec) => VEO_SEGMENT_SEC * (1 + veoExtensionCount(requestedSec));
 export const getModelDuration = (provider, model, movieDuration) => {
     const modelParams = provider2MovieAgent[provider]?.modelParams;
     const { durations } = modelParams[model];
@@ -582,7 +612,8 @@ export const gptImageOutputTokens = {
 // Spot-checked against replicate.com model pages on this date (seedance-1-lite, kling-v2.1, omni-human).
 // Some replicate prices vary by resolution/variant; price_per_sec holds the rate for the typical configuration
 // (e.g. seedance-1-lite at 720p, kling-v2.1 standard).
-const REPLICATE_PRICING_AS_OF = "2026-07-03";
+// Optional-audio models are priced without audio, since mulmocast generates them silent by default.
+const REPLICATE_PRICING_AS_OF = "2026-10-07";
 const replicateMoviePricing = Object.fromEntries(Object.entries(provider2MovieAgent.replicate.modelParams).map(([model, params]) => [
     model,
     { unit: "seconds", perSecUSD: params.price_per_sec, asOf: REPLICATE_PRICING_AS_OF },
@@ -601,22 +632,35 @@ export const modelPricing = {
         "gpt-image-1.5": { unit: "tokens", inputPerMTokensUSD: 5, outputPerMTokensUSD: 32, asOf: "2026-07-03" },
         "gpt-image-2": { unit: "tokens", inputPerMTokensUSD: 5, outputPerMTokensUSD: 30, asOf: "2026-07-03" },
         "gpt-image-1-mini": { unit: "tokens", inputPerMTokensUSD: 2, outputPerMTokensUSD: 8, asOf: "2026-07-03" },
+        // https://developers.openai.com/api/docs/pricing (text input $5/1M, image output $30/1M; image input $8/1M is not modeled)
+        "gpt-image-2.5-sunburst": { unit: "tokens", inputPerMTokensUSD: 5, outputPerMTokensUSD: 30, asOf: "2026-10-07" },
+        "gpt-image-2.5-flare": { unit: "tokens", inputPerMTokensUSD: 5, outputPerMTokensUSD: 30, asOf: "2026-10-07" },
         // https://developers.openai.com/api/docs/models/gpt-5 etc.
         "gpt-5": { unit: "tokens", inputPerMTokensUSD: 1.25, outputPerMTokensUSD: 10, asOf: "2026-07-03" },
         "gpt-5-mini": { unit: "tokens", inputPerMTokensUSD: 0.25, outputPerMTokensUSD: 2, asOf: "2026-07-03" },
+        // https://developers.openai.com/api/docs/models/gpt-5.6-sol (promotional price, "at least through November 21, 2026")
+        "gpt-5.6-sol": { unit: "tokens", inputPerMTokensUSD: 4, outputPerMTokensUSD: 20, asOf: "2026-10-07" },
+        "gpt-5.6-terra": { unit: "tokens", inputPerMTokensUSD: 2, outputPerMTokensUSD: 12, asOf: "2026-10-07" },
         "gpt-4o": { unit: "tokens", inputPerMTokensUSD: 2.5, outputPerMTokensUSD: 10, asOf: "2026-07-03" },
     },
     gemini: {
         // https://ai.google.dev/gemini-api/docs/pricing
         "gemini-2.5-flash-preview-tts": { unit: "tokens", inputPerMTokensUSD: 0.5, outputPerMTokensUSD: 10, asOf: "2026-07-03" },
         "gemini-2.5-pro-preview-tts": { unit: "tokens", inputPerMTokensUSD: 1, outputPerMTokensUSD: 20, asOf: "2026-07-03" },
+        // through 2026-12-31; $1 / $18 (flash) and $1 / $12 (lite) from 2027-01-01
+        "gemini-3.8-flash-tts": { unit: "tokens", inputPerMTokensUSD: 0.5, outputPerMTokensUSD: 9, asOf: "2026-10-07" },
+        "gemini-3.8-flash-lite-tts": { unit: "tokens", inputPerMTokensUSD: 0.5, outputPerMTokensUSD: 6, asOf: "2026-10-07" },
     },
     google: {
         // https://ai.google.dev/gemini-api/docs/pricing ($0.039/image ≒ 1290 output tokens at $30/1M)
         "gemini-2.5-flash-image": { unit: "images", inputPerMTokensUSD: 0.3, perImageUSD: 0.039, asOf: "2026-07-03" },
+        // $0.0336 per 1K-resolution image
+        "gemini-3.1-flash-lite-image": { unit: "images", inputPerMTokensUSD: 0.25, perImageUSD: 0.0336, asOf: "2026-10-07" },
         // Veo per second of generated video (720p). veo-2.0-generate-001 and veo-3.0-generate-001
         // were shut down on 2026-06-30, so they intentionally have no price.
         "veo-3.1-generate-preview": { unit: "seconds", perSecUSD: 0.4, asOf: "2026-07-03" },
+        // Vertex AI: https://cloud.google.com/vertex-ai/generative-ai/pricing (video + audio, 720p/1080p)
+        "veo-3.1-generate-001": { unit: "seconds", perSecUSD: 0.4, asOf: "2026-10-07" },
         "veo-3.1-lite-generate-preview": { unit: "seconds", perSecUSD: 0.05, asOf: "2026-07-03" },
         // Google Cloud Text-to-Speech per 1M characters, keyed by voice tier.
         // https://cloud.google.com/text-to-speech/pricing
@@ -629,15 +673,16 @@ export const modelPricing = {
     anthropic: {
         // https://platform.claude.com/docs/en/docs/about-claude/pricing
         "claude-sonnet-4-5-20250929": { unit: "tokens", inputPerMTokensUSD: 3, outputPerMTokensUSD: 15, asOf: "2026-07-03" },
+        "claude-sonnet-4-6": { unit: "tokens", inputPerMTokensUSD: 3, outputPerMTokensUSD: 15, asOf: "2026-10-07" },
     },
     elevenlabs: {
-        // https://elevenlabs.io/pricing/api ($0.10 per 1k chars for multilingual/v3, $0.05 for flash/turbo)
-        eleven_v3: { unit: "chars", perMCharsUSD: 100, asOf: "2026-07-03" },
-        eleven_multilingual_v2: { unit: "chars", perMCharsUSD: 100, asOf: "2026-07-03" },
-        eleven_turbo_v2_5: { unit: "chars", perMCharsUSD: 50, asOf: "2026-07-03" },
-        eleven_turbo_v2: { unit: "chars", perMCharsUSD: 50, asOf: "2026-07-03" },
-        eleven_flash_v2_5: { unit: "chars", perMCharsUSD: 50, asOf: "2026-07-03" },
-        eleven_flash_v2: { unit: "chars", perMCharsUSD: 50, asOf: "2026-07-03" },
+        // https://elevenlabs.io/pricing/api ($0.08 per 1k chars for multilingual/v3, $0.04 for flash/turbo)
+        eleven_v3: { unit: "chars", perMCharsUSD: 80, asOf: "2026-10-07" },
+        eleven_multilingual_v2: { unit: "chars", perMCharsUSD: 80, asOf: "2026-10-07" },
+        eleven_turbo_v2_5: { unit: "chars", perMCharsUSD: 40, asOf: "2026-10-07" },
+        eleven_turbo_v2: { unit: "chars", perMCharsUSD: 40, asOf: "2026-10-07" },
+        eleven_flash_v2_5: { unit: "chars", perMCharsUSD: 40, asOf: "2026-10-07" },
+        eleven_flash_v2: { unit: "chars", perMCharsUSD: 40, asOf: "2026-10-07" },
     },
     replicate: {
         ...replicateMoviePricing,

package/lib/types/schema.d.ts

@@ -4466,8 +4466,8 @@ export declare const mulmoOpenAIImageModelSchema: z.ZodObject<{
         [x: string]: string;
     }>>;
     quality: z.ZodOptional<z.ZodEnum<{
-        low: "low";
         high: "high";
+        low: "low";
         auto: "auto";
         medium: "medium";
     }>>;

package/lib/utils/estimate_usage.js

@@ -4,10 +4,11 @@
 // assume every asset is generated fresh.
 import { Tiktoken } from "js-tiktoken/lite";
 import o200k_base from "js-tiktoken/ranks/o200k_base";
-import { provider2TTSAgent, provider2MovieAgent, gptImageOutputTokens, getModelDuration, getModelPricing, defaultProviders, } from "../types/provider2agent.js";
+import { provider2TTSAgent, provider2MovieAgent, gptImageOutputTokens, isGptImage25Model, getModelDuration, defaultMovieModel, veoExtendedSeconds, VEO_EXTENSION_MODEL, VEO_SEGMENT_SEC, getModelPricing, defaultProviders, } from "../types/provider2agent.js";
 import { text2SpeechProviderSchema, text2MovieProviderSchema } from "../types/schema.js";
 import { MulmoPresentationStyleMethods } from "../methods/mulmo_presentation_style.js";
 import { MulmoBeatMethods } from "../methods/mulmo_beat.js";
+import { geminiTtsInputText } from "./gemini_tts.js";
 import { imagePrompt, htmlImageSystemPrompt, translateSystemPrompt, translatePrompts } from "./prompt.js";
 const MILLION = 1_000_000;
 const LATIN_CHARS_PER_TOKEN = 4;
@@ -92,15 +93,14 @@ const buildOpenAITts = ({ speaker, text, textIsFinal, beatIndex, lang }) => {
     return { process: "tts", beatIndex, lang, provider: "openai", model, inputChars: metric(text.length, textIsFinal), ...tokenMetrics };
 };
 const buildGeminiTts = ({ speaker, speechOptions, text, beatIndex, lang }) => {
-    // The gemini TTS agent wraps the transcript in a "Director's Notes" prompt when an instruction is set.
-    const prompt = speechOptions?.instruction ? `### DIRECTOR'S NOTES\n${speechOptions.instruction}\n\n#### TRANSCRIPT\n${text}` : text;
+    const model = speaker.model ?? provider2TTSAgent.gemini.defaultModel;
     return {
         process: "tts",
         beatIndex,
         lang,
         provider: "gemini",
-        model: speaker.model ?? provider2TTSAgent.gemini.defaultModel,
-        inputTokens: estimated(countHeuristicTokens(prompt)),
+        model,
+        inputTokens: estimated(countHeuristicTokens(geminiTtsInputText(model, text, speechOptions?.instruction))),
         outputTokens: estimated(estimateSpeechSec(text) * GEMINI_TTS_AUDIO_TOKENS_PER_SEC),
     };
 };
@@ -162,8 +162,8 @@ const gptImageSize = (canvasSize) => {
 const gptImageOutputMetric = (model, canvasSize, quality) => {
     const table = gptImageOutputTokens[gptImageSize(canvasSize)];
     const knownQuality = quality === "low" || quality === "medium" || quality === "high" ? quality : undefined;
-    // Unspecified quality means API "auto"; assume "high" as the conservative upper bound.
-    const tokens = table[knownQuality ?? "high"];
+    // Anything else renders at "high" (the upper bound), except an explicit "auto" on 2.5, which renders at "low".
+    const tokens = table[knownQuality ?? (quality === "auto" && isGptImage25Model(model) ? "low" : "high")];
     return metric(tokens, GPT_IMAGE_FIXED_TOKEN_MODELS.includes(model) && knownQuality !== undefined);
 };
 const buildImageRecord = ({ process, provider, model, prompt, canvasSize, quality, beatIndex, refKey }) => {
@@ -217,6 +217,8 @@ const beatDurationMetric = (beat) => {
     }
     return estimated(beat.text ? estimateSpeechSec(beat.text) : DEFAULT_MOVIE_DURATION_SEC);
 };
+// The movie agent extends these clips past one segment instead of capping them, and Google bills every generated second.
+const isVeoExtension = (provider, model, duration) => provider === "google" && model === VEO_EXTENSION_MODEL && duration.value > VEO_SEGMENT_SEC;
 const snapMovieDuration = (provider, model, duration) => {
     if (!isKeyOf(provider2MovieAgent[provider].modelParams, model)) {
         return duration;
@@ -230,9 +232,11 @@ const estimateMovieGeneration = (style, beat, beatIndex, refKey) => {
     if (!isKeyOf(provider2MovieAgent, provider) || provider === "mock") {
         return undefined;
     }
-    const model = info.movieParams?.model ?? provider2MovieAgent[provider].defaultModel;
+    const model = info.movieParams?.model ?? defaultMovieModel(provider, info.movieParams);
     const duration = beat ? beatDurationMetric(beat) : estimated(DEFAULT_MOVIE_DURATION_SEC);
-    const predictSec = snapMovieDuration(provider, model, duration);
+    const predictSec = isVeoExtension(provider, model, duration)
+        ? metric(veoExtendedSeconds(duration.value), duration.precision === "exact")
+        : snapMovieDuration(provider, model, duration);
     return attachCost({ process: refKey === undefined ? "movie" : "movieReference", beatIndex, refKey, provider, model, predictSec });
 };
 const estimateSoundEffect = (style, beat, beatIndex) => {

package/lib/utils/gemini_tts.d.ts

@@ -0,0 +1,45 @@
+export declare const GEMINI_INTERACTIONS_TTS_MODELS: ReadonlySet<string>;
+export declare const GEMINI_TTS_SAMPLE_RATE = 24000;
+export declare const geminiDirectorsNotesPrompt: (text: string, instructions?: string) => string;
+export declare const geminiTtsInputText: (model: string, text: string, instructions?: string) => string;
+export declare const usesGeminiInteractionsTts: (model: string) => boolean;
+export type GeminiInteractionsTtsInput = {
+    model: string;
+    text: string;
+    voice: string;
+    instructions?: string;
+};
+export declare const buildGeminiInteractionsTtsRequest: ({ model, text, voice, instructions }: GeminiInteractionsTtsInput) => {
+    model: string;
+    input: {
+        type: "user_input";
+        content: {
+            annotations?: {
+                type: "speech_metadata";
+                style: string;
+            }[] | undefined;
+            type: "text";
+            text: string;
+        }[];
+    }[];
+    response_format: {
+        type: "audio";
+        mime_type: string;
+        sample_rate: number;
+    };
+    generation_config: {
+        speech_config: {
+            voice: string;
+        }[];
+    };
+};
+type InteractionAudio = {
+    data?: string;
+    mime_type?: string;
+    sample_rate?: number;
+};
+export declare const pcmFromInteractionAudio: (audio: InteractionAudio | undefined) => {
+    rawPcm: Buffer;
+    sampleRate: number;
+};
+export {};

package/lib/utils/gemini_tts.js

@@ -0,0 +1,35 @@
+// Gemini 3.8 TTS takes the Interactions API: the text is read verbatim, so speaker direction goes in a
+// speech_metadata annotation instead of the prompt, and raw PCM must be asked for (the unary default is WAV).
+export const GEMINI_INTERACTIONS_TTS_MODELS = new Set(["gemini-3.8-flash-tts", "gemini-3.8-flash-lite-tts"]);
+export const GEMINI_TTS_SAMPLE_RATE = 24000;
+// 2.5 TTS has no field for direction, so it goes in the prompt: https://ai.google.dev/gemini-api/docs/speech-generation#controllable
+export const geminiDirectorsNotesPrompt = (text, instructions) => instructions ? ["### DIRECTOR'S NOTES", instructions, "", "#### TRANSCRIPT", text].join("\n") : text;
+// The text the model is given for one line: 3.8 gets the transcript plus the style annotation, 2.5 the wrapped prompt.
+export const geminiTtsInputText = (model, text, instructions) => usesGeminiInteractionsTts(model) ? [text, instructions].filter(Boolean).join("\n") : geminiDirectorsNotesPrompt(text, instructions);
+export const usesGeminiInteractionsTts = (model) => GEMINI_INTERACTIONS_TTS_MODELS.has(model);
+export const buildGeminiInteractionsTtsRequest = ({ model, text, voice, instructions }) => ({
+    model,
+    input: [
+        {
+            type: "user_input",
+            content: [{ type: "text", text, ...(instructions ? { annotations: [{ type: "speech_metadata", style: instructions }] } : {}) }],
+        },
+    ],
+    response_format: { type: "audio", mime_type: "audio/l16", sample_rate: GEMINI_TTS_SAMPLE_RATE },
+    generation_config: { speech_config: [{ voice }] },
+});
+const sampleRateOf = (audio) => {
+    if (typeof audio.sample_rate === "number" && audio.sample_rate > 0)
+        return audio.sample_rate;
+    const match = audio.mime_type?.match(/rate=(\d+)/);
+    return match ? Number(match[1]) : GEMINI_TTS_SAMPLE_RATE;
+};
+// Anything but headerless 16-bit PCM would be fed to the PCM encoder as noise, so it is an error, not a guess.
+export const pcmFromInteractionAudio = (audio) => {
+    if (!audio?.data)
+        throw new Error("No audio data returned");
+    if (!audio.mime_type?.toLowerCase().startsWith("audio/l16")) {
+        throw new Error(`Gemini TTS returned ${audio.mime_type ?? "audio of an unknown type"}, expected audio/l16`);
+    }
+    return { rawPcm: Buffer.from(audio.data, "base64"), sampleRate: sampleRateOf(audio) };
+};

package/lib/utils/remotion/network_guard.d.ts

@@ -0,0 +1,5 @@
+import type { HeadlessBrowser } from "@remotion/renderer";
+export type NetworkGuard = {
+    failures: string[];
+};
+export declare const guardSceneNetwork: (browser: HeadlessBrowser, serverPort: number, onBlocked: (url: string) => void) => Promise<NetworkGuard>;