mulmocast 2.15.0 → 2.16.0

package/README.md

@@ -103,11 +103,6 @@ Create a `.env` file in your project directory with the following API keys:
 OPENAI_API_KEY=your_openai_api_key
 ```
 
-#### (Optional) For the advanced image generation model
-```bash
-DEFAULT_OPENAI_IMAGE_MODEL=gpt-image-1 # for the advanced image generation model
-```
-
 #### (Optional) For Google's image and TTS.
 ```bash
 GEMINI_API_KEY=your_google_gemini_api_key
@@ -298,10 +293,9 @@ Optionally, you can specify *__clipboard* as the script file name to paste the s
 Verify your `.env` file contains:
 ```bash
 OPENAI_API_KEY=your_openai_api_key
-DEFAULT_OPENAI_IMAGE_MODEL=gpt-image-1 # required for high-quality Ghibli-style images
 ```
 
-> **Note:** Ensure your OpenAI organization is verified to access the gpt-image-1 model. Visit https://platform.openai.com/settings/organization/general and complete the "Verifications" section.
+> **Note:** Ensure your OpenAI organization is verified to access the GPT Image models. Visit https://platform.openai.com/settings/organization/general and complete the "Verifications" section.
 
 ### Step 2: Generate a Ghibli-style MulmoScript
 ```bash

package/lib/actions/image_references.d.ts

@@ -1,6 +1,28 @@
-import { MulmoStudioContext, MulmoBeat, MulmoImagePromptMedia } from "../types/index.js";
+import { MulmoPresentationStyleMethods } from "../methods/index.js";
+import { MulmoStudioContext, MulmoBeat, MulmoImagePromptMedia, MulmoCanvasDimension } from "../types/index.js";
 import type { ResolvedImageReference } from "../methods/mulmo_image_prompt_media.js";
 export type { ResolvedImageReference } from "../methods/mulmo_image_prompt_media.js";
+type ImageAgentInfo = ReturnType<typeof MulmoPresentationStyleMethods.getImageAgentInfo>;
+type MovieAgentInfo = ReturnType<typeof MulmoPresentationStyleMethods.getMovieAgentInfo>;
+export declare const referenceImageGraphParams: (imageAgentInfo: ImageAgentInfo, canvasSize: MulmoCanvasDimension) => {
+    model: string | undefined;
+    canvasSize: {
+        width: number;
+        height: number;
+    };
+    vertexai_project: string | undefined;
+    vertexai_location: string | undefined;
+};
+export declare const referenceMovieGraphParams: (movieAgentInfo: MovieAgentInfo, canvasSize: MulmoCanvasDimension) => {
+    model: string | undefined;
+    canvasSize: {
+        width: number;
+        height: number;
+    };
+    generateAudio: boolean | undefined;
+    vertexai_project: string | undefined;
+    vertexai_location: string | undefined;
+};
 export declare const generateReferenceImage: (inputs: {
     context: MulmoStudioContext;
     key: string;

package/lib/actions/image_references.js

@@ -6,6 +6,20 @@ import { MulmoPresentationStyleMethods, MulmoMediaSourceMethods, MulmoImagePromp
 import { ImageMediaType, } from "../types/index.js";
 import { imageOpenaiAgent, mediaMockAgent, imageGenAIAgent, imageReplicateAgent, movieGenAIAgent, movieReplicateAgent } from "../agents/index.js";
 import { agentGenerationError, imageReferenceAction, imageFileTarget, movieFileTarget } from "../utils/error_cause.js";
+// Reference media are generated outside the beat graphs, so they must carry the same Vertex AI target the beats do.
+export const referenceImageGraphParams = (imageAgentInfo, canvasSize) => ({
+    model: imageAgentInfo.imageParams.model,
+    canvasSize,
+    vertexai_project: imageAgentInfo.imageParams.vertexai_project,
+    vertexai_location: imageAgentInfo.imageParams.vertexai_location,
+});
+export const referenceMovieGraphParams = (movieAgentInfo, canvasSize) => ({
+    model: movieAgentInfo.movieParams.model,
+    canvasSize,
+    generateAudio: movieAgentInfo.movieParams.generateAudio,
+    vertexai_project: movieAgentInfo.movieParams.vertexai_project,
+    vertexai_location: movieAgentInfo.movieParams.vertexai_location,
+});
 // Resolve the normalized reference list to local file paths (with labels), in reference order.
 // imageKey is the imageRefs key of the image being generated (used for warnings and url
 // download file naming). A named reference that is missing from imageRefs warns and is
@@ -62,10 +76,7 @@ export const generateReferenceImage = async (inputs) => {
                         sessionType: "imageReference",
                     },
                 },
-                params: {
-                    model: imageAgentInfo.imageParams.model,
-                    canvasSize: image.canvasSize ?? context.presentationStyle.canvasSize,
-                },
+                params: referenceImageGraphParams(imageAgentInfo, image.canvasSize ?? context.presentationStyle.canvasSize),
             },
         },
     };
@@ -145,11 +156,7 @@ const generateReferenceMovie = async (inputs) => {
                         sessionType: "imageReference",
                     },
                 },
-                params: {
-                    model: movieAgentInfo.movieParams.model,
-                    canvasSize: context.presentationStyle.canvasSize,
-                    generateAudio: movieAgentInfo.movieParams.generateAudio,
-                },
+                params: referenceMovieGraphParams(movieAgentInfo, context.presentationStyle.canvasSize),
             },
         },
     };

package/lib/agents/image_genai_agent.js

@@ -1,6 +1,6 @@
 import fs from "fs";
 import { GraphAILogger } from "graphai";
-import { provider2ImageAgent, deprecatedGoogleImageModelHints, vertexAIGlobalOnlyImageModels, } from "../types/provider2agent.js";
+import { provider2ImageAgent, deprecatedGoogleImageModelHints, vertexAIGlobalOnlyImageModels, isGeminiImageModel, vertexImageLocation, } from "../types/provider2agent.js";
 import { apiKeyMissingError, agentIncorrectAPIKeyError, agentGenerationError, agentInvalidResponseError, imageAction, imageFileTarget, unsupportedModelTarget, hasCause, getGenAIErrorReason, resultify, } from "../utils/error_cause.js";
 import { getAspectRatio } from "../utils/utils.js";
 import { ASPECT_RATIOS, PRO_ASPECT_RATIOS } from "../types/const.js";
@@ -85,7 +85,7 @@ export const imageGenAIAgent = async ({ namedInputs, params, config, }) => {
     const apiKey = config?.apiKey;
     const ai = params.vertexai_project
         ? (() => {
-            const location = params.vertexai_location ?? "us-central1";
+            const location = vertexImageLocation(model, params.vertexai_location);
             if (vertexAIGlobalOnlyImageModels.has(model) && location !== "global") {
                 GraphAILogger.warn(`imageGenAIAgent: model "${model}" on Vertex AI is only available in location "global", but got "${location}". Set imageParams.vertexai_location to "global".`);
             }
@@ -99,7 +99,7 @@ export const imageGenAIAgent = async ({ namedInputs, params, config, }) => {
             }
             return new GoogleGenAI({ apiKey, httpOptions: { timeout: GENAI_REQUEST_TIMEOUT_MS } });
         })();
-    if (model === "gemini-2.5-flash-image" || model === "gemini-3.1-flash-image-preview" || model === "gemini-3-pro-image-preview") {
+    if (isGeminiImageModel(model)) {
         const contentParams = (() => {
             const contents = getGeminiContents(prompt, referenceImages);
             return {

package/lib/agents/movie_genai_agent.js

@@ -5,7 +5,7 @@ import { apiKeyMissingError, agentGenerationError, agentInvalidResponseError, im
 import { getAspectRatio } from "../utils/utils.js";
 import { ffmpegGetMediaDuration } from "../utils/ffmpeg_utils.js";
 import { ASPECT_RATIOS } from "../types/const.js";
-import { getModelDuration, provider2MovieAgent, AUDIO_MODE_NEVER, AUDIO_MODE_ALWAYS } from "../types/provider2agent.js";
+import { getModelDuration, provider2MovieAgent, AUDIO_MODE_NEVER, AUDIO_MODE_ALWAYS, defaultMovieModel, veoExtensionCount, VEO_EXTENSION_MODEL, VEO_SEGMENT_SEC, } from "../types/provider2agent.js";
 // Per-request timeout so a stalled GenAI video API call rejects instead of hanging.
 const GENAI_REQUEST_TIMEOUT_MS = 120_000;
 // Wall-clock cap on the long-running video operation poll loop (Veo runs minutes).
@@ -89,14 +89,12 @@ const createVeo31Payload = (model, prompt, aspectRatio, source) => ({
     ...source,
 });
 const generateExtendedVideo = async (ai, model, prompt, aspectRatio, imagePath, requestedDuration, movieFile, isVertexAI) => {
-    const initialDuration = 8;
-    const maxExtensionDuration = 8;
-    const extensionsNeeded = Math.ceil((requestedDuration - initialDuration) / maxExtensionDuration);
+    const extensionsNeeded = veoExtensionCount(requestedDuration);
     GraphAILogger.info(`Veo 3.1 video extension: ${extensionsNeeded} extensions needed for ${requestedDuration}s target`);
     const generateIteration = async (iteration, accumulatedDuration, previousVideo) => {
         const isInitial = iteration === 0;
         const remainingDuration = requestedDuration - accumulatedDuration;
-        const extensionDuration = isInitial ? initialDuration : (getModelDuration("google", model, remainingDuration) ?? maxExtensionDuration);
+        const extensionDuration = isInitial ? VEO_SEGMENT_SEC : (getModelDuration("google", model, remainingDuration) ?? VEO_SEGMENT_SEC);
         const getSource = () => {
             if (isInitial)
                 return imagePath ? { image: loadImageAsBase64(imagePath) } : undefined;
@@ -172,7 +170,7 @@ const generateStandardVideo = async (ai, model, prompt, aspectRatio, imagePath,
 export const movieGenAIAgent = async ({ namedInputs, params, config, }) => {
     const { prompt, imagePath, lastFrameImagePath, referenceImages, movieFile } = namedInputs;
     const aspectRatio = getAspectRatio(params.canvasSize, ASPECT_RATIOS);
-    const model = params.model ?? provider2MovieAgent.google.defaultModel;
+    const model = params.model ?? defaultMovieModel("google", params);
     const apiKey = config?.apiKey;
     try {
         const requestedDuration = params.duration ?? 8;
@@ -211,7 +209,7 @@ export const movieGenAIAgent = async ({ namedInputs, params, config, }) => {
                 return new GoogleGenAI({ apiKey, httpOptions: { timeout: GENAI_REQUEST_TIMEOUT_MS } });
             })();
         // Veo 3.1: Video extension mode for videos longer than 8s
-        if (model === "veo-3.1-generate-preview" && requestedDuration > 8 && params.canvasSize) {
+        if (model === VEO_EXTENSION_MODEL && requestedDuration > VEO_SEGMENT_SEC && params.canvasSize) {
             return generateExtendedVideo(ai, model, prompt, aspectRatio, imagePath, requestedDuration, movieFile, isVertexAI);
         }
         // Standard mode

package/lib/data/scriptTemplates.js

@@ -484,7 +484,7 @@ export const scriptTemplates = [
         ],
         filename: "html",
         htmlImageParams: {
-            model: "claude-sonnet-4-5-20250929",
+            model: "claude-sonnet-4-6",
             provider: "anthropic",
         },
         lang: "en",
@@ -987,7 +987,7 @@ export const scriptTemplates = [
         ],
         filename: "presentation",
         htmlImageParams: {
-            model: "claude-sonnet-4-5-20250929",
+            model: "claude-sonnet-4-6",
             provider: "anthropic",
         },
         lang: "en",

package/lib/data/templateDataSet.js

@@ -49,7 +49,7 @@ export const templateDataSet = {
         "```",
     html: "Another LLM will generate actual slides from the prompt and data for each beat. Adding optional data would help it to generate more compelling slide. Mention the reference in one of beats, if it exists. The valid type of reference is 'article', 'paper', 'image', 'video', 'audio'. Use the JSON below as a template.\n" +
         "```JSON\n" +
-        '{"$mulmocast":{"version":"1.1","credit":"closing"},"references":[{"url":"https://www.somegreatwebsite.com/article/123","title":"Title of the article we are referencing","type":"[TYPE OF ARTICLE: article, paper, image, video, audio]"}],"title":"[TITLE: Brief, engaging title for the topic]","htmlImageParams":{"provider":"anthropic","model":"claude-sonnet-4-5-20250929"},"lang":"en","beats":[{"text":"[NARRATION: Narration for the beat.]","htmlPrompt":{"prompt":"[PROMPT to create appropriate HTML page for the beat.]"}},{"text":"[NARRATION: Narration for the beat.]","htmlPrompt":{"prompt":"[PROMPT to create appropriate HTML page for the beat with the data.]","data":{"description":"DATA TO BE PRESENTED IN THIS BEAT (in any format)]","net_income":{"Q2 FY2024":320,"Q3 FY2024":333,"Q4 FY2024":350},"unit":"USD (Million)"}}}],"canvasSize":{"width":1536,"height":1024}}\n' +
+        '{"$mulmocast":{"version":"1.1","credit":"closing"},"references":[{"url":"https://www.somegreatwebsite.com/article/123","title":"Title of the article we are referencing","type":"[TYPE OF ARTICLE: article, paper, image, video, audio]"}],"title":"[TITLE: Brief, engaging title for the topic]","htmlImageParams":{"provider":"anthropic","model":"claude-sonnet-4-6"},"lang":"en","beats":[{"text":"[NARRATION: Narration for the beat.]","htmlPrompt":{"prompt":"[PROMPT to create appropriate HTML page for the beat.]"}},{"text":"[NARRATION: Narration for the beat.]","htmlPrompt":{"prompt":"[PROMPT to create appropriate HTML page for the beat with the data.]","data":{"description":"DATA TO BE PRESENTED IN THIS BEAT (in any format)]","net_income":{"Q2 FY2024":320,"Q3 FY2024":333,"Q4 FY2024":350},"unit":"USD (Million)"}}}],"canvasSize":{"width":1536,"height":1024}}\n' +
         "```",
     image_prompt: "Another AI will generate images for each beat based on the image prompt of that beat. Movie prompts must be written in English. Mention the reference in one of beats, if it exists. Use the JSON below as a template.\n" +
         "```JSON\n" +

package/lib/methods/mulmo_presentation_style.js

@@ -8,7 +8,7 @@ import { userAssert } from "../utils/utils.js";
 import { slideThemes } from "../data/slideThemes.js";
 import { text2ImageProviderSchema, text2HtmlImageProviderSchema, text2MovieProviderSchema, text2SpeechProviderSchema, mulmoCanvasDimensionSchema, mulmoTransitionSchema, mulmoFillOptionSchema, } from "../types/schema.js";
 import { MulmoBeatMethods } from "./mulmo_beat.js";
-import { provider2ImageAgent, provider2MovieAgent, provider2LLMAgent, provider2TTSAgent, provider2SoundEffectAgent, provider2LipSyncAgent, defaultProviders, getModelAudio, AUDIO_MODE_ALWAYS, } from "../types/provider2agent.js";
+import { provider2ImageAgent, provider2MovieAgent, defaultMovieModel, provider2LLMAgent, provider2TTSAgent, provider2SoundEffectAgent, provider2LipSyncAgent, defaultProviders, getModelAudio, AUDIO_MODE_ALWAYS, } from "../types/provider2agent.js";
 const defaultTextSlideStyles = [
     '*,*::before,*::after{box-sizing:border-box}body,h1,h2,h3,h4,p,figure,blockquote,dl,dd{margin:0}ul[role="list"],ol[role="list"]{list-style:none}html:focus-within{scroll-behavior:smooth}body{min-height:100vh;text-rendering:optimizeSpeed;line-height:1.5}a:not([class]){text-decoration-skip-ink:auto}img,picture{max-width:100%;display:block}input,button,textarea,select{font:inherit}@media(prefers-reduced-motion:reduce){html:focus-within{scroll-behavior:auto}*,*::before,*::after{animation-duration:.01ms !important;animation-iteration-count:1 !important;transition-duration:.01ms !important;scroll-behavior:auto !important}}',
     "body { margin: 60px; margin-top: 40px; color:#333; background-color:#fff; font-size: 30px; font-family: Arial, sans-serif; box-sizing: border-box; height: 100vh }",
@@ -160,7 +160,7 @@ export const MulmoPresentationStyleMethods = {
         const imageParams = { ...presentationStyle.imageParams, ...beat?.imageParams };
         const provider = MulmoPresentationStyleMethods.getText2ImageProvider(imageParams?.provider) ?? defaultProviders.text2image;
         const agentInfo = provider2ImageAgent[provider];
-        // The default text2image model is gpt-image-1 from OpenAI.
+        // The default text2image model is gpt-image-2.5-sunburst from OpenAI.
         const defaultImageParams = {
             provider,
             model: agentInfo.defaultModel,
@@ -189,7 +189,7 @@ export const MulmoPresentationStyleMethods = {
             return true;
         }
         const provider = text2MovieProviderSchema.parse(movieParams?.provider ?? defaultProviders.text2movie);
-        const model = movieParams?.model ?? provider2MovieAgent[provider].defaultModel;
+        const model = movieParams?.model ?? defaultMovieModel(provider, movieParams);
         return getModelAudio(provider, model)?.mode === AUDIO_MODE_ALWAYS;
     },
     getSoundEffectAgentInfo(presentationStyle, beat) {

package/lib/types/provider2agent.d.ts

@@ -43,20 +43,22 @@ export declare const provider2TTSAgent: {
 };
 export declare const gptImages: string[];
 export declare const deprecatedOpenAIImageModelHints: {
-    readonly "dall-e-2": "Use 'gpt-image-1' or another supported model.";
-    readonly "dall-e-3": "Use 'gpt-image-1' or another supported model.";
+    readonly "dall-e-2": "Use 'gpt-image-2.5-sunburst' or another supported model.";
+    readonly "dall-e-3": "Use 'gpt-image-2.5-sunburst' or another supported model.";
 };
 export type DeprecatedOpenAIImageModel = keyof typeof deprecatedOpenAIImageModelHints;
 export declare const deprecatedGoogleImageModelHints: {
-    readonly "imagen-3.0-generate-002": "Use 'gemini-2.5-flash-image' or 'gemini-3-pro-image-preview' instead.";
-    readonly "imagen-4.0-generate-001": "Use 'gemini-2.5-flash-image' or 'gemini-3-pro-image-preview' instead.";
-    readonly "imagen-4.0-ultra-generate-001": "Use 'gemini-2.5-flash-image' or 'gemini-3-pro-image-preview' instead.";
-    readonly "imagen-4.0-fast-generate-001": "Use 'gemini-2.5-flash-image' or 'gemini-3-pro-image-preview' instead.";
-    readonly "imagen-4.0-generate-preview-06-06": "Use 'gemini-2.5-flash-image' or 'gemini-3-pro-image-preview' instead.";
-    readonly "imagen-4.0-ultra-generate-preview-06-06": "Use 'gemini-2.5-flash-image' or 'gemini-3-pro-image-preview' instead.";
+    readonly "imagen-3.0-generate-002": "Use 'gemini-3.1-flash-lite-image' or 'gemini-3-pro-image' instead.";
+    readonly "imagen-4.0-generate-001": "Use 'gemini-3.1-flash-lite-image' or 'gemini-3-pro-image' instead.";
+    readonly "imagen-4.0-ultra-generate-001": "Use 'gemini-3.1-flash-lite-image' or 'gemini-3-pro-image' instead.";
+    readonly "imagen-4.0-fast-generate-001": "Use 'gemini-3.1-flash-lite-image' or 'gemini-3-pro-image' instead.";
+    readonly "imagen-4.0-generate-preview-06-06": "Use 'gemini-3.1-flash-lite-image' or 'gemini-3-pro-image' instead.";
+    readonly "imagen-4.0-ultra-generate-preview-06-06": "Use 'gemini-3.1-flash-lite-image' or 'gemini-3-pro-image' instead.";
 };
 export type DeprecatedGoogleImageModel = keyof typeof deprecatedGoogleImageModelHints;
+export declare const isGeminiImageModel: (model: string) => boolean;
 export declare const vertexAIGlobalOnlyImageModels: ReadonlySet<string>;
+export declare const vertexImageLocation: (model: string, location?: string) => string;
 export declare const provider2ImageAgent: {
     openai: {
         agentName: string;
@@ -169,34 +171,34 @@ export declare const provider2LipSyncAgent: {
 export declare const provider2LLMAgent: {
     readonly openai: {
         readonly agentName: "openAIAgent";
-        readonly defaultModel: "gpt-5";
+        readonly defaultModel: "gpt-5.6-sol";
         readonly keyName: "OPENAI_API_KEY";
         readonly baseURLKeyName: "OPENAI_BASE_URL";
         readonly apiVersionKeyName: "OPENAI_API_VERSION";
         readonly max_tokens: 8192;
-        readonly models: readonly ["gpt-5", "gpt-5-nano", "gpt-5-mini", "gpt-4.1", "gpt-4.1-mini", "gpt-4.1-nano", "o3", "o3-mini", "o3-pro", "o1", "o1-pro", "gpt-4o", "gpt-4o-mini"];
+        readonly models: readonly ["gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna", "gpt-5", "gpt-5-nano", "gpt-5-mini", "gpt-4.1", "gpt-4.1-mini", "gpt-4.1-nano", "o3", "o3-mini", "o3-pro", "o1", "o1-pro", "gpt-4o", "gpt-4o-mini"];
     };
     readonly anthropic: {
         readonly agentName: "anthropicAgent";
-        readonly defaultModel: "claude-sonnet-4-5-20250929";
+        readonly defaultModel: "claude-sonnet-4-6";
         readonly max_tokens: 8192;
-        readonly models: readonly ["claude-opus-4-1-20250805", "claude-opus-4-20250514", "claude-sonnet-4-20250514", "claude-sonnet-4-5-20250929", "claude-haiku-4-5-20251001"];
+        readonly models: readonly ["claude-sonnet-4-6", "claude-sonnet-4-5-20250929", "claude-haiku-4-5-20251001"];
         readonly keyName: "ANTHROPIC_API_KEY";
         readonly apiKeyNameOverride: "ANTHROPIC_API_TOKEN";
     };
     readonly gemini: {
         readonly agentName: "geminiAgent";
-        readonly defaultModel: "gemini-2.5-flash";
+        readonly defaultModel: "gemini-3.8-flash";
         readonly max_tokens: 8192;
-        readonly models: readonly ["gemini-3.1-pro-preview", "gemini-3-flash-preview", "gemini-2.5-pro", "gemini-2.5-flash", "gemini-2.5-flash-lite"];
+        readonly models: readonly ["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"];
         readonly keyName: "GEMINI_API_KEY";
     };
     readonly groq: {
         readonly agentName: "groqAgent";
-        readonly defaultModel: "llama-3.1-8b-instant";
+        readonly defaultModel: "openai/gpt-oss-20b";
         readonly keyName: "GROQ_API_KEY";
         readonly max_tokens: 4096;
-        readonly models: readonly ["llama-3.1-8b-instant", "llama-3.3-70b-versatile", "deepseek-r1-distill-llama-70b", "openai/gpt-oss-120b", "openai/gpt-oss-20b"];
+        readonly models: readonly ["openai/gpt-oss-20b", "openai/gpt-oss-120b"];
     };
     readonly mock: {
         readonly agentName: "mediaMockAgent";
@@ -229,6 +231,13 @@ export declare const ProviderName: {
 export declare const isImageProvider: (provider: string | undefined) => provider is keyof typeof provider2ImageAgent;
 export declare const getMaxImageReferenceImages: (provider: string | undefined, model: string) => number | undefined;
 export declare const getModelAudio: (provider: keyof typeof provider2MovieAgent, model: string) => MovieAudioSpec | undefined;
+export declare const defaultMovieModel: (provider: keyof typeof provider2MovieAgent, movieParams?: {
+    vertexai_project?: string;
+}) => string;
+export declare const VEO_EXTENSION_MODEL = "veo-3.1-generate-preview";
+export declare const VEO_SEGMENT_SEC = 8;
+export declare const veoExtensionCount: (requestedSec: number) => number;
+export declare const veoExtendedSeconds: (requestedSec: number) => number;
 export declare const getModelDuration: (provider: keyof typeof provider2MovieAgent, model: string, movieDuration?: number) => number | undefined;
 export type ModelPricing = {
     unit: "tokens" | "chars" | "seconds" | "images";

package/lib/types/provider2agent.js

@@ -44,13 +44,13 @@ 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"];
+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 +59,21 @@ 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 +81,9 @@ 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 +91,16 @@ 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 +146,6 @@ 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 +192,14 @@ 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 +223,7 @@ 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 +231,14 @@ 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 +251,13 @@ 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 +277,7 @@ 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 +292,26 @@ 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 +319,7 @@ 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 +327,7 @@ 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 +337,27 @@ 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 +377,9 @@ 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 +390,8 @@ 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 +399,14 @@ 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 +478,16 @@ 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 +505,31 @@ 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 +578,14 @@ 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 +606,8 @@ 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 },
@@ -601,9 +626,15 @@ 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: {
@@ -614,9 +645,13 @@ export const modelPricing = {
     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 +664,16 @@ 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/utils/estimate_usage.js

@@ -4,7 +4,7 @@
 // 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, 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";
@@ -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/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>;

package/lib/utils/remotion/network_guard.js

@@ -0,0 +1,48 @@
+import { WEBRTC_REMOVAL_SCRIPT, isAllowedSceneRequest } from "./network_policy.js";
+const isRecord = (value) => typeof value === "object" && value !== null;
+const isPausedRequest = (value) => isRecord(value) && typeof value.requestId === "string" && isRecord(value.request) && typeof value.request.url === "string";
+const isAttachedTarget = (value) => isRecord(value) && typeof value.sessionId === "string" && isRecord(value.targetInfo) && typeof value.targetInfo.type === "string";
+// Remotion types only the CDP commands it uses itself; Fetch and Runtime.runIfWaitingForDebugger
+// are sent through the same send(), which resolves on the reply and rejects on a CDP error.
+const sendCommand = async (channel, method, params = {}) => {
+    const send = Reflect.get(channel, "send");
+    if (typeof send !== "function")
+        throw new Error(`remotion: cannot send ${method}: the browser connection has no send()`);
+    await send.call(channel, method, params);
+};
+const FRAME_TARGET_TYPES = ["page", "iframe"];
+const answerRequest = async (connection, paused, serverPort, onBlocked) => {
+    if (isAllowedSceneRequest({ url: paused.request.url, isDocument: paused.resourceType === "Document" }, serverPort)) {
+        await sendCommand(connection, "Fetch.continueRequest", { requestId: paused.requestId });
+        return;
+    }
+    onBlocked(paused.request.url);
+    await sendCommand(connection, "Fetch.failRequest", { requestId: paused.requestId, errorReason: "BlockedByClient" });
+};
+// A target that could not be guarded stays paused, so its render fails instead of running unguarded.
+const guardTarget = async (browser, attached) => {
+    const session = browser.connection.session(attached.sessionId);
+    if (!session)
+        throw new Error(`remotion: no CDP session for a new ${attached.targetInfo.type}`);
+    if (FRAME_TARGET_TYPES.includes(attached.targetInfo.type)) {
+        await session.send("Page.enable");
+        await session.send("Page.addScriptToEvaluateOnNewDocument", { source: WEBRTC_REMOVAL_SCRIPT });
+    }
+    await sendCommand(session, "Runtime.runIfWaitingForDebugger");
+};
+export const guardSceneNetwork = async (browser, serverPort, onBlocked) => {
+    const guard = { failures: [] };
+    const fail = (error) => guard.failures.push(error instanceof Error ? error.message : String(error));
+    const connection = browser.connection;
+    connection.on("Fetch.requestPaused", (event) => {
+        if (isPausedRequest(event))
+            answerRequest(connection, event, serverPort, onBlocked).catch(fail);
+    });
+    connection.on("Target.attachedToTarget", (event) => {
+        if (isAttachedTarget(event))
+            guardTarget(browser, event).catch(fail);
+    });
+    await sendCommand(connection, "Fetch.enable", { patterns: [{ urlPattern: "*" }] });
+    await connection.send("Target.setAutoAttach", { autoAttach: true, waitForDebuggerOnStart: true, flatten: true });
+    return guard;
+};

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

@@ -0,0 +1,10 @@
+export type SceneRequest = {
+    url: string;
+    isDocument: boolean;
+};
+export declare const isAllowedSceneRequest: ({ url, isDocument }: SceneRequest, serverPort: number) => boolean;
+export declare const isAllowedMediaDownload: (src: string) => boolean;
+export declare const REMOTION_CONTENT_SECURITY_POLICY = "connect-src 'self' data: blob:";
+export declare const withContentSecurityPolicy: (html: string) => string;
+export declare const WEBRTC_GLOBALS: readonly ["RTCPeerConnection", "webkitRTCPeerConnection", "RTCDataChannel", "RTCIceCandidate", "RTCSessionDescription"];
+export declare const WEBRTC_REMOVAL_SCRIPT: string;

package/lib/utils/remotion/network_policy.js

@@ -0,0 +1,51 @@
+// What a remotion scene's browser may reach: only the bundle server started for that render, and
+// in-memory URLs. Kept free of I/O so the rules can be tested without a browser.
+const LOOPBACK_HOSTS = ["localhost", "127.0.0.1", "[::1]"];
+const IN_MEMORY_SCHEMES = ["data:", "blob:"];
+const parseUrl = (url) => {
+    try {
+        return new URL(url);
+    }
+    catch {
+        return undefined;
+    }
+};
+// Remotion's /proxy makes the bundle server download its `src` from Node, on the scene's behalf.
+const REMOTION_PROXY_PATH = "/proxy";
+// The only page that carries the CSP; any other HTML the server returns (a directory listing, a 404)
+// would run without it, so no other path may be loaded as a document.
+const SCENE_DOCUMENT_PATHS = ["/", "/index.html"];
+// Another port on localhost is another service, so only the render's own port counts as local.
+const isBundleServer = (parsed, serverPort) => parsed.protocol === "http:" && LOOPBACK_HOSTS.includes(parsed.hostname) && parsed.port === String(serverPort);
+export const isAllowedSceneRequest = ({ url, isDocument }, serverPort) => {
+    const parsed = parseUrl(url);
+    if (!parsed)
+        return false;
+    if (IN_MEMORY_SCHEMES.includes(parsed.protocol))
+        return true;
+    if (!isBundleServer(parsed, serverPort) || parsed.pathname.startsWith(REMOTION_PROXY_PATH))
+        return false;
+    return !isDocument || SCENE_DOCUMENT_PATHS.includes(parsed.pathname);
+};
+// Remotion downloads <Audio> / <Video> sources from Node while rendering, outside the browser;
+// a scene uses no media files, so only inline data survives.
+export const isAllowedMediaDownload = (src) => parseUrl(src)?.protocol === "data:";
+// The browser-level request block does not see WebSocket or EventSource connections; this does.
+export const REMOTION_CONTENT_SECURITY_POLICY = "connect-src 'self' data: blob:";
+const HEAD_TAG = /<head(\s[^>]*)?>/i;
+export const withContentSecurityPolicy = (html) => {
+    const match = HEAD_TAG.exec(html);
+    if (!match)
+        throw new Error("remotion: the bundle's index.html has no <head>, so the network policy cannot be applied");
+    const meta = `<meta http-equiv="Content-Security-Policy" content="${REMOTION_CONTENT_SECURITY_POLICY}">`;
+    const insertAt = match.index + match[0].length;
+    return [html.slice(0, insertAt), meta, html.slice(insertAt)].join("");
+};
+export const WEBRTC_GLOBALS = ["RTCPeerConnection", "webkitRTCPeerConnection", "RTCDataChannel", "RTCIceCandidate", "RTCSessionDescription"];
+// Run before any page script in every frame: WebRTC opens UDP connections that neither the request
+// block nor CSP covers. Non-configurable, so scene code cannot put the constructors back.
+export const WEBRTC_REMOVAL_SCRIPT = [
+    `for (const name of ${JSON.stringify(WEBRTC_GLOBALS)}) {`,
+    "  try { Object.defineProperty(globalThis, name, { value: undefined, writable: false, configurable: false }); } catch (_) {}",
+    "}",
+].join("\n");

package/lib/utils/remotion/render.js

@@ -1,10 +1,15 @@
 import fs from "node:fs";
 import path from "node:path";
+import net from "node:net";
 import { createRequire } from "node:module";
+import { GraphAILogger } from "graphai";
 import { REMOTION_COMPOSITION_ID, REMOTION_ENTRY_FILE, buildEntrySource } from "./claude_prompt.js";
 import { missingRemotionPackages, remotionInstallCommand } from "./packages.js";
+import { isAllowedMediaDownload, withContentSecurityPolicy } from "./network_policy.js";
+import { guardSceneNetwork } from "./network_guard.js";
 // WebGL (three.js scenes) needs the ANGLE backend; the default one cannot create a context headless.
 const CHROMIUM_OPTIONS = { gl: "angle" };
+const MAX_LOGGED_URL_LENGTH = 200;
 const loadRemotion = async () => {
     try {
         const [bundler, renderer] = await Promise.all([import("@remotion/bundler"), import("@remotion/renderer")]);
@@ -40,8 +45,7 @@ const remotionNodeModulesDir = () => {
     const remotionPackageJson = createRequire(import.meta.url).resolve("remotion/package.json");
     return path.dirname(path.dirname(remotionPackageJson));
 };
-export const renderRemotionScene = async ({ workDir, props, videoPath, stillPath, extraStills = [] }) => {
-    const { bundle, renderer } = await loadRemotion();
+const bundleScene = async (bundle, workDir) => {
     const entryPoint = path.join(workDir, REMOTION_ENTRY_FILE);
     fs.writeFileSync(entryPoint, buildEntrySource());
     const nodeModulesDir = remotionNodeModulesDir();
@@ -53,13 +57,53 @@ export const renderRemotionScene = async ({ workDir, props, videoPath, stillPath
             resolve: { ...config.resolve, modules: [...(config.resolve?.modules ?? ["node_modules"]), nodeModulesDir] },
         }),
     });
-    const composition = await renderer.selectComposition({ serveUrl, id: REMOTION_COMPOSITION_ID, inputProps: props, chromiumOptions: CHROMIUM_OPTIONS });
+    const indexHtml = path.join(serveUrl, "index.html");
+    fs.writeFileSync(indexHtml, withContentSecurityPolicy(fs.readFileSync(indexHtml, "utf8")));
+    return serveUrl;
+};
+const findFreePort = () => new Promise((resolve, reject) => {
+    const server = net.createServer();
+    server.once("error", reject);
+    server.listen(0, "127.0.0.1", () => {
+        const address = server.address();
+        const port = typeof address === "object" && address ? address.port : 0;
+        server.close(() => (port > 0 ? resolve(port) : reject(new Error("remotion: could not pick a port for the bundle server"))));
+    });
+});
+const logBlockedRequest = (url) => GraphAILogger.info(`remotion: blocked a network request from the scene: ${url}`);
+// Throwing here stops remotion before it fetches, and fails the render.
+const rejectMediaDownload = (src) => {
+    if (!isAllowedMediaDownload(src))
+        throw new Error(`remotion: a scene may not load media files (${src.slice(0, MAX_LOGGED_URL_LENGTH)})`);
+    return undefined;
+};
+// The bundle server listens on the port given here, the only local origin the guard lets the scene reach.
+const renderOutputs = async (renderer, browser, serveUrl, port, outputs) => {
+    const { props, videoPath, stillPath, extraStills = [] } = outputs;
+    const shared = { serveUrl, inputProps: props, chromiumOptions: CHROMIUM_OPTIONS, puppeteerInstance: browser, port };
+    const downloadGuard = { onDownload: rejectMediaDownload };
+    const composition = await renderer.selectComposition({ ...shared, id: REMOTION_COMPOSITION_ID });
     if (videoPath) {
-        await renderer.renderMedia({ composition, serveUrl, codec: "h264", outputLocation: videoPath, inputProps: props, chromiumOptions: CHROMIUM_OPTIONS });
+        await renderer.renderMedia({ ...shared, ...downloadGuard, composition, codec: "h264", outputLocation: videoPath });
     }
     const stills = stillPath ? [...extraStills, { frame: composition.durationInFrames - 1, path: stillPath }] : extraStills;
     await stills.reduce(async (previous, still) => {
         await previous;
-        await renderer.renderStill({ composition, serveUrl, output: still.path, frame: still.frame, inputProps: props, chromiumOptions: CHROMIUM_OPTIONS });
+        await renderer.renderStill({ ...shared, ...downloadGuard, composition, output: still.path, frame: still.frame });
     }, Promise.resolve());
 };
+export const renderRemotionScene = async ({ workDir, ...outputs }) => {
+    const { bundle, renderer } = await loadRemotion();
+    const serveUrl = await bundleScene(bundle, workDir);
+    const port = await findFreePort();
+    const browser = await renderer.openBrowser("chrome", { chromiumOptions: CHROMIUM_OPTIONS });
+    try {
+        const guard = await guardSceneNetwork(browser, port, logBlockedRequest);
+        await renderOutputs(renderer, browser, serveUrl, port, outputs);
+        if (guard.failures.length > 0)
+            throw new Error(`remotion: the network guard failed: ${guard.failures.join("; ")}`);
+    }
+    finally {
+        await browser.close({ silent: true });
+    }
+};

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mulmocast",
-  "version": "2.15.0",
+  "version": "2.16.0",
   "description": "",
   "type": "module",
   "main": "lib/index.node.js",

package/scripts/templates/html.json

@@ -13,7 +13,7 @@
   "title": "[TITLE: Brief, engaging title for the topic]",
   "htmlImageParams": {
     "provider": "anthropic",
-    "model": "claude-sonnet-4-5-20250929"
+    "model": "claude-sonnet-4-6"
   },
   "lang": "en",
   "beats": [

package/scripts/templates/presentation.json

@@ -5,7 +5,7 @@
   },
   "htmlImageParams": {
     "provider": "anthropic",
-    "model": "claude-sonnet-4-5-20250929"
+    "model": "claude-sonnet-4-6"
   },
   "title": "Sample Title",
   "references": [

package/scripts/test/test_genai.json

@@ -59,7 +59,7 @@
       "text": "movie generated by veo3",
       "moviePrompt": "a woman is walking through a busy Tokyo street at night, she is wearing dark sunglasses",
       "movieParams": {
-        "model": "veo-3.0-generate-001"
+        "model": "veo-3.1-lite-generate-preview"
       }
     },
     {
@@ -80,7 +80,7 @@
       },
       "moviePrompt": "a woman is walking through a busy Tokyo street at night, she is wearing dark sunglasses",
       "movieParams": {
-        "model": "veo-3.0-generate-001"
+        "model": "veo-3.1-lite-generate-preview"
       }
     }
   ]

package/scripts/test/test_gpt_image.json

@@ -7,7 +7,7 @@
   "title": "Testing OpenAI's new Image Generation",
   "description": "Hello",
   "imageParams": {
-    "model": "gpt-image-1",
+    "model": "gpt-image-2.5-sunburst",
     "style": "<style>Japanese animation with soft watercolor backgrounds, characters with simple rounded faces, large expressive eyes, small nose and mouth, soft jawlines, minimalist facial features, pastel color palette, detailed natural environments, whimsical magical elements, hand-drawn aesthetic, gentle lighting, flowing movement in hair and clothing, nostalgic countryside scenery with fantasy elements."
   },
   "speechParams": {

package/scripts/test/test_replicate.json

@@ -96,12 +96,12 @@
       }
     },
     {
-      "id": "kling-v1.6-pro",
-      "text": "kwaivgi/kling-v1.6-pro with gpt4-image",
+      "id": "kling-v2.1-start-image",
+      "text": "kwaivgi/kling-v2.1 with gpt4-image",
       "imagePrompt": "a woman is walking through a busy Tokyo street at night, she is wearing dark sunglasses",
       "moviePrompt": "a woman is walking through a busy Tokyo street at night, she is wearing dark sunglasses",
       "movieParams": {
-        "model": "kwaivgi/kling-v1.6-pro"
+        "model": "kwaivgi/kling-v2.1"
       }
     },
     {

package/scripts/test/test_vertexai.json

@@ -24,7 +24,7 @@
   },
   "movieParams": {
     "provider": "google",
-    "model": "veo-2.0-generate-001",
+    "model": "veo-3.1-generate-001",
     "vertexai_project": "YOUR_PROJECT_ID",
     "vertexai_location": "us-central1"
   },
@@ -76,7 +76,7 @@
       "text": "TTS: Gemini 2.5 Pro TTS Kore. Movie: Veo 3 override.",
       "moviePrompt": "Ocean waves crashing on a beach at sunset",
       "movieParams": {
-        "model": "veo-3.0-generate-001",
+        "model": "veo-3.1-generate-001",
         "vertexai_project": "YOUR_PROJECT_ID",
         "vertexai_location": "us-central1"
       }
@@ -101,7 +101,7 @@
       },
       "moviePrompt": "The robot turns its head and waves at the camera",
       "movieParams": {
-        "model": "veo-3.0-generate-001",
+        "model": "veo-3.1-generate-001",
         "vertexai_project": "YOUR_PROJECT_ID",
         "vertexai_location": "us-central1"
       },