mulmocast 2.14.0 → 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/assets/schemas/mulmo_script.json

@@ -125,7 +125,7 @@
                 "properties": {
                   "source": {
                     "type": "string",
-                    "description": "AvatarScript avatar package (a folder with avatar.json, or a mesh-avatar-studio project), relative to the script"
+                    "description": "AvatarScript avatar package (a folder with avatar.json, or a mesh-avatar-studio project): a path relative to the script, or an http(s) URL"
                   },
                   "position": {
                     "type": "object",
@@ -226,7 +226,7 @@
                       "properties": {
                         "source": {
                           "type": "string",
-                          "description": "AvatarScript avatar package (a folder with avatar.json, or a mesh-avatar-studio project), relative to the script"
+                          "description": "AvatarScript avatar package (a folder with avatar.json, or a mesh-avatar-studio project): a path relative to the script, or an http(s) URL"
                         },
                         "position": {
                           "type": "object",

package/lib/actions/avatar.d.ts

@@ -19,7 +19,7 @@ export type AvatarPlacementPlan = {
 };
 export type AvatarTrackPlan = {
     speakerId: string;
-    /** absolute path of the avatar package */
+    /** the avatar package: an absolute path, or an http(s) URL */
     source: string;
     segments: AvatarSegmentPlan[];
     /** where the avatar is shown, beat by beat; it is not shown outside these */
@@ -33,5 +33,6 @@ export type AvatarTrackPlan = {
  * avatarParams, then the defaults; during other speakers' beats the avatar stays where it was.
  */
 export declare const planAvatarTracks: (context: MulmoStudioContext) => AvatarTrackPlan[];
+export declare const avatarUrlStamp: (source: string) => Promise<(string | null)[]>;
 /** The `avatar` action: renders the avatar tracks into studio.avatarTracks (none when no speaker has an avatar). */
 export declare const avatar: (context: MulmoStudioContext) => Promise<MulmoStudioContext>;

package/lib/actions/avatar.js

@@ -7,7 +7,7 @@ import path from "node:path";
 import { GraphAILogger } from "graphai";
 import { MulmoPresentationStyleMethods, MulmoStudioContextMethods } from "../methods/index.js";
 import { getFullPath, getOutputStudioFilePath } from "../utils/file.js";
-import { localizedText } from "../utils/utils.js";
+import { isHttp, localizedText } from "../utils/utils.js";
 const DEFAULT_POSITION = { x: "84%", y: "100%", scale: "62%" };
 // Show the whole avatar image. A rig may crop the top of its image (a flat cut edge, hidden when the
 // avatar fills the frame); over a slide the cut would run through the head.
@@ -43,7 +43,7 @@ export const planAvatarTracks = (context) => {
         tracks.set(found.speakerId, {
             speakerId: found.speakerId,
             avatar: found.avatar,
-            source: getFullPath(context.fileDirs.mulmoFileDirPath, found.avatar.source),
+            source: isHttp(found.avatar.source) ? found.avatar.source : getFullPath(context.fileDirs.mulmoFileDirPath, found.avatar.source),
             segments: [],
             placements: [],
             duration,
@@ -105,6 +105,62 @@ const folderStamp = (dir) => {
         .sort()
         .map((file) => [path.relative(dir, file), fileStamp(file)]);
 };
+// An avatar at a URL: its small JSON files (avatar.json, the rig, layers.json, sprites.json) change
+// whenever the avatar is edited or rebuilt, so they stand in for the files' stamps. They are fetched
+// on every run; avatarscript caches the images.
+// Redirects are followed only within the same origin, so a public host cannot send these requests
+// to another one (an internal service, say). GitHub's github.com/…/raw/… links redirect to
+// raw.githubusercontent.com: use the latter.
+const MAX_REDIRECTS = 5;
+const fetchSameOrigin = async (url) => {
+    let location = url;
+    for (let hops = 0; hops <= MAX_REDIRECTS; hops++) {
+        const response = await fetch(location, { redirect: "manual", signal: AbortSignal.timeout(60_000) });
+        const next = response.status >= 300 && response.status < 400 ? response.headers.get("location") : null;
+        if (!next)
+            return response;
+        const target = new URL(next, location);
+        if (target.origin !== url.origin)
+            throw new Error(`avatar: ${url.href} redirects to another host (${target.origin}); use the final URL`);
+        location = target;
+    }
+    throw new Error(`avatar: ${url.href}: too many redirects`);
+};
+const fetchText = async (url) => {
+    const response = await fetchSameOrigin(url);
+    if (response.status === 404)
+        return null;
+    if (!response.ok)
+        throw new Error(`avatar: ${url.href}: HTTP ${response.status}`);
+    return response.text();
+};
+export const avatarUrlStamp = async (source) => {
+    const url = new URL(source);
+    const folder = new URL(url);
+    folder.pathname = url.pathname.endsWith("/avatar.json") ? url.pathname.slice(0, -"avatar.json".length) : url.pathname.replace(/\/?$/, "/");
+    // the source's query (a version, a signature) goes with every file that has no query of its own
+    // like avatarscript, fetch only files inside the package's folder (no "../", no other host)
+    const at = (file, base) => {
+        const location = new URL(file, base);
+        if (location.origin !== folder.origin || !location.pathname.startsWith(folder.pathname)) {
+            throw new Error(`avatar: ${file} is outside the avatar package ${folder.href}`);
+        }
+        if (!location.search)
+            location.search = url.search;
+        return location;
+    };
+    const manifestText = await fetchText(at("avatar.json", folder));
+    const manifest = manifestText ? JSON.parse(manifestText) : undefined;
+    const root = at((manifest?.root ?? ".").replace(/\/?$/, "/"), folder);
+    // as avatarscript: a manifest names its files (sprites are optional); without one, a
+    // mesh-avatar-studio project's default paths
+    const assets = manifest ? (manifest.assets ?? {}) : { rig: "rig.json", layers: "built/layers.json", sprites: "built/sprites/sprites.json" };
+    const files = [assets.rig, assets.layers, assets.sprites];
+    // every location is checked before anything is fetched
+    const locations = files.map((file) => (file === undefined ? undefined : at(file, root)));
+    return [manifestText, ...(await Promise.all(locations.map((location) => (location ? fetchText(location) : null))))];
+};
+const sourceStamp = (source) => (isHttp(source) ? avatarUrlStamp(source) : folderStamp(source));
 const even = (n) => Math.max(2, Math.round(n / 2) * 2); // video codecs need even sizes
 /** Renders one track (or reuses it), returning where it goes on the canvas. */
 const renderAvatarTrack = async (plan, context, getAvatarScript) => {
@@ -112,7 +168,7 @@ const renderAvatarTrack = async (plan, context, getAvatarScript) => {
     // rendered once, at the largest size it is shown; smaller placements scale it down
     const height = even(Math.max(...plan.placements.map((p) => percent(p.position.scale, canvas.height))));
     const segments = plan.segments.map((segment) => ({ ...segment, audioStamp: fileStamp(segment.audio) }));
-    const identity = { version: 4, segments, source: plan.source, sourceStamp: folderStamp(plan.source), duration: plan.duration, height, padTop: PAD_TOP };
+    const identity = { version: 4, segments, source: plan.source, sourceStamp: await sourceStamp(plan.source), duration: plan.duration, height, padTop: PAD_TOP };
     const hash = createHash("sha256").update(JSON.stringify(identity)).digest("hex").slice(0, 12);
     const dir = MulmoStudioContextMethods.getImageProjectDirPath(context);
     const name = `avatar_${plan.speakerId.replace(/[^\w-]/g, "_")}_${hash}`;

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/index.remotion.d.ts

@@ -1,3 +1,6 @@
 export { REMOTION_COMPONENT_GUIDE } from "./utils/remotion/system_prompt.js";
 export { REMOTION_PACKAGES, REMOTION_RENDER_PACKAGES, REMOTION_SCENE_PACKAGES, remotionInstallCommand } from "./utils/remotion/packages.js";
 export { ensureRemotionPackages, missingInstalledRemotionPackages } from "./utils/remotion/render.js";
+export { REMOTION_REVIEW_FRACTIONS } from "./utils/remotion/frames.js";
+export { renderRemotionFrames } from "./utils/remotion/render_frames.js";
+export type { RemotionFrame, RemotionFramesRequest } from "./utils/remotion/render_frames.js";

package/lib/index.remotion.js

@@ -2,3 +2,5 @@
 export { REMOTION_COMPONENT_GUIDE } from "./utils/remotion/system_prompt.js";
 export { REMOTION_PACKAGES, REMOTION_RENDER_PACKAGES, REMOTION_SCENE_PACKAGES, remotionInstallCommand } from "./utils/remotion/packages.js";
 export { ensureRemotionPackages, missingInstalledRemotionPackages } from "./utils/remotion/render.js";
+export { REMOTION_REVIEW_FRACTIONS } from "./utils/remotion/frames.js";
+export { renderRemotionFrames } from "./utils/remotion/render_frames.js";

package/lib/index.remotion_guide.d.ts

@@ -0,0 +1 @@
+export { REMOTION_COMPONENT_GUIDE } from "./utils/remotion/system_prompt.js";

package/lib/index.remotion_guide.js

@@ -0,0 +1,3 @@
+// Entry point for code that only needs the remotion component guide — a tool description built into
+// a browser bundle, for example. `mulmocast/remotion` also carries the renderer, which needs Node.
+export { REMOTION_COMPONENT_GUIDE } from "./utils/remotion/system_prompt.js";

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";