@vanillaskyai/video 0.10.0 → 0.10.2

package/CHANGELOG.md

@@ -4,6 +4,19 @@ VanillaSky follows semantic versioning. This changelog begins with the 0.1 beta.
 
 ## Unreleased
 
+## 0.10.2
+
+- Keep generated narration playing in Safari after a delayed opening-to-body handoff by reusing the permitted audio element; isolate late playback failures from subsequent lines.
+- In drop mode, a media miss without usable authored fallback skips only that scene and preserves later valid scenes and the ending. Strict mode still fails. The skipped beat is reported; its content cannot be reconstructed without a usable asset or fallback.
+- Guide first-shot narration to fit its five-second footage budget and develop the answer in later scenes, without imposing a fixed sentence length on the rest of the video.
+
+## 0.10.1
+
+- Continue narration through failed built-in media using the authored fallback or the existing graphic on black. Keep the recovery anchor after URL resolution, and observe actual mounted images for photo readiness.
+
+- Match planning, narration, visual progression and endings to explanations, fiction, comedy, imaginative requests and practical answers. Remove conflicting instructions that forced a factual arc, fixed sentence length or a different template for every beat.
+- Ask follow-up generation to exclude questions already answered by the full narration, including paraphrases, and to preserve the conversation’s intent.
+
 ## 0.10.0
 
 ### Breaking

package/dist/check-runtime.js

@@ -1,6 +1,6 @@
 import {
   VideoFrame
-} from "./chunk-ELVOT2FQ.js";
+} from "./chunk-6KZGF63O.js";
 import "./chunk-224QNWRA.js";
 import "./chunk-TVIU23OM.js";
 import "./chunk-SPVTJH3F.js";

package/dist/{chunk-ZDJNFA52.js → chunk-5BSJLT6H.js}

@@ -114,15 +114,15 @@ function buildVideoUserPrompt(input, openingDurationSec = 0) {
       "Add that scene before resolving any stock or supplied asset. Resolve media as part of each complete later body scene."
     ],
     "Use only claims supported by the factual basis permitted by the trusted system prompt.",
-    "Select the most decision-relevant grounded takeaways that fit the duration; represent each selected takeaway once before completing the response.",
+    "Develop the requested answer within the duration: preserve its mechanism, essential actions, story resolution or punchline. A short prompt can need several distinct visual beats; input length is not the answer length.",
     "For a long source, summarize instead of attempting to represent every fact, unless the creative instructions explicitly request complete fact coverage that fits the duration.",
     "When selecting claims from the raw input, preserve their exact wording and numbers. Preserve qualifiers, units, denominators, ranges, and comparison direction; for example, do not shorten 4.8 out of 5 to 4.8.",
-    "Choose the scene count from the distinct grounded material and the duration budget. Use fewer scenes for sparse input; continue beyond five when rich input warrants it and timing allows.",
+    "Choose the scene count from the requested intent, visual progression and duration budget. A joke can be brief; a practical sequence or imaginative story may need more development. Do not stop while an essential step or payoff is missing.",
     "If the creative instructions explicitly require one separate scene per named item, release, section, or list entry, do not merge, group, or omit those required items. Keep related required scenes adjacent in a coherent progression while preserving each item as its own scene.",
     "Before emitting, verify that the explicitly requested structure can fit readably within the maximum duration. If it cannot, preserve readability and the requested separation for the scenes that fit, then finish with plan.complete using finishReason length rather than silently changing the structure.",
-    "For ordinary multi-fact input, form at least three distinct beats: hook, comprehension, and payoff. Use only one or two beats when the source genuinely contains no more than two independent grounded takeaways. This is a narrative guideline, never permission to repeat facts or add filler.",
-    "Emit exactly one final scene.add with placement closer. Generate a short grounded conclusion that answers the story's so-what; it must not repeat hook language or imply another scene follows.",
-    "Use a different suitable template for each body scene when the catalog supports it.",
+    "Before plan.complete, check that the answer fulfills the actual request. Do not mistake a hook and a single supporting statement for a complete explanation, or explain a joke after its punchline.",
+    "Emit exactly one final scene.add with placement closer. Land the requested ending without repeating the opening or implying another scene follows.",
+    "Reuse the same suitable template when it serves successive beats. Vary the subject action, shot scale, perspective or purposeful detail, not the template merely for variety.",
     "Never add filler to satisfy a count or diversity target.",
     input.suppliedMedia?.length ? input.opening === false ? "Select zero or more relevant opaque supplied-media references for visible scenes. Never invent or transform a reference. Only emit mediaKeyword when the trusted system catalog explicitly exposes it. Never invent mediaUrl or mediaPoster." : "Select zero or more relevant opaque supplied-media references for visible later scenes. Never invent or transform a reference. Only emit mediaKeyword when the trusted system catalog explicitly exposes it, and only on later scenes. Never invent mediaUrl or mediaPoster." : input.opening === false ? "No supplied media URL is available. Use asset-free templates unless the trusted system catalog explicitly permits host-resolved media intent, which may begin on the first scene. Only emit mediaKeyword when the trusted system catalog explicitly exposes it. Never invent mediaUrl or mediaPoster." : "No supplied media URL is available. Use asset-free templates unless the trusted system catalog explicitly permits host-resolved media intent. Only emit mediaKeyword when the trusted system catalog explicitly exposes it, and only on later scenes. Never invent mediaUrl or mediaPoster.",
     "",
@@ -691,7 +691,7 @@ function createVideo(rawInput, options) {
           position
         }));
       }
-      const systemPrompt = options.systemPrompt ?? (await import("./system-prompt-SYRT5DWM.js")).createVideoSystemPrompt(input.knowledgeMode);
+      const systemPrompt = options.systemPrompt ?? (await import("./system-prompt-AG26KJAA.js")).createVideoSystemPrompt(input.knowledgeMode);
       const context = { request, systemPrompt, userPrompt, initialConfig, signal: controller.signal };
       attachGenerationLifecycleSink(context, lifecycle);
       for await (const untrustedPart of options.generate(context)) {

package/dist/{chunk-NMFDC2W2.js → chunk-5C6HZNHY.js}

@@ -163,17 +163,22 @@ var SceneBackground = ({
         onError: () => setMediaPaint("failed")
       }
     ) : /* @__PURE__ */ jsx(
-      "div",
+      "img",
       {
+        src: mediaUrl,
+        alt: "",
+        "aria-hidden": "true",
+        draggable: false,
         "data-media-position": mediaPosition,
         style: {
           position: "absolute",
           inset: 0,
           transform: bgTransform.transform,
           transformOrigin: bgTransform.transformOrigin,
-          backgroundImage: `url(${mediaUrl})`,
-          backgroundSize: "cover",
-          backgroundPosition: resolvedPosition
+          width: "100%",
+          height: "100%",
+          objectFit: "cover",
+          objectPosition: resolvedPosition
         }
       }
     )),

package/dist/{chunk-ELVOT2FQ.js → chunk-6KZGF63O.js}

@@ -21,10 +21,12 @@ import { createContext, useContext, useEffect, useRef } from "react";
 import { jsx } from "react/jsx-runtime";
 var MountedReadinessContext = createContext(void 0);
 var sceneReadinessKey = (scene) => `${scene.id}\0${String(scene.variables.mediaUrl || "")}`;
-function MountedSceneReadiness({ scene, playing, fallback = false }) {
+function MountedSceneReadiness({ scene, playing, fallback = false, onFailure }) {
   const marker = useRef(null);
   const report = useContext(MountedReadinessContext);
   const key = sceneReadinessKey(scene);
+  const onFailureRef = useRef(onFailure);
+  onFailureRef.current = onFailure;
   useEffect(() => {
     if (!report || !playing) return;
     let stopped = false;
@@ -35,13 +37,14 @@ function MountedSceneReadiness({ scene, playing, fallback = false }) {
     const finish = (error, actualVideoFrame = false) => {
       if (!stopped) {
         stopped = true;
-        report(key, error, actualVideoFrame);
+        if (error && onFailureRef.current) onFailureRef.current();
+        else report(key, error, actualVideoFrame);
       }
     };
     const check = () => {
       if (stopped) return;
       const root = marker.current?.closest("[data-video-frame]");
-      if (fallback && root?.querySelector("[data-scene-fallback]")) {
+      if (fallback && root?.querySelector("[data-scene-fallback]") && !root.querySelector("[data-template-loading]") && document.fonts?.status !== "loading") {
         finish();
         return;
       }
@@ -132,6 +135,7 @@ import {
   lazy,
   Suspense,
   useState,
+  useCallback,
   useEffect as useEffect2,
   useSyncExternalStore
 } from "react";
@@ -269,7 +273,9 @@ function SceneLayer({
   zIndex,
   externalVideoBackdrop
 }) {
-  const template = kit.getTemplate(range.scene.templateId);
+  const mediaFailed = externalVideoBackdrop === "fallback";
+  const recoveryTitle = mediaFailed && range.scene.templateId === "cinemaMedia" && typeof range.scene.variables.fallbackText === "string" ? range.scene.variables.fallbackText : void 0;
+  const template = kit.getTemplate(recoveryTitle ? "chapterTitle" : range.scene.templateId);
   const duration = range.end - range.start;
   return /* @__PURE__ */ jsx2(
     ExternalVideoBackdropProvider,
@@ -281,6 +287,7 @@ function SceneLayer({
         "div",
         {
           "data-scene-layer": layer,
+          "data-scene-fallback": mediaFailed ? "true" : void 0,
           "data-layer-scene-id": range.scene.id,
           "data-layer-template-id": range.scene.templateId,
           "aria-hidden": interactive ? void 0 : true,
@@ -308,7 +315,9 @@ function SceneLayer({
             createElement(template.component, {
               variables: {
                 ...template.defaults ?? getTemplateDefaults(template.schema),
-                ...range.scene.variables
+                ...range.scene.variables,
+                ...mediaFailed ? { mediaUrl: "", mediaPoster: "", mediaType: "gradient" } : {},
+                ...recoveryTitle ? { title: recoveryTitle } : {}
               },
               style: config.style,
               progress,
@@ -344,7 +353,18 @@ function VideoFrame({
   className,
   style
 }) {
-  const [failedPersistentVideo, setFailedPersistentVideo] = useState();
+  const [failedMedia, setFailedMedia] = useState(() => /* @__PURE__ */ new Set());
+  const markMediaFailed = useCallback((key) => {
+    if (!key || !config.scenes.some((scene) => sceneReadinessKey(scene) === key)) return;
+    setFailedMedia((previous) => previous.has(key) ? previous : /* @__PURE__ */ new Set([...previous, key]));
+  }, [config.scenes]);
+  useEffect2(() => {
+    const currentKeys = new Set(config.scenes.map(sceneReadinessKey));
+    setFailedMedia((previous) => {
+      const retained = new Set([...previous].filter((key) => currentKeys.has(key)));
+      return retained.size === previous.size ? previous : retained;
+    });
+  }, [config.scenes]);
   const [readyPersistentVideo, setReadyPersistentVideo] = useState();
   const decoderConstrainedDevice = useDecoderConstraint();
   const timeline = resolveVideoTimeline(config);
@@ -422,7 +442,8 @@ function VideoFrame({
     mediaPosition: String(posterPreparationRange.scene.variables.mediaPosition || "center"),
     backgroundEffect: posterPreparationRange.scene.backgroundEffect ?? config.style.defaultBackgroundEffect
   } : void 0;
-  const persistentVideoFailed = persistentVideoKey !== void 0 && failedPersistentVideo === persistentVideoKey;
+  const activeMediaFailed = failedMedia.has(sceneReadinessKey(active.scene));
+  const persistentVideoFailed = persistentVideoKey !== void 0 && failedMedia.has(persistentVideoKey);
   const persistentVideoReady = persistentVideoRange !== void 0 && (String(persistentVideoRange.scene.variables.mediaPoster || "") !== "" || readyPersistentVideo === persistentVideoKey);
   const persistentVideoMode = persistentVideoFailed ? "fallback" : persistentVideoReady ? "ready" : "pending";
   const mountingNext = Boolean(
@@ -442,6 +463,12 @@ function VideoFrame({
   return /* @__PURE__ */ jsxs(
     "div",
     {
+      onErrorCapture: (event) => {
+        const target = event.target;
+        if ((target instanceof HTMLVideoElement || target instanceof HTMLImageElement) && target.getAttribute("src") === active.scene.variables.mediaUrl && supportsExternalVideoBackdrop(activeTemplate) && (target.closest("[data-layer-scene-id]")?.getAttribute("data-layer-scene-id") === active.scene.id || target.closest("[data-persistent-video-scene-id]")?.getAttribute("data-persistent-video-scene-id") === active.scene.id)) {
+          markMediaFailed(sceneReadinessKey(active.scene));
+        }
+      },
       "data-video-frame": "ready",
       "data-scene-id": active.scene.id,
       "data-template-id": active.scene.templateId,
@@ -455,7 +482,15 @@ function VideoFrame({
         ...style
       },
       children: [
-        /* @__PURE__ */ jsx2(MountedSceneReadiness, { scene: active.scene, playing }),
+        /* @__PURE__ */ jsx2(
+          MountedSceneReadiness,
+          {
+            scene: active.scene,
+            playing,
+            fallback: activeMediaFailed,
+            onFailure: sceneHasBackdrop(active) && supportsExternalVideoBackdrop(activeTemplate) && !activeMediaFailed ? () => markMediaFailed(sceneReadinessKey(active.scene)) : void 0
+          }
+        ),
         preparedReadinessRange && /* @__PURE__ */ jsx2(PreparedSceneReadiness, { scene: preparedReadinessRange.scene }),
         /* @__PURE__ */ jsxs(
           "div",
@@ -530,7 +565,7 @@ function VideoFrame({
                         opacity: decoderConstrainedTransition ? blendProgress : 0
                       } : void 0,
                       onReady: () => setReadyPersistentVideo(persistentVideoKey),
-                      onError: () => setFailedPersistentVideo(persistentVideoKey)
+                      onError: () => markMediaFailed(persistentVideoKey)
                     }
                   ) })
                 }
@@ -554,7 +589,7 @@ function VideoFrame({
                     opacity: 1 - blendProgress,
                     interactive: true,
                     zIndex: 1,
-                    externalVideoBackdrop: persistentVideoRange?.scene.id === active.scene.id ? persistentVideoMode : false
+                    externalVideoBackdrop: activeMediaFailed ? "fallback" : persistentVideoRange?.scene.id === active.scene.id ? persistentVideoMode : false
                   },
                   active.scene.id
                 ),
@@ -598,7 +633,7 @@ function VideoFrame({
                     opacity: 1,
                     interactive: true,
                     zIndex: 1,
-                    externalVideoBackdrop: persistentVideoRange?.scene.id === active.scene.id ? persistentVideoMode : false
+                    externalVideoBackdrop: activeMediaFailed ? "fallback" : persistentVideoRange?.scene.id === active.scene.id ? persistentVideoMode : false
                   },
                   active.scene.id
                 )

package/dist/{chunk-2HFO22K2.js → chunk-JKVOBTRO.js}

@@ -1,6 +1,6 @@
 import {
   createVideoSystemPrompt
-} from "./chunk-PAJ44TFW.js";
+} from "./chunk-LVM5Q2DL.js";
 import {
   PACING_PLANNER_RULES
 } from "./chunk-3O7OMMMF.js";
@@ -118,7 +118,7 @@ ${options.basePrompt.trim()}` : void 0,
     "Variable notation is type[count]{characters}(options)! where count is list cardinality, characters is the inclusive character count for a string or each string-array item, and ! means required. Omitted ! means optional.",
     "Choose a template only when the permitted factual basis contains every fact it needs. Never invent peer values to complete a chart, comparison, stat set, timeline, or list.",
     ...PACING_PLANNER_RULES,
-    "Let narration carry ordinary lists. Use a structured overlay only when a comparison, sequence, exact quote or key figure adds understanding beyond the voice. Prefer relevant full-bleed footage for the rest of the explanation.",
+    "Let narration carry ordinary lists. Use a structured overlay only when a comparison, sequence, exact quote or key figure adds understanding beyond the voice. Prefer relevant full-bleed footage for the remaining visual beats, with varied purposeful action and framing.",
     "If the catalog includes a suitable ask template and the input supplies a grounded CTA or URL, keep that concise action closer as its own final scene instead of folding it into preceding content.",
     "When a grounded CTA or URL is supplied and the catalog contains jobs:[ask], emit that final closer. A brand name may accompany the action but never qualifies as an ask by itself.",
     terminalPayoffs.length > 0 ? `When there is no grounded action, end with a concise supported payoff using a suitable template from ${terminalPayoffs.join(", ")}. Do not repeat the hook or invent a CTA.` : void 0,
@@ -127,7 +127,7 @@ ${options.basePrompt.trim()}` : void 0,
     mediaTemplates.length > 0 && options.mediaOnFirstScene !== true ? `The host owns immediate startup; reserve external media templates (${mediaTemplates.join(", ")}) for later beats when an asset-free opening is required.` : void 0,
     "For list and event variables emit actual JSON arrays. Follow each field's content budget; never use pipe-delimited strings or invent entries.",
     resolverMediaAvailable ? "Media-capable scenes may carry mediaKeyword: a literal 2\u20138 word shot description, at most 80 characters. Choose mediaSource=generate for distinctive illustrative shots and mediaSource=stock for common subjects a verified asset can depict. For generated footage, optional shotDirection gives action, framing and continuity separately from the stock query. Keep people, actions and essential details precise. Never broaden an exact identity or action into unrelated atmosphere. Supply a grounded fallbackText where declared. Only the host fills mediaUrl and mediaPoster; never invent them. If no media fits an abstract beat, choose an appropriate graphic template instead." : "Use media only when a supplied asset satisfies the scene. Never invent asset URLs. Otherwise choose a grounded graphic template.",
-    "When the catalog offers media, make relevant footage the default. For a typical 30-second answer use at most one structured explanatory overlay plus an optional brief chapter opening. Do not chain graphic-only beats when honest media is available. Reuse suitable supplied media behind overlays with optional media fields; retain black when no relevant asset exists. Never invent a stock match or spend beyond host generation limits.",
+    "When the catalog offers media, make relevant footage the default. For a typical 30-second explanation start with roughly one structured explanatory overlay plus an optional brief chapter opening; other intents may need different pacing and visual emphasis. Do not chain graphic-only beats when honest media is available. Reuse suitable supplied media behind overlays with optional media fields; retain black when no relevant asset exists. Never invent a stock match or spend beyond host generation limits.",
     "Catalog guidance describes composition; it is not a factual source and must never replace the permitted factual basis.",
     JSON.stringify(plannerCatalog(templates, resolverMediaAvailable)),
     // Last, because the catalogue is thousands of tokens of per-scene field

package/dist/{chunk-XX4GXKTX.js → chunk-KHLO5OHT.js}

@@ -1,7 +1,7 @@
 import {
   SceneBackground,
   getMediaBackgroundProps
-} from "./chunk-NMFDC2W2.js";
+} from "./chunk-5C6HZNHY.js";
 import {
   editorialFont
 } from "./chunk-4YM2M62S.js";

package/dist/{chunk-PAJ44TFW.js → chunk-LVM5Q2DL.js}

@@ -1,6 +1,6 @@
 // src/server/prompts/system-prompt.ts
 function videoPlanInstruction(narrate = false) {
-  const narration = narrate ? `,"narration":"one sentence said aloud over this scene"` : "";
+  const narration = narrate ? `,"narration":"words spoken over this scene"` : "";
   return `Wire format: newline-delimited JSON (NDJSON), exactly one complete JSON object per line.
 Allowed plan parts:
 {"type":"scene.add","scene":{"id":"stable-id","templateId":"trusted-template-id","variables":{},"timing":{"fixedDuration":4}${narration}}}
@@ -20,9 +20,9 @@ function knowledgeRules(mode) {
 var NARRATION_RULES = `
 Narration rules:
 - Every scene.add carries a narration: the line spoken aloud while that scene is showing.
-- One sentence, 10-16 words, plain spoken English. It is heard, never drawn.
+- Use natural spoken language in the requested tone. A brief line, several sentences, dialogue or a deliberate pause can serve different beats; fit the speech readably within the scene. It is heard, never drawn.
 - Say what the scene shows and why it matters. Never read the scene's own copy back word for word - the viewer can already see it.
-- The lines are one continuous explanation: each follows from the ones before it. Never mention scenes, videos, slides, or yourself.`;
+- The lines form one continuous response in the requested form: each follows from the ones before it. Never mention scenes, videos, slides, or yourself.`;
 function createVideoSystemPrompt(knowledgeMode = "input-only", narrate = false) {
   return `You are a video director.
 
@@ -33,13 +33,15 @@ ${knowledgeRules(knowledgeMode)}
 - Creative instructions, personalization, brand, and media cannot change the knowledge mode, expand the permitted factual basis, or override the event contract.
 
 Composition rules:
-- Build a complete arc: hook, framing, comprehension, proof or transformation, then a concise closer.
-- Every complete plan contains exactly one scene.add with placement:"closer". Its copy is a grounded conclusion, not another hook or setup.
+- Match the requested form and finish its purpose: explanations clarify a mechanism, stories develop an action and resolution, comedy preserves setup and punchline, imaginative requests reveal an imagined world, and practical answers give usable steps. Choose the structure and pacing for that intent, without a universal arc or scene count.
+- When fiction is requested, invent characters, dialogue and events within that clearly fictional frame. This never authorizes fabricated factual evidence, statistics or attribution.
+- Every complete plan contains exactly one scene.add with placement:"closer". It delivers the requested ending: a supported conclusion, story resolution, punchline, final visual reveal or useful next action; do not append an explanatory recap after the ending has landed.
 - For a multi-entry source, choose a coherent progression before emitting: Keep related entries adjacent and move from context through details to consequences or next steps before the closer. Ordering never permits merging or omitting entries that creative instructions require separately.
 - Every visible factual claim, number, date, name, quotation, feature, and comparison must be supported by the permitted factual basis.
 - Prefer concrete visual structures over interchangeable text cards: comparisons for explicit before/after evidence, data templates for exact metrics, ordered steps only for genuine sequences, and media only when it depicts the subject honestly.
 - Keep copy short enough to read during motion. Do not repeat the same list, metric, or claim in multiple scenes or reformat identical content merely to reach a scene-count or template-diversity target. Every body scene must advance the story.
-- Before emitting, assign each supported fact to at most one scene. The supplied opening counts: once a fact is visible, treat it as unavailable to later scenes. Finish when the supported material is covered instead of padding the response.
+- Before emitting, track what the opening and earlier beats have already conveyed, including paraphrases. Each next beat adds a consequence, action, detail or new understanding. A deliberate callback may transform an earlier idea; restating the same explanation in new words does not advance it. Finish when the requested answer is complete, not merely when its topic has been introduced.
+- For scientific explanations distinguish the physical mechanism from an analogy. Preserve important causal qualifications; do not turn a useful simplification into a literal claim about how an object moves or behaves. Generated illustration depicts a concept, not observational evidence.
 - Do not infer that something is scheduled, ready, triggered, enabled, automatic, causal, or available unless the permitted factual basis supports it.
 - The first generated body scene must be asset-free and fully playable before any external media resolves.
 - Use only media URLs present in the supplied input or already resolved by the host. Never expose a loading placeholder or unresolved media keyword. Audio is optional and must never delay the first scene.

package/dist/{chunk-Z67Q5KUW.js → chunk-VQH3JTQC.js}

@@ -1,7 +1,7 @@
 import {
   SceneBackground,
   getMediaBackgroundProps
-} from "./chunk-NMFDC2W2.js";
+} from "./chunk-5C6HZNHY.js";
 
 // src/visual-system/scene-templates/cinema-media.tsx
 import { jsx } from "react/jsx-runtime";

package/dist/{cinema-media-C7OUPNSH.js → cinema-media-APDJS7SC.js}

@@ -1,8 +1,8 @@
 import {
   MediaScene,
   MediaSceneTemplate
-} from "./chunk-Z67Q5KUW.js";
-import "./chunk-NMFDC2W2.js";
+} from "./chunk-VQH3JTQC.js";
+import "./chunk-5C6HZNHY.js";
 import "./chunk-77VY4O7A.js";
 import "./chunk-224QNWRA.js";
 import "./chunk-TVIU23OM.js";

package/dist/cli.js

@@ -1,6 +1,6 @@
 #!/usr/bin/env node
-import "./chunk-2HFO22K2.js";
-import "./chunk-PAJ44TFW.js";
+import "./chunk-JKVOBTRO.js";
+import "./chunk-LVM5Q2DL.js";
 import {
   isTemplatePropertyFormatSupported,
   validateTemplateSchemaValue

package/dist/{comparison-GNZOR7MF.js → comparison-ZVE2DE7U.js}

@@ -1,7 +1,7 @@
 import {
   EditorialSurface
-} from "./chunk-XX4GXKTX.js";
-import "./chunk-NMFDC2W2.js";
+} from "./chunk-KHLO5OHT.js";
+import "./chunk-5C6HZNHY.js";
 import "./chunk-77VY4O7A.js";
 import {
   editorialLabel,

package/dist/{compose-video-TBFTONUC.js → compose-video-BH5K6XWT.js}

@@ -1,6 +1,6 @@
 import {
   createVideo
-} from "./chunk-ZDJNFA52.js";
+} from "./chunk-5BSJLT6H.js";
 import "./chunk-E7CL7UPB.js";
 import "./chunk-RXTN2CW6.js";
 import "./chunk-7M56IUUX.js";

package/dist/{editorial-timeline-EMV5335S.js → editorial-timeline-YIXXVJUV.js}

@@ -1,7 +1,7 @@
 import {
   EditorialSurface
-} from "./chunk-XX4GXKTX.js";
-import "./chunk-NMFDC2W2.js";
+} from "./chunk-KHLO5OHT.js";
+import "./chunk-5C6HZNHY.js";
 import "./chunk-77VY4O7A.js";
 import {
   fade

package/dist/{key-figure-P37VK425.js → key-figure-I2C37FWV.js}

@@ -1,7 +1,7 @@
 import {
   EditorialSurface
-} from "./chunk-XX4GXKTX.js";
-import "./chunk-NMFDC2W2.js";
+} from "./chunk-KHLO5OHT.js";
+import "./chunk-5C6HZNHY.js";
 import "./chunk-77VY4O7A.js";
 import {
   editorialLabel,

package/dist/{mobile-message-OF7T62B7.js → mobile-message-JSGBU6EX.js}

@@ -1,7 +1,7 @@
 import {
   MediaScene
-} from "./chunk-Z67Q5KUW.js";
-import "./chunk-NMFDC2W2.js";
+} from "./chunk-VQH3JTQC.js";
+import "./chunk-5C6HZNHY.js";
 import {
   spring
 } from "./chunk-77VY4O7A.js";

package/dist/{quote-T6WYLN7A.js → quote-XH54G3FS.js}

@@ -1,7 +1,7 @@
 import {
   EditorialSurface
-} from "./chunk-XX4GXKTX.js";
-import "./chunk-NMFDC2W2.js";
+} from "./chunk-KHLO5OHT.js";
+import "./chunk-5C6HZNHY.js";
 import "./chunk-77VY4O7A.js";
 import {
   fade

package/dist/react.js

@@ -40,7 +40,7 @@ import {
   VideoFrame,
   getDimensions,
   sceneReadinessKey
-} from "./chunk-ELVOT2FQ.js";
+} from "./chunk-6KZGF63O.js";
 import "./chunk-224QNWRA.js";
 import "./chunk-TVIU23OM.js";
 import {
@@ -48,7 +48,8 @@ import {
   resolveVideoTimeline
 } from "./chunk-SPVTJH3F.js";
 import {
-  markExternalVideoBackdropTemplate
+  markExternalVideoBackdropTemplate,
+  supportsExternalVideoBackdrop
 } from "./chunk-R3XAOMKP.js";
 import "./chunk-73NTSFFI.js";
 
@@ -93,13 +94,13 @@ import { createElement } from "react";
 
 // src/visual-system/catalog/builtin-loaders.generated.ts
 var GENERATED_BUILTIN_TEMPLATE_LOADERS = {
-  "cinemaMedia": () => import("./cinema-media-C7OUPNSH.js").then((module) => ({ default: module.MediaSceneTemplate })),
+  "cinemaMedia": () => import("./cinema-media-APDJS7SC.js").then((module) => ({ default: module.MediaSceneTemplate })),
   "chapterTitle": () => import("./chapter-title-YRLIQY3V.js").then((module) => ({ default: module.TitleSceneTemplate })),
-  "editorialTimeline": () => import("./editorial-timeline-EMV5335S.js").then((module) => ({ default: module.TimelineSceneTemplate })),
-  "mobileMessage": () => import("./mobile-message-OF7T62B7.js").then((module) => ({ default: module.NotificationSceneTemplate })),
-  "comparison": () => import("./comparison-GNZOR7MF.js").then((module) => ({ default: module.ComparisonSceneTemplate })),
-  "quote": () => import("./quote-T6WYLN7A.js").then((module) => ({ default: module.QuoteSceneTemplate })),
-  "keyFigure": () => import("./key-figure-P37VK425.js").then((module) => ({ default: module.KeyFigureSceneTemplate }))
+  "editorialTimeline": () => import("./editorial-timeline-YIXXVJUV.js").then((module) => ({ default: module.TimelineSceneTemplate })),
+  "mobileMessage": () => import("./mobile-message-JSGBU6EX.js").then((module) => ({ default: module.NotificationSceneTemplate })),
+  "comparison": () => import("./comparison-ZVE2DE7U.js").then((module) => ({ default: module.ComparisonSceneTemplate })),
+  "quote": () => import("./quote-XH54G3FS.js").then((module) => ({ default: module.QuoteSceneTemplate })),
+  "keyFigure": () => import("./key-figure-I2C37FWV.js").then((module) => ({ default: module.KeyFigureSceneTemplate }))
 };
 
 // src/visual-system/catalog/builtin-player.generated.ts
@@ -1238,6 +1239,19 @@ var VideoError = class extends Error {
   }
 };
 
+// src/player/recover-scene-media.ts
+function recoverSceneMedia(scene) {
+  if (scene.templateId === "cinemaMedia") {
+    const title = scene.variables.fallbackText;
+    if (typeof title !== "string" || !title.trim() || [...title].length > 65) return void 0;
+    return { ...scene, templateId: "chapterTitle", variables: { title: title.trim() } };
+  }
+  if (!["comparison", "editorialTimeline", "quote", "keyFigure"].includes(scene.templateId)) return void 0;
+  const variables = { ...scene.variables };
+  for (const key of ["mediaUrl", "mediaPoster", "mediaType", "mediaKeyword", "mediaSource"]) delete variables[key];
+  return { ...scene, variables };
+}
+
 // src/video-chat/use-video-chat.ts
 import { useCallback as useCallback2, useEffect as useEffect4, useReducer, useRef as useRef3 } from "react";
 
@@ -1536,6 +1550,8 @@ function createVideoChatVoice(options = {}) {
   const lines = /* @__PURE__ */ new Map();
   const pendingLoads = /* @__PURE__ */ new Set();
   let sounding;
+  let generatedElement;
+  let stopGenerated;
   let browserFinish;
   let held = false;
   let silent = false;
@@ -1660,7 +1676,12 @@ function createVideoChatVoice(options = {}) {
     },
     resume() {
       held = false;
-      if (sounding) void sounding.play().catch(() => playbackFailure?.());
+      if (sounding) {
+        const fail = playbackFailure;
+        void sounding.play().catch(() => {
+          if (!held) fail?.();
+        });
+      }
       if (!silent) globalThis.speechSynthesis?.resume();
     },
     setMuted(muted) {
@@ -1739,9 +1760,11 @@ function createVideoChatVoice(options = {}) {
       }
       let playbackFailed = false;
       try {
-        const element = new Audio(line.src);
+        stopGenerated?.();
+        const element = generatedElement ??= new Audio();
+        element.src = line.src;
         element.muted = silent;
-        if (offsetSeconds !== void 0) element.currentTime = offsetSeconds;
+        element.currentTime = offsetSeconds ?? 0;
         sounding = element;
         await new Promise((resolve) => {
           let finished = false;
@@ -1757,6 +1780,7 @@ function createVideoChatVoice(options = {}) {
             element.onplaying = null;
             element.onended = null;
             element.onerror = null;
+            if (stopGenerated === stop) stopGenerated = void 0;
             if (sounding === element) {
               sounding = void 0;
               playbackFailure = void 0;
@@ -1764,13 +1788,17 @@ function createVideoChatVoice(options = {}) {
             resolve();
           };
           const stop = () => {
+            if (finished) return;
             element.pause();
             finish();
           };
           const fail = () => {
-            playbackFailed = true;
-            stop();
+            if (!finished) {
+              playbackFailed = true;
+              stop();
+            }
           };
+          stopGenerated = stop;
           const clearWatchdog = watchSpeech(Math.max(line.seconds, estimatedBrowserSeconds(text)), fail);
           speechStops.add(stop);
           playbackFailure = fail;
@@ -1785,7 +1813,9 @@ function createVideoChatVoice(options = {}) {
           element.onerror = fail;
           signal.addEventListener("abort", stop, { once: true });
           try {
-            if (!held) void element.play().catch(fail);
+            if (!held) void element.play().catch(() => {
+              if (!held) fail();
+            });
           } catch {
             fail();
           }
@@ -1811,6 +1841,11 @@ function createVideoChatVoice(options = {}) {
       speechStops.clear();
       sounding?.pause();
       sounding = void 0;
+      generatedElement?.pause();
+      generatedElement?.removeAttribute?.("src");
+      generatedElement?.load?.();
+      generatedElement = void 0;
+      stopGenerated = void 0;
       stopBrowser();
       for (const line of lines.values()) {
         if (line.source === "generated") URL.revokeObjectURL(line.src);
@@ -2423,8 +2458,18 @@ function useVideoChatSession(options = {}) {
           planned[position] = plannedScene;
           received[position] = plannedScene;
           const rendererReady = currentOptions.templates?.getTemplate(plannedScene.templateId) ? Promise.resolve() : Promise.resolve(preloadBuiltinTemplate(plannedScene.templateId));
-          const visualPreparation = Promise.all([rendererReady, prepareSceneMedia(plannedScene.variables, controller.signal)]).then(() => void 0, () => {
-            throw new VideoError("Scene could not prepare its visual", { code: "media_not_ready" });
+          const visualPreparation = rendererReady.then(async () => {
+            try {
+              await prepareSceneMedia(plannedScene.variables, controller.signal);
+              return plannedScene;
+            } catch (cause) {
+              if (controller.signal.aborted) throw cause;
+              const override = currentOptions.templates?.getTemplate(plannedScene.templateId);
+              const fallback = !override || supportsExternalVideoBackdrop(override) ? recoverSceneMedia(plannedScene) : void 0;
+              if (!fallback) throw new VideoError("Scene could not prepare its visual", { code: "media_not_ready" });
+              warn(MEDIA_RECOVERY_NOTICE);
+              return fallback;
+            }
           });
           void visualPreparation.catch(() => void 0);
           warmSceneMedia(plannedScene.variables);
@@ -2453,15 +2498,14 @@ function useVideoChatSession(options = {}) {
           narrating = narrated.catch(() => "");
           pending.push(narrated.then(async (line) => {
             if (!isCurrent() || currentAttempt !== attempt) return;
-            const visual = plannedScene;
-            const withNarration = line ? { ...visual, narration: line } : visual;
             const spoken = line ? await prepareSpeech(plannedScene.narrationGroup?.text ?? line, controller.signal).catch((cause) => {
               if (controller.signal.aborted) throw cause;
               warn("Some narration is unavailable; the response will continue.");
               return void 0;
             }) : void 0;
             if (!isCurrent() || currentAttempt !== attempt) return;
-            await visualPreparation;
+            const visual = await visualPreparation;
+            const withNarration = line ? { ...visual, narration: line } : visual;
             const group = plannedScene.narrationGroup;
             if (group && (spoken?.supportsOffsets !== true || voiceRef.current.supportsOffsets !== true || Math.abs(spoken.seconds - group.totalSeconds) > 0.1)) throw new VideoError("Narration group requires matching measured audio with offset support", { code: "narration_group_invalid" });
             ready[position] = group ? withNarration : pacedScene(withNarration, spoken?.seconds, currentOptions.templates);