@vanillaskyai/video 0.10.16 → 0.10.18

package/CHANGELOG.md

@@ -4,6 +4,16 @@ VanillaSky follows semantic versioning. This changelog begins with the 0.1 beta.
 
 ## Unreleased
 
+## 0.10.18
+
+- Preserve later valid shots after a malformed body record containing an unescaped newline, while keeping malformed initial containers rejected. Bound unfinished JSON records independently of provider chunk sizes.
+- Keep the outgoing scene visible while a canonical image backdrop prepares or reaches its bounded chapter recovery.
+
+## 0.10.17
+
+- Let video playback prepare without waiting for an optional poster, while retaining actual video-frame and narration readiness checks.
+- Guide shorter, useful spoken openings and compact streamed briefs, with clearer direct answers, comparisons, and narration-aligned actions.
+
 ## 0.10.16
 
 - Start the first fully prepared scene without waiting for an eight-second startup buffer, while preserving narration and visual readiness.

package/dist/check-runtime.js

@@ -1,6 +1,6 @@
 import {
   VideoFrame
-} from "./chunk-MXLVTOGR.js";
+} from "./chunk-M4QTEJTK.js";
 import "./chunk-5JBMYQP6.js";
 import "./chunk-224QNWRA.js";
 import "./chunk-SPVTJH3F.js";

package/dist/{chunk-MXLVTOGR.js → chunk-M4QTEJTK.js}

@@ -448,11 +448,15 @@ function VideoFrame({
   }
   const previousIndex = timeline.findIndex((range) => sceneReadinessKey(range.scene) === displayedKey.current);
   const previous = timeline[previousIndex];
-  const canPrepare = (range) => Boolean(range && mediaAudioMuted && sceneHasVideoBackdrop(range) && supportsExternalVideoBackdrop(kit.getTemplate(range.scene.templateId)));
+  const canPrepare = (range) => Boolean(range && (mediaAudioMuted || !sceneHasVideoBackdrop(range)) && sceneHasBackdrop(range) && supportsExternalVideoBackdrop(kit.getTemplate(range.scene.templateId)));
   const canRetain = (range) => canPrepare(range) || range?.scene.templateId === "chapterTitle";
   const hasPlayableMedia = (range) => {
     if (!range || !preparedMedia.has(sceneReadinessKey(range.scene))) return false;
     const layer = [...recoveryRoot.current?.querySelectorAll("[data-layer-scene-id]") ?? []].find((node) => node.getAttribute("data-layer-scene-id") === range.scene.id);
+    if (!sceneHasVideoBackdrop(range)) {
+      const image = [...layer?.querySelectorAll("img") ?? []].find((element) => element.getAttribute("src") === range.scene.variables.mediaUrl);
+      return Boolean(image && image.getAttribute("src") === range.scene.variables.mediaUrl && image.complete && image.naturalWidth > 0);
+    }
     const video = layer?.querySelector("video");
     return Boolean(video && video.getAttribute("src") === range.scene.variables.mediaUrl && video.currentSrc === video.src && video.readyState >= HTMLMediaElement.HAVE_FUTURE_DATA);
   };

package/dist/react.js

@@ -40,7 +40,7 @@ import {
   VideoFrame,
   getDimensions,
   sceneReadinessKey
-} from "./chunk-MXLVTOGR.js";
+} from "./chunk-M4QTEJTK.js";
 import "./chunk-5JBMYQP6.js";
 import "./chunk-224QNWRA.js";
 import {
@@ -1501,9 +1501,7 @@ async function prepareSceneMedia(variables, signal) {
   const url = typeof variables.mediaUrl === "string" ? variables.mediaUrl.trim() : "";
   if (!url || variables.mediaType === "gradient") return "graphic";
   const video = variables.mediaType === "video" || /\.(mp4|webm|mov)(?:[?#]|$)/i.test(url);
-  const poster = typeof variables.mediaPoster === "string" ? variables.mediaPoster.trim() : "";
-  if (video && !poster) return "awaiting-video-frame";
-  let posterFailed = false;
+  if (video) return "awaiting-video-frame";
   await new Promise((resolve, reject) => {
     const image = new Image();
     let settled = false;
@@ -1526,13 +1524,10 @@ async function prepareSceneMedia(variables, signal) {
       void (image.decode ? image.decode() : Promise.resolve()).then(() => finish(), () => finish(new Error("Could not prepare scene image")));
     };
     image.onerror = () => finish(new Error("Could not prepare scene image"));
-    image.src = video ? poster : url;
+    image.src = url;
     if (image.complete && image.naturalWidth > 0) image.onload(new Event("load"));
-  }).catch((error) => {
-    if (!video || signal.aborted) throw error;
-    posterFailed = true;
   });
-  return video ? posterFailed ? "awaiting-video-frame" : "poster-bridge" : "image";
+  return "image";
 }
 
 // src/video-chat/voice.ts

package/dist/server.js

@@ -1074,6 +1074,13 @@ function createChatShotPlanner(options) {
           for (; cursor < buffer.length; cursor++) {
             const character = buffer[cursor];
             if (quoted) {
+              if ((character === "\n" || character === "\r") && brief && depth === 1 && buffer[0] === "{") {
+                const raw = buffer.slice(0, cursor + 1);
+                buffer = buffer.slice(cursor + 1);
+                cursor = depth = 0;
+                quoted = escaped = false;
+                return raw;
+              }
               if (escaped) escaped = false;
               else if (character === "\\") escaped = true;
               else if (character === '"') quoted = false;
@@ -1094,17 +1101,22 @@ function createChatShotPlanner(options) {
           for await (const delta of upstream) {
             context.signal.throwIfAborted();
             if (typeof delta !== "string") throw new Error("The LLM adapter returned a non-text delta");
-            buffer += delta;
-            if (buffer.length > 32768) throw new Error("Chat plan line exceeds the bounded stream limit");
-            let raw = takeFrame();
-            while (raw !== void 0) {
-              try {
-                const part = line(raw);
-                if (part) yield JSON.stringify(part) + "\n";
-              } catch (cause) {
-                reject(cause);
+            for (let offset = 0; offset < delta.length; ) {
+              const capacity = 32768 - buffer.length;
+              if (capacity <= 0) throw new Error("Chat plan line exceeds the bounded stream limit");
+              const piece = delta.slice(offset, offset + capacity);
+              buffer += piece;
+              offset += piece.length;
+              let raw = takeFrame();
+              while (raw !== void 0) {
+                try {
+                  const part = line(raw);
+                  if (part) yield JSON.stringify(part) + "\n";
+                } catch (cause) {
+                  reject(cause);
+                }
+                raw = takeFrame();
               }
-              raw = takeFrame();
             }
           }
           if (buffer.trim()) {
@@ -1217,18 +1229,21 @@ function createVideoChatResponseInstructions(generatedVideoAvailable, openingAlr
   return [
     'Use the exact record type "answer" for the first brief and "shot" for developing beats. Output JSON records only, with no prose outside them, including when explaining a limitation.',
     "Write a complete, intentful video answer as newline-delimited JSON. Match the user's form and tone; mixed intents can combine directions.",
-    `First write one brief: {"type":"answer","intent":"explanation|story|comedy|imagination|practical","opening":"a short inviting spoken introduction of 6\u20139 words","subject":"literal visual subject","development":"the essential development of this answer","visualDirection":"consistent subjects, appearance and visual approach","ending":{"title":"short meaningful chapter title, at most 65 characters","narration":"the authored payoff","subject":"literal subject","action":"visible action or change","durationSec":${clipDurationSec},"continuity":"cut|continue"}}.`,
+    `First write one brief: {"type":"answer","intent":"explanation|story|comedy|imagination|practical","opening":"one useful spoken line of 4\u20137 ordinary words","subject":"literal visual subject","development":"the essential development of this answer","visualDirection":"consistent subjects, appearance and visual approach","ending":{"title":"short meaningful chapter title, at most 65 characters","narration":"the authored payoff","subject":"literal subject","action":"visible action or change","durationSec":${clipDurationSec},"continuity":"cut|continue"}}.`,
+    "The opening should take roughly 2\u20133 seconds at a natural pace: give the core answer, a useful starting cue, or the story's immediate situation. No greeting, topic announcement, promise to explain, or description of loading. Never compress away an essential qualifier just to hit the word target.",
     `Then stream each developing shot on its own line: {"type":"shot","title":"short meaningful chapter title, at most 65 characters","narration":"the exact spoken beat","subject":"2\u20138 literal filmable words, at most 80 characters","action":"concrete subject, action or visible change and useful framing","durationSec":${clipDurationSec},"continuity":"cut|continue"}.`,
     `The selected footage mode is ${mode === "pexels" ? "Pexels stock search: use literal filmable subjects; never imply stock proves a mechanism or depicts fictional events exactly" : `AI video, with at most ${generatedVideoAvailable ? maxGeneratedVideos : 0} generation attempts`}. Missing footage becomes the authored chapter title, with complete narration. Never truncate already-authored narration when footage fails. The host selects providers; do not make source choices.`,
     ...mode === "cinematic" ? [generatedVideoAvailable && maxGeneratedVideos > 0 ? `Plan at most ${maxGeneratedVideos} generated-video beats in total, including the saved ending. Use at most ${maxGeneratedVideos - 1} developing shot records; the ending uses the remaining beat. The opening chapter does not consume a generated clip. Before writing, choose a concise, complete treatment that fits this budget: combine related ideas, preserve essential facts and qualifiers, and finish the requested answer. Do not plan an extra chapter tail simply because generation attempts will run out.${maxGeneratedVideos === 1 ? " Put the complete answer in the saved ending, set development to an empty string, and emit no developing shot records." : ""}` : "No generated-video attempts are available; plan a complete chapter-led answer with a useful ending. Do not omit the answer to satisfy a zero clip budget."] : [],
     ...mode === "pexels" ? ["Stock queries must retain the essential subject, activity and distinguishing equipment in the shot's subject field, within its word limit. That field alone is the search query; action and visualDirection do not refine it. Prefer common observable actions with usable framing. Do not replace the required actor or activity with scenery, a different sport or a loosely related setting. Preserve fictional or comic narration, but do not depend on stock showing an exact invented expression or sequence; choose an illustrative action that supports the beat."] : [],
     ...mode === "pexels" ? ['Include stockSelection on every shot and the saved ending when the essential subject is known: "stockSelection":{"subject":"essential actor or object category","activity":"optional literal activity","equipment":"optional distinguishing equipment","exclude":["optional contradictory subject or activity"]}. Each phrase must be 1\u20134 words and at most 48 characters; exclude has at most 3 phrases. The essential subject is separate from the setting: do not use scenery, mood, camera framing or incidental appearance as the actor. Keep the search query broad enough to find footage; the optional hint helps select results without substituting a different actor or task. Use exclusions only for actual contradictions, not every detail absent from the story. Omit unknown fields or the whole hint rather than inventing an anchor. This is selection guidance, not verification that footage depicts the exact narration.'] : [],
+    "Keep development to one concise sentence and visualDirection to the few details needed for consistency. Emit the complete brief, then the first developing shot immediately when developing shots are needed and allowed by the budget; otherwise end after the brief. Continue the same stream without an outline, recap or second planning pass. The saved ending must still contain the complete payoff before the brief is emitted.",
     "For a very short answer whose ending alone fulfills the request, development may be empty and no developing shots are needed. Otherwise, develop the essential content before the ending.",
     "The brief's ending is saved and played after your developing shots. Do not repeat it as a shot. Stop writing after the last developing shot. No technical events, identifiers, template choices, media providers, URLs or unlisted fields.",
     "Every shot uses moving footage with separate narration and subtitles. Generated footage is silent: do not ask its subjects to speak or render words. No headline cards or on-screen explanatory text.",
     `Each clip has at most ${clipDurationSec} seconds. Write spoken beats that fit naturally, usually ${Math.floor(clipDurationSec * 1.6)}\u2013${Math.floor(clipDurationSec * 2)} words per shot. Split longer ideas across purposeful shots, preserving facts and qualifiers. Never truncate a claim to meet a word target. Use only the shots needed within the total duration, including the ending; do not pad to a fixed count.`,
-    "Identify the full answer and its ending before developing shots. Each action must support what is said: camera movement alone is not progression. Vary scale, viewpoint and meaningful details while keeping subjects consistent.",
-    "Explanations: clarify the actual causal mechanism, separating physical cause from a metaphor. Generated cutaways and animation illustrate ideas; they are not factual evidence. Preserve uncertainty, quantities and conditions; never invent evidence or quotations.",
+    "Identify the full answer and its ending before developing shots. Each shot should carry one clear action or change, timed to the narration of that beat; do not describe an outcome before its shot. Use framing that lets the viewer see the relevant action, not just its setting. Each action must support what is said: camera movement alone is not progression. Vary scale, viewpoint and meaningful details while keeping subjects consistent.",
+    "Explanations: answer the actual question first, then show the essential causal link rather than a tour of the topic. Clarify the actual causal mechanism, separating physical cause from a metaphor. Generated cutaways and animation illustrate ideas; they are not factual evidence. Preserve uncertainty, quantities and conditions; never invent evidence or quotations.",
+    "Comparisons and choices: Compare the same criteria for both alternatives, using only supported or supplied differences. Finish with the requested choice and the condition that makes it appropriate; if evidence is insufficient, say what is missing. Do not invent scores, advantages or a winner. Use explanation or practical intent as appropriate, not a new record type.",
     "Stories: portray characters making choices and experiencing consequences; use consistent character descriptions and an earned resolution, not a promised next scene.",
     "Comedy: establish the premise, time the visual or spoken reveal, allow a reaction beat, and stop on the payoff without explaining the joke.",
     "Imagination: make the impossible action concrete, establish the world's internal rules and keep its imagery consistent. Do not replace imagination with an explanation of it.",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vanillaskyai/video",
-  "version": "0.10.16",
+  "version": "0.10.18",
   "description": "Open-source voice-and-video chat SDK for AI applications.",
   "keywords": [
     "video-chat",

package/starters/video-chat/package.json

@@ -9,7 +9,7 @@
     "preview": "vite preview"
   },
   "dependencies": {
-    "@vanillaskyai/video": "0.10.16",
+    "@vanillaskyai/video": "0.10.18",
     "react": "^19.2.8",
     "react-dom": "^19.2.8",
     "@ai-sdk/anthropic": "^3.0.0",