@koda-sl/baker-cli 0.293.0-dev.cf9d5462d → 0.295.0-dev.f941e60fb

package/dist/cli.js

@@ -16,6 +16,7 @@ import {
   RunAbortedError,
   SEEDANCE_DURATIONS,
   SEEDANCE_PROFILE,
+  SPOKEN_WORDS_PER_SECOND,
   VIDEO_GENERATE_MODELS,
   ValidationError,
   YtDlpError,
@@ -53,6 +54,7 @@ import {
   sha256Hex,
   spineInputFlags,
   spineInputOps,
+  spokenWordsIn,
   supportsLastFrame,
   supportsParam,
   supportsReferenceToVideo,
@@ -60,7 +62,7 @@ import {
   ulid,
   validateCanvasDeep,
   ytDlpBlockSignal
-} from "./chunk-LTXNRRXW.js";
+} from "./chunk-ER3KCAW3.js";
 import {
   csvOrJson,
   daysAgoIso,
@@ -57379,6 +57381,15 @@ import { defineCommand as defineCommand208 } from "citty";
 // src/commands/studio/animate.ts
 import { defineCommand as defineCommand200 } from "citty";
 
+// src/commands/studio/avatarModel.ts
+function pinnedModelHint(input) {
+  if (!input.avatar || !input.model) return null;
+  if (input.model === DEFAULT_VIDEO_GENERATE_MODEL) return null;
+  const speaks = input.generateAudio ?? true;
+  if (!speaks) return null;
+  return `You pinned --model ${input.model} on a speaking avatar. Without it this clip routes to ${DEFAULT_VIDEO_GENERATE_MODEL}, which was chosen for exactly this case: the better picture, both orientations, and audio always on. Drop --model unless the user asked for a specific one \u2014 and note ${input.model} may not take the length or the reference you are counting on.`;
+}
+
 // src/commands/studio/batch.ts
 function projectBatch(generation, full) {
   const compact3 = {
@@ -57593,6 +57604,21 @@ function uploadFailure(path47) {
   };
 }
 
+// src/commands/studio/spokenPace.ts
+var WORDS_PER_SECOND2 = SPOKEN_WORDS_PER_SECOND;
+var spokenWords = spokenWordsIn;
+function wordsThatFit(seconds) {
+  return Math.floor(seconds * WORDS_PER_SECOND2);
+}
+function spokenPaceHint(prompt, seconds) {
+  if (!seconds) return null;
+  const words2 = spokenWords(prompt);
+  const fits = wordsThatFit(seconds);
+  if (words2 === 0 || words2 <= fits) return null;
+  const rate = (words2 / seconds).toFixed(1);
+  return `The line is ${words2} words and this clip is ${seconds}s \u2014 that is ${rate} words a second against a natural 2.5. The model will not run over, it will RUSH: syllables swallowed, endings clipped, and the take unusable however good the picture. Cut the line to about ${fits} words, or give it the length it needs (${Math.ceil(words2 / 2.5)}s). Do not re-render the same line at the same length and hope.`;
+}
+
 // src/commands/studio/animate.ts
 var MODEL_LIST = VIDEO_MODEL_IDS;
 var DEFAULT_MAX_WAIT_MS = 6e5;
@@ -57644,10 +57670,10 @@ registerSchema({
     },
     duration: {
       type: "number",
-      // Left unset this SNAPS onto the model's set rather than failing: the lane
-      // default is 5s and Veo renders only 4/6/8, so an unstated length becomes 4
-      // there. An explicit length the model cannot render is refused by name.
-      description: `Clip length in seconds \u2014 per-model. Unset it and the lane default (5s) snaps onto what the model renders. Past 15s only bytedance/seedance-2.5 will do it, out to 30. Across the roster: ${CLIP_DURATIONS.join(", ")}`,
+      // Left unset this SNAPS onto the model's set rather than failing: Veo renders
+      // only 4/6/8, so an unstated length lands on the nearest of those. An explicit
+      // length the model cannot render is refused by name.
+      description: `Clip length in seconds \u2014 per-model. LEAVE IT OFF when someone speaks: the line sizes the clip itself, at ~2.5 words a second (40 words needs 16s), because a model handed more words than fit does not run over, it rushes and swallows syllables. Unset with nobody speaking is 5s. Past 15s only bytedance/seedance-2.5 will do it, out to 30. Across the roster: ${CLIP_DURATIONS.join(", ")}`,
       required: false
     },
     quality: {
@@ -57821,6 +57847,10 @@ function costHintsFor(body) {
   const hints2 = [
     count === 1 ? `This clip costs about ${total} credits \u2014 roughly 45x an image.` : `${count} takes at about ${perTake} credits each \u2014 about ${total} credits.`
   ];
+  const pinned = pinnedModelHint({ avatar: body.avatar, model: body.model, generateAudio: body.generateAudio });
+  if (pinned) hints2.push(pinned);
+  const pace = spokenPaceHint(body.prompt, body.duration);
+  if (pace) hints2.push(pace);
   if (count > 1) {
     hints2.push(
       "Ask for more than one take only when the MOTION is the risk \u2014 for a look you are unsure about, iterate on the still first."
@@ -57864,7 +57894,7 @@ var animateCommand = defineCommand200({
     "aspect-ratio": { type: "string", description: "Aspect ratio (default 9:16)", required: false },
     duration: {
       type: "string",
-      description: `Seconds, per-model \u2014 unset snaps the 5s default onto the model's set. Roster: ${CLIP_DURATIONS.join(" | ")}`,
+      description: `Seconds, per-model \u2014 leave it off when someone speaks and the line sizes it (~2.5 words/s); 5s when nobody does. Roster: ${CLIP_DURATIONS.join(" | ")}`,
       required: false
     },
     quality: {