@nodaro/prompts 1.29.0 → 1.32.0

package/src/caption-treatments.ts

@@ -0,0 +1,109 @@
+// packages/prompts/src/caption-treatments.ts
+import type { CaptionPlan, SupportedFontName } from "@nodaro/shared"
+import { ADD_CAPTIONS_PRESETS } from "./factory-presets/video-edit.js"
+
+export const HOOK_PLATE_PRESET_ID = "add-captions/hook-plate"
+export const BODY_CAPTIONS_PRESET_ID = "add-captions/body-captions"
+
+/** Levers a spoken language needs on every caption segment; a language with no entry adds none. */
+export interface CaptionLanguageLevers {
+  readonly fontFamily?: SupportedFontName
+  readonly uppercase?: boolean
+}
+
+/**
+ * Keyed by primary language tag. Hebrew: one face that carries Hebrew and Latin, so a Latin
+ * brand inside a Hebrew line is drawn in the same face, and no upper-casing, so the brand keeps
+ * its spelling.
+ */
+export const CAPTION_LEVERS_BY_LANGUAGE: Readonly<Record<string, CaptionLanguageLevers>> = {
+  he: { fontFamily: "Rubik", uppercase: false },
+}
+
+/** The add-captions segment keys a lever set may carry: the segment schema minus times and words. */
+export const CAPTION_SEGMENT_LEVER_KEYS: readonly string[] = [
+  "style", "position", "fontSize", "color", "backgroundColor", "look", "fontFamily", "fontWeight",
+  "strokeColor", "strokeWidth", "highlightColor", "uppercase", "positionY", "animate", "maxWordsPerLine",
+]
+
+export type CaptionSegmentRole = "hook" | "body"
+
+export interface HookPlateSegmentsOptions {
+  /** The body's levers; default = the Body Captions preset. Keys outside CAPTION_SEGMENT_LEVER_KEYS are ignored. */
+  readonly bodyLevers?: Readonly<Record<string, unknown>>
+}
+
+export interface HookPlateSegmentsResult {
+  /** Route-shaped (camelCase) `segments`. Empty means "no captions". */
+  readonly segments: Record<string, unknown>[]
+  /** Segments left out, in order hook, body. */
+  readonly dropped: CaptionSegmentRole[]
+}
+
+const LEVER_KEYS: ReadonlySet<string> = new Set(CAPTION_SEGMENT_LEVER_KEYS)
+
+/** A lever set: segment lever keys only, without null or undefined values. */
+function pickLevers(source: Readonly<Record<string, unknown>> | undefined): Record<string, unknown> {
+  const out: Record<string, unknown> = {}
+  if (!source) return out
+  for (const [k, v] of Object.entries(source)) if (LEVER_KEYS.has(k) && v !== undefined && v !== null) out[k] = v
+  return out
+}
+
+function presetLevers(id: string): Record<string, unknown> {
+  const preset = ADD_CAPTIONS_PRESETS.find((p) => p.id === id)
+  if (!preset) throw new Error(`The ${id} caption preset is missing from the catalog.`)
+  return pickLevers(preset.data)
+}
+
+const PLATE_LEVERS = presetLevers(HOOK_PLATE_PRESET_ID)
+const BODY_LEVERS = presetLevers(BODY_CAPTIONS_PRESET_ID)
+
+function languageLevers(tag: string | undefined): CaptionLanguageLevers {
+  if (typeof tag !== "string") return {}
+  const primary = tag.trim().split(/[-_]/)[0]!.toLowerCase()
+  return Object.hasOwn(CAPTION_LEVERS_BY_LANGUAGE, primary) ? CAPTION_LEVERS_BY_LANGUAGE[primary]! : {}
+}
+
+const isTime = (n: unknown): n is number => typeof n === "number" && Number.isFinite(n)
+
+/**
+ * One caption plan → the `segments` of one add-captions request: the opening line as the Hook
+ * Plate, then the body's words in the Body Captions style (or the caller's own body levers).
+ * The styling is read from the two presets, never restated here. Pure: no I/O.
+ */
+export function hookPlateCaptionSegments(plan: CaptionPlan, opts: HookPlateSegmentsOptions = {}): HookPlateSegmentsResult {
+  const hookEnd = plan.hookEndMs
+  if (!isTime(hookEnd) || hookEnd < 0) return { segments: [], dropped: ["hook", "body"] }
+  const language = languageLevers(plan.language)
+  const limit = typeof plan.videoDurationMs === "number" ? plan.videoDurationMs : Number.POSITIVE_INFINITY
+  const segments: Record<string, unknown>[] = []
+  const dropped: CaptionSegmentRole[] = []
+
+  const plateEnd = Math.min(hookEnd, limit)
+  if (plateEnd > 0 && typeof plan.hookText === "string" && /\S/.test(plan.hookText)) {
+    segments.push({ startMs: 0, endMs: plateEnd, text: plan.hookText, ...PLATE_LEVERS, ...language })
+  } else {
+    dropped.push("hook")
+  }
+
+  // The body starts where the opening line ends, even when no plate is shown, so it never
+  // draws over the opening line.
+  const bodyEnd = Math.min(plan.bodyEndMs, limit)
+  const words = Array.isArray(plan.captions) ? plan.captions : []
+  const chosen = pickLevers(opts.bodyLevers)
+  const bodyLevers = Object.keys(chosen).length > 0 ? chosen : BODY_LEVERS
+  if (bodyEnd > hookEnd && words.length > 0) {
+    segments.push({
+      startMs: hookEnd,
+      endMs: bodyEnd,
+      captions: words.map((w) => ({ text: w.text, startMs: w.startMs, endMs: w.endMs })),
+      ...bodyLevers,
+      ...language,
+    })
+  } else {
+    dropped.push("body")
+  }
+
+  return { segments, dropped }
+}

package/src/factory-presets/sfx.ts

@@ -53,6 +53,27 @@ export const TEXT_TO_AUDIO_PRESETS: readonly FactoryPreset[] = [
     group: "Ambiences (loopable)",
     data: { provider: "elevenlabs-sfx", duration: 22, loop: true, promptInfluence: 0.5, prompt: "low sci-fi ambient drone, ominous spaceship hum" },
   },
+  {
+    id: "text-to-audio/living-room-tone",
+    name: "Living Room Tone",
+    description: "Quiet, lived-in room.",
+    group: "Ambiences (loopable)",
+    data: { provider: "elevenlabs-sfx", duration: 22, loop: true, promptInfluence: 0.4, prompt: "quiet living room room tone, faint air and a soft distant household hum, steady and even, an empty room at rest" },
+  },
+  {
+    id: "text-to-audio/kitchen-hum",
+    name: "Kitchen Hum",
+    description: "Fridge hum in a still kitchen.",
+    group: "Ambiences (loopable)",
+    data: { provider: "elevenlabs-sfx", duration: 22, loop: true, promptInfluence: 0.4, prompt: "home kitchen room tone, a low steady refrigerator hum with faint air, calm and continuous, an empty kitchen at rest" },
+  },
+  {
+    id: "text-to-audio/cafe-murmur",
+    name: "Café Murmur",
+    description: "Soft, indistinct café chatter.",
+    group: "Ambiences (loopable)",
+    data: { provider: "elevenlabs-sfx", duration: 22, loop: true, promptInfluence: 0.4, prompt: "café background ambience, a soft indistinct murmur of distant conversation and the occasional faint cup clink, steady and even, no clear words" },
+  },
 
   // ── UI & Stingers ────────────────────────────────────────────────────────
   {

package/src/factory-presets/text.ts

@@ -80,6 +80,24 @@ export const LLM_CHAT_PRESETS: readonly FactoryPreset[] = [
       temperature: 0.7,
     },
   },
+  {
+    id: "llm-chat/hook-generator",
+    name: "Hook Generator",
+    description: "10 opening lines for a talking-to-camera video.",
+    group: "Writing & Marketing",
+    data: {
+      systemPrompt:
+        "You write opening lines for short vertical videos in which one person talks straight to the camera about a product, a site or an idea. Given the user's topic, write 10 different opening lines.\n\n" +
+        "Every line:\n" +
+        "- is one short spoken clause of at most 8 words;\n" +
+        "- leads with the payload the viewer can check: a number, a price, a time saved, a before and after, a contrast with the usual way, or a named result;\n" +
+        "- opens on the claim itself — no greeting, no filler word, no introduction of the product before the claim;\n" +
+        "- sounds said, not written: plain words, contractions welcome, no hashtags, no emoji, no stage directions;\n" +
+        "- spells the product or brand exactly as the user gives it and writes numbers as digits.\n\n" +
+        "Vary the angle across the 10: a surprising number, a price, time saved, a mistake to avoid, a before and after, a contrast, a named result, a direct question. Write in the language the user writes in. Return only the numbered list, one line each.",
+      temperature: 0.9,
+    },
+  },
 
   // ── Utility ──────────────────────────────────────────────────────────────
   {
@@ -273,6 +291,17 @@ export const IMAGE_TO_TEXT_PRESETS: readonly FactoryPreset[] = [
     group: "Extraction",
     data: { detailLevel: "detailed", customPrompt: "Write a compelling e-commerce product description based on this product image." },
   },
+  {
+    id: "image-to-text/site-screenshot-summary",
+    name: "Site Screenshot Summary",
+    description: "Section, headings and figures from a web page.",
+    group: "Extraction",
+    data: {
+      detailLevel: "structured",
+      customPrompt:
+        "Describe this website screenshot as five labelled lines. Section: which part of the site it is (for example the top of the home page, a feature, how it works, reviews, pricing, sign-up, navigation or footer). Headings: every heading exactly as written. Figures: every number, price, rating and date exactly as written, each with what it refers to. Brand: the product or company name exactly as spelled. First look: what a viewer notices first, in one sentence. Write \"none\" on a line with nothing to report.",
+    },
+  },
 
   // ── Creative ─────────────────────────────────────────────────────────────
   {

package/src/factory-presets/video-edit.ts

@@ -95,6 +95,20 @@ export const ADD_CAPTIONS_PRESETS: readonly FactoryPreset[] = [
     group: "Caption Styles",
     data: { style: "subtitle", position: "top", fontSize: 36, color: "#FFFFFF", autoTranscribe: true },
   },
+  {
+    id: "add-captions/hook-plate",
+    name: "Hook Plate",
+    description: "Your opening line as a bold plate near the top.",
+    group: "Caption Styles",
+    data: { style: "subtitle", look: "clean", position: "top", positionY: 20, fontSize: 32, fontWeight: 800, maxWordsPerLine: 4, color: "#111111", backgroundColor: "#FFFFFF", autoTranscribe: false },
+  },
+  {
+    id: "add-captions/body-captions",
+    name: "Body Captions",
+    description: "Word-by-word captions under the face.",
+    group: "Caption Styles",
+    data: { style: "word-highlight", look: "outline", position: "bottom", positionY: 72, fontSize: 32, maxWordsPerLine: 3, color: "#FFFFFF", autoTranscribe: true },
+  },
 ]
 
 export const COMBINE_VIDEOS_PRESETS: readonly FactoryPreset[] = [

package/src/factory-presets/voice.ts

@@ -115,6 +115,20 @@ export const VOICE_DESIGN_PRESETS: readonly FactoryPreset[] = [
     group: "Narration & Character",
     data: { voiceDescription: "A calm, soft-spoken meditation guide with a slow, breathy, reassuring delivery." },
   },
+  {
+    id: "voice-design/casual-creator-female",
+    name: "Casual Creator (female)",
+    description: "Relaxed, warm, everyday.",
+    group: "Narration & Character",
+    data: { voiceDescription: "An adult woman in her late twenties talking casually to the viewer as if telling a friend about something she tried: relaxed, warm and upbeat, a natural conversational pace with small pauses, a light neutral American accent, recorded close and dry in a quiet room, never polished or announcer-like." },
+  },
+  {
+    id: "voice-design/casual-creator-male",
+    name: "Casual Creator (male)",
+    description: "Relaxed, warm, everyday.",
+    group: "Narration & Character",
+    data: { voiceDescription: "An adult man in his late twenties talking casually to the viewer as if telling a friend about something he tried: relaxed, warm and upbeat, a natural conversational pace with small pauses, a light neutral American accent, recorded close and dry in a quiet room, never polished or announcer-like." },
+  },
 
   // ── Professional & Assistant ─────────────────────────────────────────────
   {

package/src/index.ts

@@ -86,8 +86,10 @@ export * from "./image-reference-doctrine.js"
 export * from "./scene3d-reference-doctrine.js"
 export * from "./prompt-wizard-categories.js"
 export * from "./resolve-prompt.js"
+export { stripAudioTags } from "./audio-tags.js"
 export * from "./node-prompt-fields.js"
 export * from "./factory-presets.js"
+export * from "./caption-treatments.js"
 export * from "./style-presets.js"
 export * from "./object-asset-presets.js"
 export * from "./factory-snippets/index.js"
@@ -103,3 +105,6 @@ export * from "./ad-creative-analysis.js"
 
 // --- Picker art: option pictures + topic icons (editor and API) ---
 export * from "./picker-art/index.js"
+
+// --- What a text-requiring node would send (the orchestrator's empty-input skip reads it) ---
+export * from "./node-send-text.js"

package/src/node-send-text.ts

@@ -0,0 +1,120 @@
+import { SOCIAL_POST_NODE_TYPES } from "@nodaro/shared"
+import { computeLlmChatFields, computeNodePrompt, computeScriptTopic } from "./resolve-prompt.js"
+
+/**
+ * Output nodes that send exactly what is WIRED into them — a webhook's
+ * parameters, a social post's caption and media. When every wire produced
+ * nothing in this run (a Filter List that kept no row, a node that was itself
+ * skipped), there is nothing to send: the node is skipped with
+ * `skipReason: "empty_input"` instead of posting an empty payload (a site that
+ * validates its input answers 400 and alerts its owner) or failing the run
+ * (a publish route refuses a post with no text and no media). A typed caption
+ * with no wire at all is the author's choice and still runs, as before. ONE
+ * list, read by the orchestrator's skip rule (`empty-input-skips.ts`).
+ */
+export const WIRED_OUTPUT_NODE_TYPES: ReadonlySet<string> = new Set(["webhook-output", ...SOCIAL_POST_NODE_TYPES])
+
+/**
+ * Node types that send TEXT to a model or a voice and fail on an empty one —
+ * `userInput: Too small` (llm-chat), "no text found" (text-to-speech) — so a
+ * run with nothing upstream can skip them instead of failing. Never an image
+ * or video node (an empty prompt is a legal request there) and never a social
+ * post (an empty caption is the author's choice). ONE list, read by the
+ * orchestrator's skip rule (`empty-input-skips.ts`); the engine never names a
+ * type on its own.
+ */
+export const TEXT_REQUIRED_NODE_TYPES: ReadonlySet<string> = new Set([
+  "llm-chat",
+  "ai-writer",
+  "generate-script",
+  "text-to-speech",
+  "generate-music",
+  "text-to-audio",
+  // Save to Collection refuses an empty record ("nothing to save"): a feed
+  // with nothing new wired straight into it must skip it, not fail the run.
+  "collection-write",
+  // Generate Image: an empty prompt is a legal request in general (references
+  // alone), but when the text it was WIRED produced nothing this run — the skip
+  // rule's precondition — an empty prompt is a paid picture of nothing.
+  "generate-image",
+])
+
+/**
+ * The node's data without its prompt pre/post text: the skip decision is about
+ * the CORE text. A prefix alone ("Write a news item about:") is not something
+ * to send — with the affixes applied, an empty core read as non-empty and an
+ * idle tick paid for a model call on the prefix.
+ */
+const withoutAffixes = (data: Record<string, unknown>): Record<string, unknown> => ({ ...data, promptPrefix: undefined, promptSuffix: undefined })
+
+export interface NodeSendTextArgs {
+  /** A list fan-out item (highest precedence). */
+  override?: string
+  /** The text wired into the node's main text input. */
+  wired?: string
+  /** llm-chat only: the text wired into `system-prompt`. */
+  wiredSystemPrompt?: string
+  refMap: ReadonlyMap<string, string>
+}
+
+const present = (s?: string): s is string => typeof s === "string" && s.trim().length > 0
+
+/**
+ * ai-writer's user input — the rule its sync-HTTP body applies: a fan-out
+ * item, else the wired text, else the typed `userInput`, else the legacy
+ * `prompt`. (ai-writer is renamed to llm-chat on load; saved graphs that
+ * never reloaded still run it.)
+ */
+export function computeAiWriterInput(
+  data: Record<string, unknown>,
+  { override, wired }: Pick<NodeSendTextArgs, "override" | "wired">,
+): string {
+  return [override, wired, data.userInput as string | undefined, data.prompt as string | undefined].find(present) ?? ""
+}
+
+/**
+ * The text a text-requiring node would send, by the SAME rule its executor
+ * uses (`computeLlmChatFields`, `computeScriptTopic`, `computeNodePrompt`,
+ * `computeAiWriterInput`) — so "would this node run on nothing?" is answered
+ * by the function that decides what it sends, not by a second reading of it.
+ * Undefined for any other node type.
+ */
+export function computeNodeSendText(
+  nodeType: string,
+  data: Record<string, unknown>,
+  args: NodeSendTextArgs,
+): string | undefined {
+  const bare = withoutAffixes(data)
+  switch (nodeType) {
+    case "llm-chat":
+      return computeLlmChatFields(bare, {
+        override: args.override,
+        wiredUserInput: args.wired,
+        wiredSystemPrompt: args.wiredSystemPrompt,
+        refMap: args.refMap,
+      }).userInput
+    case "ai-writer":
+      return computeAiWriterInput(bare, args)
+    case "generate-script":
+      return computeScriptTopic(bare, { override: args.override, wired: args.wired, refMap: args.refMap })
+    case "generate-music": {
+      // A music node sings typed lyrics with no prompt; only both empty is nothing to send.
+      const prompt = computeNodePrompt(nodeType, bare, { override: args.override, wired: args.wired, refMap: args.refMap })
+      return present(prompt) ? prompt : present(data.lyrics as string | undefined) ? (data.lyrics as string) : ""
+    }
+    case "text-to-speech":
+    case "text-to-audio":
+    case "generate-image":
+      return computeNodePrompt(nodeType, bare, { override: args.override, wired: args.wired, refMap: args.refMap })
+    case "collection-write": {
+      // The record's content: the item (a fan-out row, else what reached `in`),
+      // else a title, text or link typed on the node. A picture alone is a
+      // record too, but it never starves the node: only a wired TEXT input that
+      // produced nothing can (the skip rule's precondition).
+      const typed = [data.title, data.text, data.link].find((v): v is string => present(v as string | undefined))
+      return [args.override, args.wired, typed].find(present) ?? ""
+    }
+    default:
+      return undefined
+  }
+}