@agentmedia/schema 0.2.2 → 0.4.0

package/src/video.ts

@@ -221,18 +221,54 @@ export const SubtitleSchema = z.object({
 
 export type SubtitleInput = z.infer<typeof SubtitleSchema>;
 
-export const ProductReviewSchema = z.object({
-  product_url: z.string().url(),
+export const SaasReviewSchema = z.preprocess(flattenGroupedParams, z.object({
+  script: z.string().min(50).max(3000).optional(),
+  prompt: z.string().min(1).max(1000).optional(),
+  product_url: z.string().url().optional(),
   angle: z.enum(REVIEW_ANGLES).optional(),
-  tone: z.enum(TONES).optional(),
   actor_slug: z.string().min(1).max(100).optional(),
+  persona_slug: z.string().min(1).max(100).optional(),
+  face_photo_url: z.string().url().optional(),
+  voice: z.string().max(100).optional(),
   target_duration: z.number().refine(
     (v) => (DURATIONS as readonly number[]).includes(v),
     { message: 'target_duration must be 5, 10, or 15' },
   ).optional(),
-});
+  style: z.string().max(50).optional(),
+  tone: z.enum(TONES).optional(),
+  voice_speed: z.number().min(0.7).max(1.5).optional(),
+  music: z.enum(MUSIC_GENRES).optional(),
+  cta: z.string().max(100).optional(),
+  aspect_ratio: z.enum(ASPECT_RATIOS).optional(),
+  template: z.string().max(50).optional(),
+  composition_mode: z.literal('pip').optional(),
+  pip_options: PipOptionsSchema.optional(),
+  allow_broll: z.boolean().optional(),
+  broll_model: z.enum(BROLL_MODELS).optional(),
+  broll_images: z.array(z.string().url()).max(10).optional(),
+  dub_language: z.string().max(10).optional(),
+  webhook_url: z.string().url().optional(),
+  scenes: z.array(
+    z.object({
+      type: z.enum(SCENE_TYPES).optional(),
+    }).passthrough(),
+  ).max(30).optional(),
+}).refine(
+  (data) => data.script !== undefined || data.prompt !== undefined || data.product_url !== undefined,
+  { message: 'script, prompt, or product_url is required' },
+));
 
-export type ProductReviewInput = z.infer<typeof ProductReviewSchema>;
+export type SaasReviewInput = z.infer<typeof SaasReviewSchema>;
+
+/**
+ * @deprecated Use SaasReviewSchema. Kept for API/SDK compatibility.
+ */
+export const ProductReviewSchema = SaasReviewSchema;
+
+/**
+ * @deprecated Use SaasReviewInput. Kept for API/SDK compatibility.
+ */
+export type ProductReviewInput = SaasReviewInput;
 
 export const SHOW_YOUR_APP_WORDS_PER_SECOND = 3;
 export const SHOW_YOUR_APP_DURATIONS = [5, 10, 15] as const;
@@ -262,6 +298,60 @@ export const ShowYourAppSchema = z.object({
 
 export type ShowYourAppInput = z.infer<typeof ShowYourAppSchema>;
 
+// ── Laptop UGC ──────────────────────────────────────────────────────────────
+
+export const LAPTOP_UGC_WORDS_PER_SECOND = 3;
+export const LAPTOP_UGC_DURATIONS = [15, 20] as const;
+export const LAPTOP_UGC_MAX_RECORDING_BYTES = 10 * 1024 * 1024; // 10 MB
+export const LAPTOP_UGC_MAX_RECORDING_SECONDS = 10;
+export const LAPTOP_UGC_RECORDING_MIME_TYPES = ['video/mp4', 'video/quicktime'] as const;
+export const LAPTOP_UGC_MAX_IMAGE_BYTES = 5 * 1024 * 1024; // 5 MB
+export const LAPTOP_UGC_IMAGE_MIME_TYPES = ['image/png', 'image/jpeg', 'image/webp'] as const;
+
+export const LaptopUgcSchema = z.object({
+  app_screen_recording_url: z.string().url().optional().describe('Public URL of a screen recording of your app (mp4 or mov). Max 10 MB, max 10 seconds. Provide this OR app_screen_image_url, not both.'),
+  app_screen_image_url: z.string().url().optional().describe('Public URL of a static screenshot of your app (png, jpeg, or webp). Max 5 MB. Provide this OR app_screen_recording_url, not both.'),
+  actor_slug: z.string().min(1).max(100).optional().describe('Actor slug from GET /v1/actors. Random actor if omitted.'),
+  script: z.string().min(5).max(3000).describe('Full voiceover script. Auto-split at sentence boundary into two halves for the two face shots.'),
+  duration: z.number().refine(
+    (v) => (LAPTOP_UGC_DURATIONS as readonly number[]).includes(v),
+    { message: 'duration must be 15 or 20' },
+  ).optional().default(20),
+  subtitle_style: z.enum(['hormozi', 'none']).optional().default('hormozi'),
+  webhook_url: z.string().url().optional(),
+}).superRefine((val, ctx) => {
+  // Exactly one of recording / image must be provided
+  const hasRecording = !!val.app_screen_recording_url;
+  const hasImage = !!val.app_screen_image_url;
+  if (hasRecording && hasImage) {
+    ctx.addIssue({
+      code: z.ZodIssueCode.custom,
+      path: ['app_screen_recording_url'],
+      message: 'Provide app_screen_recording_url OR app_screen_image_url, not both',
+    });
+  }
+  if (!hasRecording && !hasImage) {
+    ctx.addIssue({
+      code: z.ZodIssueCode.custom,
+      path: ['app_screen_recording_url'],
+      message: 'Either app_screen_recording_url or app_screen_image_url is required',
+    });
+  }
+  // Word-count cap
+  const duration = val.duration ?? 20;
+  const maxWords = duration * LAPTOP_UGC_WORDS_PER_SECOND;
+  const wordCount = val.script.trim().split(/\s+/).filter(Boolean).length;
+  if (wordCount > maxWords) {
+    ctx.addIssue({
+      code: z.ZodIssueCode.custom,
+      path: ['script'],
+      message: `Script has ${wordCount} words; max ${maxWords} for ${duration}s at ${LAPTOP_UGC_WORDS_PER_SECOND} words/sec`,
+    });
+  }
+});
+
+export type LaptopUgcInput = z.infer<typeof LaptopUgcSchema>;
+
 export const PRODUCT_ACTING_WORDS_PER_SECOND = 3;
 export const PRODUCT_ACTING_DURATIONS = [5, 10, 15] as const;
 export const PRODUCT_ACTING_TEMPLATES = [
@@ -327,3 +417,167 @@ export const ProductActingSchema = z.object({
 });
 
 export type ProductActingInput = z.infer<typeof ProductActingSchema>;
+
+// ── Character Video ────────────────────────────────────────────────────────
+//
+// 3-step pipeline:
+//   Step 1: Character reference sheet (PNG)
+//             — POST /v1/character/sheet-generate
+//             — actor_slug returns portrait_url (free), description calls
+//               gpt-image-2 (~$0.04)
+//   Step 2: Storyboard sheet (PNG)
+//             — POST /v1/character/storyboard-generate
+//             — gpt-image-2 paints numbered panels using the character
+//               sheet as a reference so panels stay on-character (~$0.04)
+//   Step 3: Final video (this generator)
+//             — Seedance 2.0 multimodal: image_urls = [character, storyboard]
+//               + short action_prompt + duration/ratio settings
+//
+// All three URLs must be public HTTPS so Seedance can fetch them.
+
+export const CHARACTER_VIDEO_DURATIONS = [5, 10] as const;
+export const CHARACTER_VIDEO_RATIOS = ['9:16', '16:9', '1:1'] as const;
+
+export const CharacterVideoSchema = z.object({
+  character_sheet_url: z.string().url().describe(
+    'Public HTTPS URL of the character reference sheet PNG. Generate one via POST /v1/character/sheet-generate or upload your own.',
+  ),
+  storyboard_url: z.string().url().describe(
+    'Public HTTPS URL of the storyboard panels PNG. Generate one via POST /v1/character/storyboard-generate or upload your own.',
+  ),
+  action_prompt: z.string().min(1).max(2000).optional().describe(
+    'Optional scene/action description sent to Seedance as the primary scene driver (e.g. "Marco the chef in his Brooklyn kitchen at golden hour, takes a bite of fresh bread"). When omitted, the api-v2 server backstops it from the same session\'s storyboard job (script or beats) — without a real scene description Seedance picks the location arbitrarily.',
+  ),
+  duration: z.number().refine(
+    (v) => (CHARACTER_VIDEO_DURATIONS as readonly number[]).includes(v),
+    { message: 'duration must be 5 or 10' },
+  ).optional().default(10).describe('Video duration in seconds. Valid values: 5 or 10. Max 10s — Seedance i2v quality degrades past 10s with multimodal inputs.'),
+  aspect_ratio: z.enum(CHARACTER_VIDEO_RATIOS).optional().default('9:16').describe('Output aspect ratio. Default 9:16.'),
+  generate_audio: z.boolean().optional().default(true).describe('Whether Seedance synthesizes synchronized audio (ambient sounds, breath, etc.).'),
+  session_id: z.string().uuid().optional().describe('Optional UUID linking the three pipeline steps (sheet → storyboard → video) into one wizard session. Same value across calls makes inserts idempotent and lets api-v2 backstop action_prompt from the storyboard step.'),
+  webhook_url: z.string().url().optional().describe('HTTPS URL to receive a callback when the job completes or fails.'),
+});
+
+export type CharacterVideoInput = z.infer<typeof CharacterVideoSchema>;
+
+// ── Text-to-Video (pure prompt, no character / no storyboard) ──────────────
+//
+// Drives Seedance 2.0 text-to-video. The prompt IS the whole creative —
+// style, subject, mood, composition all baked into one string. Used by
+// the "Use prompts" batch-schedule mode and as a one-shot generator
+// for Claude Code / MCP / CLI workflows where the agent already has a
+// fully-formed video prompt.
+
+export const TEXT_TO_VIDEO_DURATIONS = [5, 10, 15] as const;
+export const TEXT_TO_VIDEO_RATIOS = ['9:16', '16:9', '1:1'] as const;
+
+export const TextToVideoSchema = z.object({
+  prompt: z.string().min(20).max(1000).describe(
+    'The full video prompt. Style, subject, mood, composition, lighting, lens, motion — all in one string. Seedance reads this verbatim. Min 20 / max 1000 chars.',
+  ),
+  duration: z.number().refine(
+    (v) => (TEXT_TO_VIDEO_DURATIONS as readonly number[]).includes(v),
+    { message: 'duration must be 5, 10, or 15' },
+  ).optional().default(10).describe('Video duration in seconds. Valid: 5, 10, or 15.'),
+  aspect_ratio: z.enum(TEXT_TO_VIDEO_RATIOS).optional().default('9:16').describe(
+    'Output aspect ratio. 9:16 portrait (TikTok / Reels / Shorts), 1:1 square (feed posts), 16:9 landscape (YouTube / X / LinkedIn). Default 9:16.',
+  ),
+  generate_audio: z.boolean().optional().default(true).describe(
+    'Whether Seedance synthesizes synchronized audio. No extra charge.',
+  ),
+  webhook_url: z.string().url().optional().describe(
+    'HTTPS URL to receive a callback when the job completes or fails.',
+  ),
+  /** When set, on completion the webhook-provider fans the rendered MP4
+   *  out to these Postiz integrations (X, LinkedIn, etc.). Use the IDs
+   *  returned from GET /v1/integrations/postiz/accounts. */
+  postiz_integration_ids: z.array(z.string()).optional().describe(
+    'Postiz integration IDs to auto-publish to once the video is rendered. Get them from GET /v1/integrations/postiz/accounts.',
+  ),
+  /** How to compose the social caption. 'static' = use `caption` verbatim. 'ai' = Claude writes one using `caption_guidance` as bias. */
+  caption_mode: z.enum(['ai', 'static']).optional().default('static').describe(
+    "Caption mode for auto-publish. 'static' (default) uses the literal `caption` string. 'ai' has Claude Opus generate one each time, biased by `caption_guidance`.",
+  ),
+  caption: z.string().min(1).max(2000).optional().describe(
+    'Literal caption text for the social post. Only used when caption_mode is omitted or "static".',
+  ),
+  caption_guidance: z.string().min(1).max(1000).optional().describe(
+    'Tone / hashtag / length hints for the AI caption writer. Only used when caption_mode is "ai".',
+  ),
+});
+
+export type TextToVideoInput = z.infer<typeof TextToVideoSchema>;
+
+// ── Step 1 of 3: Character Sheet generation ─────────────────────────────────
+//
+// POST /v1/character/sheet-generate. Async — returns 202 + job_id.
+// Pricing: 12 credits ($0.12) for actor/reference (1 gpt-image-2 call),
+//          20 credits ($0.20) for description-only (paint portrait + sheet).
+
+export const CharacterSheetSchema = z.object({
+  description: z.string().min(3).max(400).optional().describe(
+    'Free-text description of the character (e.g. "Marco, a 35yo Italian chef with curly black hair, white uniform"). When provided alongside a reference image, it adds context. When provided alone (no actor_slug, no reference_image_url), the worker first paints a portrait via gpt-image-2 text-to-image, then turns it into the sheet.',
+  ),
+  actor_slug: z.string().min(1).optional().describe(
+    'Slug of an actor from the agent-media library. The actor\'s portrait is used as the reference image. List actors via GET /v1/actors. Mutually exclusive with reference_image_url.',
+  ),
+  reference_image_url: z.string().url().optional().describe(
+    'Public HTTPS URL of a portrait/reference image (PNG/JPEG/WebP, <5MB recommended). Mutually exclusive with actor_slug.',
+  ),
+  session_id: z.string().uuid().optional().describe('Optional UUID linking the three pipeline steps. Same value across calls makes inserts idempotent.'),
+}).refine(
+  (v) => Boolean(v.description) || Boolean(v.actor_slug) || Boolean(v.reference_image_url),
+  { message: 'Provide at least one of: description, actor_slug, or reference_image_url' },
+).refine(
+  (v) => !(v.actor_slug && v.reference_image_url),
+  { message: 'actor_slug and reference_image_url are mutually exclusive' },
+);
+
+export type CharacterSheetInput = z.infer<typeof CharacterSheetSchema>;
+
+// ── Step 2 of 3: Storyboard generation ──────────────────────────────────────
+//
+// POST /v1/character/storyboard-generate. Async — returns 202 + job_id.
+// Pricing: 12 credits ($0.12).
+// Provide EITHER `beats` (3-10 short strings, gpt-image-2 paints them as
+// numbered panels) OR `script` (free-text up to 1500 chars; gpt-image-2
+// splits it into 4-6 panels itself).
+
+export const STORYBOARD_RATIOS = ['9:16', '16:9', '1:1'] as const;
+
+export const CharacterStoryboardSchema = z.object({
+  character_sheet_url: z.string().url().describe(
+    'Public HTTPS URL of the character sheet PNG (from /v1/character/sheet-generate). Used as the visual reference so the character stays on-model across panels.',
+  ),
+  beats: z.array(z.string().min(3).max(200)).min(3).max(10).optional().describe(
+    'Ordered list of 3-10 short beat descriptions, one per panel (e.g. ["Marco walks into kitchen", "pulls bread from oven", "takes a bite", "thumbs up"]). Mutually exclusive with `script`.',
+  ),
+  script: z.string().min(10).max(1500).optional().describe(
+    'Free-text script (10-1500 chars). gpt-image-2 splits it into 4-6 sequential panels itself. Mutually exclusive with `beats`.',
+  ),
+  ratio: z.enum(STORYBOARD_RATIOS).optional().default('9:16').describe('Storyboard sheet aspect ratio. Default 9:16.'),
+  session_id: z.string().uuid().optional().describe('Optional UUID linking the three pipeline steps.'),
+}).refine(
+  (v) => Boolean(v.beats) !== Boolean(v.script),
+  { message: 'Provide exactly one of: beats or script' },
+);
+
+export type CharacterStoryboardInput = z.infer<typeof CharacterStoryboardSchema>;
+
+// ── Helper: AI-suggested beats ─────────────────────────────────────────────
+//
+// POST /v1/character/storyboard-suggest. Sync — returns immediately.
+// No credits. Returns 3 distinct beat-sequence options for the wizard.
+
+export const StoryboardSuggestSchema = z.object({
+  actor_slug: z.string().min(1).optional().describe('Library actor whose persona drives the suggestions. Mutually exclusive with character_description.'),
+  character_description: z.string().min(1).max(200).optional().describe('1-200 char description of the character. Mutually exclusive with actor_slug.'),
+  vibe: z.string().min(1).max(200).optional().describe('Optional vibe note that biases all 3 options (e.g. "wholesome", "chaotic", "cinematic").'),
+  duration: z.number().refine((v) => v === 5 || v === 10, { message: 'duration must be 5 or 10' }).optional().default(10),
+  n_panels: z.number().int().min(4).max(10).optional().default(6).describe('Panels per option (the storyboard sheet will be an n_panels-panel grid).'),
+}).refine(
+  (v) => Boolean(v.actor_slug) !== Boolean(v.character_description),
+  { message: 'Provide exactly one of: actor_slug or character_description' },
+);
+
+export type StoryboardSuggestInput = z.infer<typeof StoryboardSuggestSchema>;