category:
  id: video
  label: Video
  description: Video generation
  defaultPlatform: runway
  defaultMode: detailed

# Currency policy (1.16.0): every entry touched by a fact-verification pass
# carries `verifiedAt` (the date its syntax facts were checked against the
# platform's current docs) and, where useful, `source`. Entries without a
# date have not been re-verified recently — treat their hints as suspect.

platforms:
  - id: runway
    label: Runway (Gen-4.5)
    description: Gen-4.5 Turbo flagship — best prompt adherence, character References, Act-Two motion capture; multi-model platform that also hosts Kling, Veo, Seedance, Nano Banana Pro
    verifiedAt: "2026-10-05"
    source: runway.com
    syntaxHints:
      - natural-language scene description (no flags)
      - References — upload a character/object image for consistency across generations
      - Act-Two — apply recorded human motion to a character reference
      - camera controls (pan, tilt, zoom, orbit) and motion brush
      - up to 4K; extend for longer clips
      - editor-first workflow (timeline, keyframing)

  - id: pika
    label: Pika (2.2)
    description: Fast social-first stylized video — Pikaframes, Pikaswaps, Pikadditions, Pikaformance lip-sync
    verifiedAt: "2026-10-05"
    syntaxHints:
      - Pikaframes — first/last keyframe interpolation
      - Pikaswaps (replace objects), Pikadditions (insert elements), Pikaffects (stylized effects)
      - Pikaformance — fast lip-synced talking images
      - action-focused, concise prompts; fast iteration
      - up to 1080p

  - id: kling
    label: Kling (3.0 / Omni)
    description: Kuaishou flagship — best physics and complex motion, native 4K/60fps; Omni adds native audio + 5-language lip-sync
    verifiedAt: "2026-10-05"
    syntaxHints:
      - natural-language cinematic prose
      - native 4K at 60fps (Kling 3.0)
      - Omni variant — native audio, lip-synced dialogue in five languages
      - multi-shot storyboards (up to 6 shots per clip)
      - strong fluid dynamics, dance, sports, action
      - extend to ~60s

  - id: luma
    label: Luma Dream Machine (Ray 2)
    description: Fast, keyframe control, natural camera language
    verifiedAt: "2026-10-05"
    syntaxHints:
      - natural language
      - keyframes / start-end frames
      - camera motion (orbit, push in)

  - id: minimax
    label: Minimax / Hailuo (2.3)
    description: Expressive character motion, best value per second, native audio (Hailuo 2.3)
    verifiedAt: "2026-10-05"
    syntaxHints:
      - natural language
      - character animation and expressions
      - native audio

  - id: veo
    label: Google Veo (3.1)
    description: DeepMind flagship — 4K, native/spatial audio, strongest prompt adherence; Veo 3.5 adds camera-control API and lip-synced dialogue
    verifiedAt: "2026-10-05"
    source: ai.google.dev
    syntaxHints:
      - natural-language cinematic description
      - tiers: Standard / Fast / Lite (Lite = 720p/1080p volume tier)
      - 4K (Standard/Fast); 8s clips, extendable
      - native audio; spatial audio on Standard
      - legible text-in-video while the camera moves
      - reference images and start/end frame control

  - id: seedance
    label: Seedance (2.0 / 2.5)
    description: ByteDance flagship — 20s clips, complex human motion, multi-reference input, product/logo consistency
    verifiedAt: "2026-10-05"
    syntaxHints:
      - long-form natural-language prose (action + camera + lighting + mood)
      - up to 20 seconds per clip
      - "aspect ratios: 16:9, 9:16, 1:1, 4:3, 3:4, custom"
      - up to 12 reference files (images, video, audio) for identity and brand consistency
      - native audio-video sync
      - multi-shot storyboarding
      - strong on dance, sports, action, product close-ups

  - id: wan
    label: Wan (2.6)
    description: Open source, versatile, image-to-video and text-to-video
    syntaxHints:
      - natural language
      - open source
      - image-to-video
      - text-to-video

  - id: heygen
    label: HeyGen
    description: AI avatar videos, talking heads
    syntaxHints:
      - avatar selection
      - script input
      - voice pairing
      - gestures
      - background

  - id: synthesia
    label: Synthesia
    description: Enterprise AI avatar videos
    syntaxHints:
      - avatar selection
      - script input
      - multi-language
      - brand templates
      - slides

  - id: cogvideox
    label: CogVideoX
    description: Open weights, high quality text-to-video
    syntaxHints:
      - natural language
      - open weights
      - text-to-video
      - image-to-video
      - detailed descriptions

  - id: higgsfield
    label: Higgsfield (multi-model platform)
    description: Multi-model video platform via MCP / CLI / Skills. Routes one natural-language prompt to Cinema Studio (Higgsfield-original), Veo 3.1, Kling 3.0, WAN 2.6, or Seedance 2.0. Sora 2 dropped after OpenAI shut the Videos API on 2026-09-24. No API key — auth via Higgsfield account.
    verifiedAt: "2026-10-05"
    syntaxHints:
      - "models: cinema-studio | veo-3.1 | kling-3.0 | wan-2.6 | seedance-2.0"
      - long-form natural-language prose (camera move + composition + lighting + mood)
      - duration up to 15s
      - image-to-video from uploaded UUID
      - Soul ID for character consistency across clips
      - "modes: Cinema Studio, UGC Factory, Lipsync Studio"