Features
Natural language video search across entire libraries
Multimodal indexing (visual, audio, speech) at ~60x real-time speed
Any-to-any retrieval via Marengo model (video, image, text, audio)
Video-to-text generation via Pegasus model (analyze, describe, summarize)
Automatic scene segmentation and temporal boundary detection
Compliance and brand safety detection with explainable AI
Highlight creation from long-form video based on natural language queries
Embed API for generating multimodal embeddings
Search API with SOTA composite accuracy (78.5% composite on Marengo)
MCP integrations for developer workflows
SOC 2 Type II certified, encrypted data handling
SDKs for Python and other languages
Integration with AWS Bedrock and Snowflake AI Data Cloud
Index up to 10k+ hours of video per day (Developer tier)
#1 on Video-MME benchmark for video understanding
Ray 3.2: frame-by-frame video control and continuity across cuts
Uni-1.1: brand intelligence with style and character references
Luma Skills: reusable creative workflows for consistent output
Modify Video: video-to-video editing
Reframe: outpaint images and reframe videos
Camera Motion Concepts: learnable camera controls
Automatic continuity across video cuts
Cinematic quality finishing
Agent-driven research and refinement across video, image, audio, text
Team members and admins management
Fast variant generation: ten directions before standup
API access for Ray 3.2 and Uni-1.1
Third-party model integration (Veo 3.1, Kling Omni, GPT Image 2)
Image generation with GPT Image 2, Nano Banana, Seedream
Audio generation: ElevenLabs TTS, SFX, music
