Developer Yang Liu (@delexw) has released Kiteva, a personal creative tool that transforms raw ideas into short motion-graphics films using local AI agents. The project, which debuted on Hacker News on October 17, 2026, runs exclusively on macOS and leverages Claude Opus 5.5 or GPT-6.1 Sol to handle the heavy lifting of creative direction and rendering. Unlike cloud-based video generators, Kiteva operates as a local desktop application, giving users full control over the asset pipeline from brief to final MP4 delivery.

The Workflow: From Idea to Render

Kiteva structures the creative process into distinct stages: idea ingestion, brief generation, motion graphic combination, previewing, and final film rendering. The system uses a small model to analyze user-provided files and website screenshots, generating a structured brief that guides the main creative model. For the actual composition, the tool draws from a catalog of 136 Motion Graphics and 46 Motion patterns. An evaluator named JEV, provided by Typesafe, scores potential directions, allowing the user to select from 1 to 5 distinct visual approaches before committing to a full render.

Technical Stack and Local Control

The application is built on a robust local stack, requiring Node.js 24+, Chrome or Chromium for headless browsing, and FFmpeg for video processing. The default renderer is HyperFrames 0.8.144, with Multiplane 0.3.10 available as a trial and Remotion 4.x as a fallback. This setup ensures that visual generation is code-based rather than relying on external stock footage or AI image generators. The tool also integrates local narration via Kokoro or Pocket TTS, keeping the entire production loop on the user’s machine. Governance is handled via a 'Governance report' that tracks every API call, token count, and cost estimate, providing transparency on resource usage.

Cost and Licensing Realities

Liu shared specific cost data from three development projects, which ranged from $16.46 to $37.42 in API-equivalent model usage. The majority of these costs were attributed to Claude Opus 5.5 generating previews and films, while preparation and review tasks using Claude Haiku 5.5 cost mere cents. It is crucial to note that Kiteva is licensed under the Kiteva Non-Commercial License. While the source is available and based on MIT terms, it explicitly prohibits commercial use, such as offering it as a paid service or using the output for paid client work without separate permission. Internal hackathons and personal projects remain free to use.

Key Takeaways

  • Kiteva is a macOS-only tool that uses Claude Opus 5.5 or GPT-6.1 Sol to generate motion graphics from text briefs.
  • The project includes a catalog of 136 Motion Graphics and uses a local evaluator (JEV) to score creative directions.
  • Rendering is handled locally via HyperFrames, Multiplane, or Remotion, avoiding external stock footage or AI image generation.
  • The license is non-commercial, meaning users cannot sell the tool or use its outputs for paid client work without permission.
  • Estimated costs for a single film project ranged from $16.46 to $37.42 in API usage, with most costs driven by the primary LLM.

The Bottom Line

Kiteva is a compelling proof-of-concept for local, code-driven creative AI, but its non-commercial license and high token costs limit its immediate utility for professional studios. It serves best as a powerful personal tool for engineers and hobbyists who value transparency and local control over cloud convenience.