Hypit lets performance marketers rebuild a competitor's best-performing ad around their own product, then generate a large batch of variants with different people, scenes, and languages. You give a coding agent the reference ad and a product photo. The agent reconstructs the shots and swaps in your product, keeping the structure that already worked. This launch targets the same job as subscription ad tools such as Arcads, Higgsfield, and Creatify: more creative tests, without rebuilding each cut by hand. The workflow installs as a skill for agents such as Claude Code and Codex. Hypit writes the ad as an editable composition of footage, dialogue, captions, B-roll, and effects, timed to words so a rewritten line retimes the edit. Public docs describe starting from a Meta Ad Library winner, swapping the product, and shipping new hooks while the body stays fixed. Sample clones in the repository reuse one format across hosts, products, and languages. An 18-second podcast-style spot is listed at $1.07 in model cost, and a 26-second street interview at $1.09. The open-source core has no seat fee and can be forked or self-hosted. Model generation is billed separately, through your own API keys or Hypit's credits, where annual plans are listed from $9.90 a month. Captions and motion graphics can also be rendered in code with no generation call. Alvin Chen and cofounder Kashorin are behind the Hong Kong company. The public repository has roughly 29,000 stars.