Anorak, founded by former Hutch executives, raises $5M pre-seed from Play Ventures, Backed VC, Supercell and MTG for interactive generative video games

• No named model, version or public product. Anorak posts weekly "playable experiments" on its blog and says it is building its own inference technology to make generative video "faster, cheaper and more controllable" (Tech.eu).
• Demos shown so far: dress-up, boxing, and an alien character that looks at user-uploaded images and tries to resell them (Mobilegamer.biz).
• Planned creator tool: the company says it will deliberately not be a chatbot that turns a prompt into a game.

Nothing about Anorak's technology is documented. There is no paper, model card, base model, latency or frame-rate figure, or evaluation, and trade coverage confirms that no technical metrics were disclosed (Mobilegamer.biz).

• Stated approach: optimise inference so the product can shift step by step from "pre-generated" content to real-time generation (Tech.eu). That framing suggests a hybrid pipeline in the near term, though this is an inference, not something the company has stated.
• Where the work happens: the AI lab is in Chengdu and is led by co-founder Tak Fung (Chief of AI).
• Reproducibility: none is possible at this stage.

• $5M pre-seed, co-led by Play Ventures and Backed VC, with Supercell and MTG: announced by the company and repeated by independent trade press (Mobilegamer.biz; Runtime Wire). No filing confirms it.
• Valuation: not disclosed.
• Zynga co-founder Justin Waldron as an angel investor and David Amor as COO: reported only by Mobilegamer.biz.
• "Technical breakthroughs … in the past 12 months": this comes from Alex Brunicki, founding partner at Backed VC, one of the round's lead investors. It is investor commentary with no published evidence behind it (Tech.eu).
• Harri Manninen's praise of the team: he is founding partner at Play Ventures, the other lead investor.
• 20-person team: reported by the company.

Interactive generative video sits next to research on "world models", AI systems that generate an explorable environment frame by frame in response to user input. Google DeepMind's Genie 3 (August 2025) generated interactive worlds at 720p and 24 frames per second that stayed consistent for minutes. It was released only as a limited research preview (The Decoder; Neowin).

Anorak's angle is different. Rather than building a general-purpose world simulator, it is designing new game mechanics around short, specific interactions. Other AI-native game studios are also raising pre-seed money, for example Surge Games' $3M round.

• Known limits of the field: world models still struggle to stay coherent over long sessions, can handle only a narrow range of agent actions, and find multiple interacting agents hard to model (BD Tech Talks). Anorak has published nothing showing how it handles these problems.
• Little scrutiny so far: coverage of the round has been uncritical, with no skeptical analysis (Mobilegamer.biz).
• Unaddressed safety risks: the image-reselling demo and the planned user-generated platform raise content-moderation and IP questions that the reviewed sources do not discuss.

• Strategic interest from incumbents: Supercell and MTG investing points to interest in generative formats, not just generative tools. MTG is backing its own former executives.
• The cost bet: the thesis depends on inference costs falling far enough that per-player real-time video can pay for itself with niche audiences. The company has disclosed no unit economics.
• What developers can watch: the weekly blog experiments and the early-2027 creator platform will be the first things anyone can actually test.
• Taste as the moat: Rutland argues that when making content gets cheap, value moves to curation. If that holds, distribution and community ownership matter more than the model itself (Tech.eu).