ai& launches video generation service built on its Japan infrastructure

ai& has launched ai& video, a new service that turns text, photos and existing clips into generated video in as little as 12 seconds on the company’s infrastructure in Japan. The launch expands ai& inference into video and gives enterprises a way to build custom models from their own licensed assets.

Why it matters: - ai& video extends the company’s inference platform from language and research use cases into AI video generation. - The service is built for creators and enterprises that want faster iteration on scenes, camera moves and visual concepts without long waits or high generation costs. - Enterprise customers can train dedicated models on assets they already have rights to use, which can help preserve brand identity and visual style.

What happened: - ai& launched ai& video on October 22, 2026. - The service generates new footage from text, photos and existing video clips. - Users can describe a scene in Japanese, animate a single photo, or continue an existing clip. - ai& says video generation can take as little as 12 seconds on its own infrastructure in Japan.

The details: - ai& video is available through a web browser interface and an API. - The browser tool lets users enter text or upload materials to generate videos directly. - The API lets companies plug video generation into production tools or their own applications. - Pricing is pay-as-you-go and varies by model, with no monthly subscription required. - On supported models, ai& video generates audio alongside video. - The service supports additional training using characters, footage and other licensed assets. - That training produces a dedicated model available only to the enterprise customer that trained it. - Additional training runs in a dedicated environment that other users cannot access. - Materials uploaded for standard video generation are not used to train models. - All video generation runs on ai& infrastructure in Japan. - An in-house moderation system is designed to reduce inappropriate generation and potential intellectual property infringement.

Between the lines: - ai& is positioning infrastructure efficiency as the product edge, not just the model itself. - The company says video is computationally intensive because models must keep people and objects moving coherently over time. - ai& says it tunes inference for specific model-and-chip combinations and uses runtime software to make full use of NVIDIA GPUs and processors from other vendors. - The company says that hardware selection plus in-house optimization delivers high throughput at low cost. - Faster generation can matter because creators often want to test multiple ideas after seeing the first clip. - Shimpei Hara, co-founder and president of ai&, said the company wants to build an environment where waiting time and cost do not stop experimentation.

What's next: - ai& plans to add support for new video generation models. - The company also plans to keep improving generation speed and processing efficiency. - ai& says it will broaden the range of models and domains available through ai& inference as AI expands into new forms of information and expression. - The company directs users to its technical blog and the ai& video service for more information.

The bottom line: - ai& is using its own compute stack in Japan to make video generation faster, cheaper and more customizable for both creators and enterprises.