Top story
AI video finally gets a control surface developers can measure
https://blog.google/innovation-and-ai/technology/developers-tools/build-with-gemini-omni-1-1-flash/
Gemini Omni 1.1 Flash moves into the Gemini API with changes that matter more than another highlight reel. Google says it can condition on ten seconds of prior video instead of the last second, extend a scene in ten-second steps to forty seconds total, and generate the transition between specified first and last frames. A 360p draft mode is claimed to run up to 60% faster at one-third the cost of 720p, while final output can be upscaled to 1080p or 4K.
The useful launch is the control surface, not the “production-ready” label. In the 43-comment HN thread, simonw connected Google’s continued video investment to possible world-model ambitions; that is an interesting hypothesis, not a reason to choose an API. Media-tool teams should instead measure continuity across repeated extensions, identity drift, and cost per accepted clip. Google provides no independent quality comparison or failure rate here, so I would trial Omni on storyboards and draft pipelines before letting it drive an unattended production step.
Two signals worth keeping
Small models are changing product economics. One personalized-news experiment reportedly fell from roughly $1 with a Sonnet-class model to about $0.10 with Luna. HN user hartator argued that less-capable models may only be a hardware transition. I would route by error cost: cheap models for repetitive throughput, stronger models and workload-specific evals for consequential decisions.
A governed CMS API beats decorative AI. Wagtail 8.0 previews its v3 API with more than 50 typed admin operations, an OpenAPI schema, Markdown, bearer authentication, and existing permissions. That is a sensible agent boundary: automation through an inspectable interface instead of generation buttons everywhere. Because the API remains a preview ahead of 8.1, test the workflow on staging first.
— Tin