First look into upcoming AI-generated Video Overviews from Google

With Google I/O just days away, an overlooked experimental project called Illuminate is quietly revealing what could become one of the biggest AI media updates yet. While Illuminate has been around for some time, offering audio overviews of research papers, a broader version was recently rolled out, exposing a homepage that lists these AI-generated summaries. However, most of its new capabilities remain hidden behind feature flags.

Previously, testers discovered that Illuminate allows the creation of customizable audio overviews, letting users select hosts, modify prompts, or even override the entire conversation. Now, something much larger seems to be emerging. While still hidden, the interface hints at support for audio overviews not only from research papers but also classic books like Frankenstein, Alice in Wonderland, and The Great Gatsby, following the same generation format. Experimental controls such as an Edit button, caption toggles, and even image generation for cover photos are also present but inaccessible to the general public.

The most compelling discovery lies at the top of the page: a new section called Sparks, marked as Early Preview. Its description reads, “Imagine any question could be instantly transformed into a short video, 100% AI-generated.” Below this are samples of vertical videos, typically one to three minutes long, covering various topics. While the generation tool isn’t publicly available and seems restricted to internal Google accounts, the phrase “100% AI-generated” suggests these videos are produced by a single model capable of generating synchronised video and audio from input, eliminating the need for separate pipelines.

Although we can’t confirm the exact model behind it, the high quality of the results raises the possibility of a connection to Veo 3 or a multi-modal Gemini (Ultra?) variant. Moreover, since the NotebookLM video overview feature is confirmed to involve two AI hosts and shares a similar format, it’s very likely that the same tech stack is behind both. If so, NotebookLM could soon support Video Overviews based on uploaded sources, delivered as fully generated conversational clips.

While most of this remains speculative and hidden behind feature flags, the glimpse into Sparks offers a strong indication of where Google is headed: towards seamless, multi-modal content generation from a single prompt.