Take a first look at the upcoming video overviews in front of Google I/O

With Google I/O just a few days away, an overlooked experimental project called Illuminate quietly reveals, which could become one of the largest KI media updates. While Illuminate has been there for some time and offers audio overviews through research work, a broader version was recently introduced, in which a homepage was uncovered on which these summaries are listed by AI generated. However, most of his new skills remain hidden behind Feate Flags.

Previously, testers found that Illuminate enables the creation of adaptable audio overviews to select the users, change input requests or even overwrite the entire conversation. Now something bigger seems to be performing. The interface is still hidden and indicates the support for audio overviews not only from research work, but also classic books such as FrankensteinPresent Alice in WonderlandAnd The big gatsbyfollows the same format of the same generation. Experimental controls such as processing buttons, captions and even image generation for cover photos are also available, but are not open to the general public.

The most convincing discovery is on the top of the page: a new section called Sparks, which is marked as an early preview. In his description it says: “Imagine that a question could be converted into a short video immediately, which is 100% generated with AI.In the following you will find examples of vertical videos that are usually for one to three minutes and cover different topics. While the generation tool is not publicly available and seems to be limited to internal Google accounts, the expression “100% ai-generated” suggests that these videos are created by a single model that can generate synchronization videos and audio from input in order to eliminate the necessities for separate pipines.

Although we cannot confirm the exact model behind it, the high quality of the results increases the possibility of a connection to VEO 3 or a multimodal Gemini variant (Ultra?). Since the “NotebooKlm video overview” function is confirmed for two AI hosts and a similar format, it is very likely that the same tech stack is behind both. In this case, NotebooKlm could soon support video overviews based on uploaded sources that are delivered as fully generated conversation clips.

While most of this remains speculative and hidden behind feature flags, the view in Sparks offers a strong indication of where Google leads: in the direction of seamless, multimodal generation of content from a single input request.