Gemini gets faster image and video models

Google wants AI media to move faster

Google is cutting the wait time for AI image generation to 4 seconds.

The company launched Nano Banana 2 Lite, a cheaper and faster Gemini image model, alongside Gemini Omni Flash for video generation and conversational editing. Nano Banana 2 Lite costs $0.034 per 1K image, while Omni Flash is priced at $0.10 per second of video output.

This looks aimed at developers who care less about perfect output on the first try and more about fast iteration inside real products. The useful detail is that Google is pushing both models across AI Studio, the Gemini API, enterprise tools, and consumer surfaces like Search, Photos, Flow, and Ads.

The pairing matters because image generation can now feed directly into short video workflows, with up to three sequential edits kept in session context. That could make lightweight creative apps easier to build, especially for commerce, design, and social content teams working through drafts on a laptop screen.

The bigger test is whether users accept the limits once the demos turn into daily work.

AWS is selling hands-on AI execution

AWS says it will spend $1 billion to put forward deployed AI engineers directly inside customer teams.

The new AWS Forward Deployed Engineering group will embed thousands of experts with companies to build agentic AI systems in production. Amazon says the goal is to move deployments from months to days, with customers including the NFL, NBA, Allen Institute, Ricoh, Cox Automotive, and Southwest Airlines.

This is Amazon admitting that many companies do not just need models, APIs, or dashboards. They need people in the room, looking at the same screens, helping turn messy internal processes into working systems.

Speed is the sell. The stronger claim is self-sufficiency, because AWS says customers should leave with runbooks, knowledge graphs, internal champions, and systems they can operate themselves.

This could make AWS more valuable to regulated industries where AI pilots often stall in security reviews and governance meetings. It also puts pressure on cloud rivals to offer more than infrastructure when customers ask for production results.

The question is how many companies can really become independent after the AWS team leaves.

Claude Science is built for messy research

Anthropic is taking Claude deeper into the daily grind of science.

Claude Science is now in beta for Pro, Max, Team, and Enterprise users on macOS and Linux. It brings literature review, code, figures, manuscripts, specialist agents, and compute access into one app, with more than 60 curated skills and connectors.

The interesting part is the audit trail, because scientific AI tools are only useful when researchers can check exactly how a result was made. I also think Anthropic is aiming at a real pain point here: scientists already live inside PubMed, Jupyter, R, terminals, cluster jobs, and half-finished figures at 11 p.m.

The early examples are strong, especially the Allen Institute workflow that helped produce about 10 long reviews with reviewer agents checking citations. Labs may start treating AI less like a chat assistant and more like a research operating layer.

The open question is whether scientists trust the reviewer agent as much as the main one.

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.