Google just shipped two new AI models over the weekend. Nano Banana 2 Lite is a cheap image generator aimed at beating OpenAI on price. Gemini Omni Flash is a video model that people are calling the Nano Banana of video. The Lite version is supposed to understand what you actually mean when you describe an image, not just spit out generic results. The video one is generating clips that look shockingly good for a Flash tier model. Both are available to try right now. If you have been waiting for a sign that the cheap end of Google AI is finally worth using, this is probably it.

Heads up if you run a business on AI right now. Austria is pushing the EU to host Anthropic inside Europe, basically as a hedge against the US cutting off access to top models. The EU Commission is already warning that any export controls should not be discriminatory. Translation for non-technical founders: the model you use today might not be available in every market tomorrow, and pricing or features could shift based on where you operate. Worth keeping in mind before you build your whole workflow around one provider.

ByteDance just pushed its Seedance video model to handle 30 second clips, which sounds small until you remember most AI video tools cap out around 8 to 10 seconds. Longer coherent clips means way fewer Frankenstein edits when you want a real ad, a product demo, or a 30 second social hook. The AI video race is not just about quality anymore, it is about length.

Adobe is buying Topaz Labs, the popular AI photo and video enhancement tools. If you use Topaz for upscaling, denoising, or sharpening footage, expect it to fold into Creative Cloud over the next year. Probably a price bump, but also a lot more polish. Either way, the big creative suites are eating every serious indie AI tool right now.

Google just launched the Interactions API, billed as the primary interface for talking to Gemini models and agents. Think of it as a unified way to send messages, handle long running tasks, and stream tool calls back. One endpoint, no glue code. If you are building agent workflows on Gemini this cuts a lot of plumbing. Worth a look if you are tired of stitching SDKs together.