Posts by Date – Israeli AI Community
10.12
Friends, after the initial excitement, we tested Google's new model (based on Gemini 3) to understand if it's a gimmick or a tool. The conclusion? The new toolkit solves most of the problems that have been frustrating us until now with AI. Let's start with Style Reference, the feature we've been waiting for. Do you have a picture with perfect lighting or a unique brand illustration style? You can upload it as a "reference." The model will not copy the content, but rather "learn" the style (color palette, mood, texture) and apply it to any new idea you have. This is a game-changer for maintaining a consistent design language.
Surgical editing (Inpainting) - Did you create an amazing image but there's an unnecessary potted plant in the background? A hand turned out weird? No need to regenerate and hope for the best. You can mark a specific area and ask the model to fix only that. Changing a suit to a t-shirt or removing an object – it all happens in seconds without compromising the composition.
The days of models getting "lost" with complex prompts are over. The "Nano Banana" demonstrates impressive spatial understanding: ask for "three people in a circle, the one on the right holding an iPad and the one on the left pointing to the sky" – and it will place them exactly there. Less trial and error, more results.
Character Consistency. The big news: the ability to generate a character (presenter) and run with it through a whole series of images in different situations, thanks to the ability to remember up to 14 reference images.
Hebrew without gibberish – The model knows how to integrate text into images, in various fonts, and in full, standard Hebrew. Signs, invitations, and banners.
