a lot of people asked for the code for the generative maps, so i'm open-sourcing it in the hopes that y'all can take it and vibe something cool!

this is why our startup culture is the best in the world; nobody is telling us to do this to ourselves, we do it willingly. we enjoy it. we're insane, broken people and that's why we're the richest nation in the world

a lot of Americans getting baited by this, but you should revel in the fact that our society want to fly across the world and drive for 12hrs for the sake of it. this is absolutely unhinged behavior that signifies a society built to brutalize themselves in the pursuit of a goal.

i just see a whole world of filmmakers focusing on their shots in terms of composition/voice, then going to a good image model like nano banana in post to apply a visual universe on top of that. obvi there's 100x more to a final product than that, but there's so much potential

god i really want to do this one over. i keep seeing it and seeing exactly how i'd improve it. needs an intermediary step to strip the layout from the original pane to a layout sketch, then go to the desired style is anyone doing experiments on image to video? this feels

prompt: Arrange three (3) of these 5" santoku knives, three (3) of the 3.5" paring knives on a cutting block, professional photo for fine dining magazine. Shot should be directly overhead, no perspective. Arrange them left to right.

underrated usage of nano banana pro for me is taking different products and arranging them in a unified way so that you can really get a sense of them apples-to-apples. product photography is all over the place w/ lighting/quality/angles. this is almost as good as seeing them

once i had a style i liked i just gave it new product photos from google and had it make variations

nano banana pro vouchers for theme park merch. we wanted to gift something physical, so we'll print these out on nice paper and wrap them.

100-ish frames rendered here. i didn't look too closely at pricing but I ballparked it at 25c per (probably wrong) and need to double-check it on gcp. So still very expensive, but if you were able to do it well, it could be worth it for the effects you can pull off.

i tweaked this prompt forward a bit, so it got ~better later in the render.

the fidelity to the layout of the source frames is very good. the missing bit is getting it to *really* consider the style and details of the previous frame's re-render

frame-by-frame transformations with nano banana pro! it's not quite there yet but it'd probably be doable with the right prompting and curation. @GoogleDeepMind @Google @NanoBanana

@GoogleDeepMind larger example from NB v1 but you can see the difference in layout consistency between frames. lots of shifting, glitching, added details

nano banana pro: "make it a little better" x 14 feedback loops i'm gobsmacked by how strong this is; it seems to track details at a very low level and fixes consistency issues without my intervention. very very very good model. @GoogleDeepMind

a basic test, but having Nano Banana Pro add style-appropriate teenage mutant ninja turtles to a given image. the critical thing that nothing else has got right is that all 4 turtles are there, 2.5 usually duplicated the characters or left one out. @GoogleDeepMind

here we go again dot gif

i'm about to lose a week of my life again, thanks demis

good luck out there humans

this necromancy idea has been obvious for a while now, ever since voice cloning got good. but to actually make it is perverse. that is not your mom talking, it's a ghost, and not hers, stuffed into her form. you made her into a puppet, and you don't know where the hand came from.

humanity reckoning with the fact that we can now make wraiths. if you're not building existentially good applications with AI today because you're morally opposed to it, you're leaving the playing field clear for companies like this, who think black mirror was a blueprint.