A dark-fantasy visual anthology. Inspired by the atmosphere and world-building of 1980s fantasy cinema. Original music and imagery combined into seven chapters.
https://t.co/t7QCGZtcA2
Tools used: @midjourney, @Kling_ai, Veo 3.1, Sora 2, @udiomusic, After Effects, Premiere.
I built the AI Filmmaker Index (@filmmakerindex) as a curated directory of AI directors, studios and motion artists, because hiring for this kind of work comes down to taste more than tools, since everyone has access to the same models, and the quickest way to judge taste is to look at finished films side by side before you write the brief.
Brands and agencies can browse it and post a brief for free, filmmakers list for free once their work is approved, and the festival page keeps more than a hundred open AI film calls in one place, sorted by deadline.
@JudeFawley27@tupacabra Cool images are an end in themselves, and hard to create, even in an age of AI. 99.99% of stories made by the experts in Hollywood are virtually unwatchable slop themselves. I'm ready for the disruption of that exclusive fiefdom.
@colin_corr@tupacabra Right, but it's still just copying the structural device of a trailer from a couple years ago. It's not a rights argument I'm making, just creative ethics.
ByteDance Seed released Seedream 5.5 on Monday, and it's already in @BytePlusGlobal's playground. The feature we'd use first is native RGBA output, which means the model can generate an image with an alpha channel, partial transparency included, so an element like smoke, glass or a cut-out product arrives ready to composite instead of having to be keyed or rotoscoped off a background. The other claim worth testing is that it leaves the unedited parts of an image unchanged at the pixel level across as many as 100 rounds of edits, which matters on commercial work, because the frame a client has already approved is the one you can least afford to have drift while you fix a single note on it.
@higgsfield has announced Miracle, which it describes as a way to edit an existing clip while it plays rather than generate a new one, so you type a change, like a new color on a single object, a different time of day, or one thing removed from the shot, and watch it land without a render queue.
It's only in early testing with selected partners, so we'll judge it once we can use it, but editing is the right starting point for commercial work, because most notes on an approved shot are about one element in it, and regenerating the whole shot to fix that element risks changing the parts the client already signed off on.
3D will have a place in the future of images for a while, but the future of everything feels fully generative. The utility of 3D for AI filmmaking and VFX will be niche and then disappear, and gaming will likely follow.
Not that this isn't amazing. If you'd told me in 2024 that in a two short years we'd be able to instantly generate on demand, complex models for Blender for essentially free, I'd have cried tears of joy. But then diffusion models leapfrogged the very notion of mathematically representing reality with 3D software.
It feels like the future of all media will still be generative, and the age of polygons is coming to an end.
3D models will be useful for planning out shoots or giving video models reference for motion, blocking scenes and scale reference, all to be fed to diffusion models for finishing.
Real world lighting physics and realistic, cinematic imagery are already far better in diffusion models like Seedance and Midjourney than in the best renderers in the world, and the rate of improvement continues to be exponential.
We will have full realism, great performances and incredibly minute visual control fully generated live by 2030. We will be able to direct the scene verbally and run it as many times as we want.
We will just describe what we want, and models will change it in real time.
Gemini 4 Argon is now globally available.
Built to solve your hardest problems from start to finish, the model brings frontier capabilities across coding, enterprise workflows, and cyber defense.
We’ve built a surgical robot to automate key steps in the process of receiving a Neuralink implant to promote safety, reliability, and scalability.
Learn more about the role of the robot and its future potential from the engineers helping develop it.