It still baffling to me that people still think AI video is rolling a dice. We're so beyond that. Jonathan Degueldre from South Africa, clearly shows how you can direct using very low poly assets, and by combining high quality assets, you can scope a car commercial for pennies.... Car folks know this is expensive to shoot IRL, so anything you can do pre-viz or even to remove a location shoot all together, is worth exploring.show more

You still have the face to map as well. What I also find is that character consistency is still an issue in many cases so multiple generations are still required. So technically, we still need to roll the dice. The stats I have for Seedance 2.0-2.5 is that we get exactly what we want 60% or the time. This is based on thousands of generations.

So where's this Joanthan. Interested in the process.

Would have been awesome if the dog was facing the other way

Beyond? Funny how you present a video with an obvious error, the dog sitting backwards.. 😄 then, this is on shot, try to craft a scene and you'll quickly realize we're not beyond it at all - this is just the wrong approach ;)

lol sure, what these people never mention is how many generations they had to do to get this result, and even here the dog is facing the other way. Don't get me wrong, it is still amazing what AI can do now, but stop pretending everything is peachy.

Car folks ( in Adv ) will also tell you one critical things for cars: models still change details. the grills, some of the lines, etc. you need to do some post processing, image there's ways, so you can do for video.

While the visual fidelity has improved orders of magnitude, the workflow is still fundamentally flawed and is in fact still based on predictive math rather than true physical/temporal understanding. We need a different toolset to get to real intentionality and true control.

I know even better idea than this

The low-poly blocking step is what clicks for me, it's basically previz that turns straight into final shots. Curious how it holds up once a client asks for revisions though. Keeping the exact same car, paint and reflections across 20 shots is where I'd expect it to get messy.

These camera movement lack alot of dynamics, they look very lazy. I'd give it to control only... you still need a good prompt for proper output, in lazy input, lazy output.

Holy shit. That looks so good

Seems like a lot of extra work when there is faster methods.

Dogs facing the wrong way..

Try regenerating this now using the same prompt, same reference images but remove the reference video and show the result. Not throwing shade but I bet it won’t be very far off and might have a better camera movement.

Low poly assets scoping a car commercial for pennies. The dice story was always cope. Direction is the scarce part.

It's an incredible workflow. I havent tried it yet, but will. It's similar to the "oldschool" workflow of previz and animatics before final production.

I want the boring numbers. Did low poly blockout plus AI beat a shoot on cost and revisions? If so, the dice crowd owes an apology.

Ping

Should have had the dog looking out the open window then sweep the cam through the window. Ire interesting to see the dogs face than its back

This is incredible, I think people are going to want something that “feels real” in the sense that AI is now able to do the old style of car ads. What’s the new style version ?

I'm seriously considering doing workspace renders for my next peice. Houdini sims, keyframes - the works!