I saw an example on Twitter recently of very well done text to video renders that look as good as Pixar animation. But these tools work in minutes apparently. How does it work when Pixar needs supercomputers to render every single frame in their movies and it’s very time consuming?
In: Technology
They are using a totally different method of rendering. Pixar would render this by tracing every path of light for every pixel a hundred times. And they would probably do it on CPUs which aren’t optimized for this because they want greater precision than a GPU can give. They also simulate materials, physics, water, air/fog, transparency, and after effects.
Meanwhile rendering video using AI is all done on GPUs. Also, it has no lighting or physics simulation. It has no ray tracing. It is guessing based on pattern recognition in that has been trained into the algorithm. This can give much much less reliable but fast and impressive results.
It is also worth noting that these videos were also probably rendered on supercomputers if they were done at decent quality very quickly.
When Pixar builds a movie, it is creating a detailed 3D space and rendering it out to a specific camera.
When current AI video models make a video, they are just drawing pixels. They don’t have any understanding of 3D space, they have no knowledge of lighting sources, they can’t track what is happening off-screen.
This makes them very quick (relatively) because there’s a lot less processing needed. But, at least for now, the Pixar method is much more reliable.
Think of it like the difference between building a model house and drawing a house. If you build the house, you can film it from any angle, move the camera, do whatever you want. But it takes a long time.
If you draw the house, you have the house from that specific perspective in that specific moment, and it’s fast. Nothing is being simulated. You don’t have other angles, you can’t change the lighting.
Truly explaining like you’re five – Pixar is cooking ramen in a broth that’s taken years to perfect with hand-pulled noodles, the best quality pork, and the freshest ingredients. This AI tool is heating up Maruchan Instant Ramen in a styrofoam cup in 4 minutes.
The average idiot will not know the difference but most will tell.
Pixar is rendering very complex 3D scenes one frame at a time. They can go back to the scene and do whatever edits and animations they want. You can revisit the files and say change the colour of a car and that model would have the new look throughout the animation.
Ai stuff is just a drawing, a picture. You can’t pan around and look or shift the light and materials etc, It’s just a generated picture. Just like old Disney films are drawings, when one was done it is set in stone, you need to redo everything otherwise.
Latest Answers