Topic: Ai Image Upscaling

AI video generation models are rapidly improving but face significant challenges for professional filmmaking, primarily concerning control and consistency in location and character. Current models often produce H.264 (MPEG) compressed videos, which are unsuitable for theatrical projection requiring high-quality DNG sequences. A potential pipeline involves breaking H.264 videos into PNG image sequences, upscaling them, and reassembling into DNG, but this demands immense GPU resources, posing a major hurdle for the film industry.

Theatrical projection prioritizes bitrate and data per image over mere resolution, requiring millions of colors and rich gradient information, which is why expensive cameras like Arri and Red are used. While AI image upscaling models are improving, they currently process 24 images per second, making full theatrical-quality video generation challenging due to immense processing power and data requirements. Small filmmakers might remain safe as the cost for high-end AI film production is substantial.

Ben Affleck's analogy for AI states that training data typically consists of 10% best, 10% worst, and 80% average material, leading to average output if not refined. This concept applies to upscaling, where tools like Topaz (used by video editors) enhance individual frames. The discussion highlights how artists can apply AI, using tools like Comfy UI (an image workflow), to break down videos into 24 frames per second for individual image upscaling, then reassemble them for high-quality video.