Encore

// ERR: Fault Detected

Over the years we had periodic issues where some of our render nodes wouldn't load all of the texture maps, models, or other assets. This would cause flickers in the rendered animation as parts of the scene change in any frame rendered by the effected machine. In the past we would catch these glitches while processing our footage, but as we sought to automate more and more of our post-production pipeline we needed a tool to help identify them.

// Solution

I began work on this project at a time when we, as a studio, were pushing to acclimate to AI-assisted workflows more. As a result, I prototyped the entire tool using GPT at first and then Gemini, in an effort to test the capabilities of these models compared to Claude (which was used by the rest of the studio). While the models really wanted to bloat the project with image-recognition libraries and other AI-driven methods of detecting changes, I directed it to develop an objective analysis of the frames compared to their neighbors using a Difference filter. Thresholds for flagging frames would be determined automatically based on the frame sequence's baseline variation.

After a day of guiding the LLMs through developing a usable interface and analyzing the frames, we actually had a working prototype. It ended up designing it in Python and QT, and while functional performance was fairly slow even for analyzing a single shot of ~120 frames. I'm currently re-building this program in C++ for both performance and additional practice with the language, using the techniques prototyped here as a foundation.