This article is available in: French
AIs keep moving forward. Photorealistic or stylized pictures created by text-to-image tools such as Stable Diffusion, chatbots such as ChatGPT are getting used by a wider audience, and reactions are mixed: some love it, while other highlight the fact that some AIs are trained on data without the consent of the artists/people behind it (our article on this pushback), major companies invest while other are looking closely at the legal ramifications of such tools.
Jiacheng Xu, Lead Compositor who wokerd at MPC, DNEG and is currently working for beloFX, just released ChatGPTforNuke, a tool that allows Nuke artists to use ChatGPT. As a reminder, ChatGPT is a chatbot developped by OpenAI that allows you to write to the chatbot and get answers. You can, for example, ask specific questions or ask ChatGPT to complete a task.
Why would you use ChatGPT as a Nuke user? Not really to chat, but rather to take advantage of the fact that ChatGPT can also generate code and re-hash useful information. For example, you can ask ChatGPT to create a specific script and let it write it for you.
You can also use ChatGPT as an assistant, ask questions about Nuke and specific tools/functions, or give ChatGPT a script you don’t understand and ask it to explain some code. Last, but not least, you can use your own data to feed ChatGPT, then as specific questions such as where the latest sequence cut is stored.
To get a better idea of what can be achieved, here’s a video of ChatGPTforNuke in action:
ChatGPTforNuke is free. You will find installation instructions over on Jiacheng Xu’s website. Please note that Python 3.6 or above is required.
Of course, this kind of tool raises several questions. On one hand, this could be a huge time-saming tool. On the other, artists will have to avoid relying blindly on AIs to create scripts, and make sure they still understand what the code stands for. Furthermore, there is no guarantee that ChatGPT will generate code that will do exactly what you asked for, so you might stumble upon a few suprising results or errors. Last, but not least, the issue of what data has been used to train ChatGPT remains.
In the long run, me wouldn’t be surprised to see software editors releasing similar features for their creative tools, using ChatGPT or another AI. We will keep a close eye on this topic.