I’ve finally found the time over the last three weeks to fully immerse myself in the practical application of AI. While familiar tools sometimes appear, the mechanics are unlike anything we’ve worked with before. Look at the image in the publication — it took me about 5 hours to create, of which assembling the geometry sketch with our logo in 3Ds Max took 15−20 minutes and the same amount of time was spent on post-production. The rest was an engaging generation process.
A cold start with the "trial and error" method won’t work with AI, and thoughtful work requires a thorough study of the subject matter. Discussing Stable Diffusion with colleagues (SD — the most flexible and powerful neural network), some compare it to learning Photoshop from scratch, but to me, it feels more complex, akin to V-Ray. Rarely am I surprised by anything in the archviz industry, but the initial results brought indescribable excitement, fluctuating between "God, this is the very 'beauty button'!" and "It seems all artists are doomed."
In practice, we’ve already adjusted our approach to work with the introduction of SD, and according to feedback from two clients, they share our excitement.
The developers of the source code at Stability AI have released the genie from the bottle, and numerous enthusiasts have picked up on this process. It’s gratifying that our compatriots make a colossal contribution to this. And the person from whom one should start learning SD is named Khachatur (Xrist).
SD has given us a new direction in the market alongside traditional Full CG and Matte paint approaches, which I would call Full AI. Whether you like it or not, studying and implementing these tools is essential. Otherwise, you’ll hopelessly fall behind your peers.