Objective: To have fun around using 3D, digital painting and AI image generation. Testing the resources needed, the modularity, expediency of the process, and how accurate the results could be. Idea: I wanted something like a that reminisced old timey pulp magazine cover, but with a modern look. Ideally, it would look dreamy and foggy. Assets: Characters modelled on Genesis 8 base, with morphs and textures made by myself. Background is a mix of custom 3D objects, plus store bought scenary (IIRC Grovebrook Park and Urban Sprawl). On the AI side, I used the models Chillout Mix and NeverEnding Dream. I also used free images here and there. Process: I started with a 3D render, the ones I usually do. I tried many poses, my lighting options (turns out that doesn't make a difference). Then I started splicing the image in multiple parts by theme (Lois, background, Clark, etc...) and started tweaking them with AI, and digitally painting over when needed, doing many generations, a lot of repainting and blending. The objective was to take these bunch of different layers and blend them to something somewhat coherent. Problems: It's definitively more labor intensive than just painting, or using AI, or manipulating 3D objects then rendering, mostly because all three methods hate each other I guess. And I'm not very good. For one, AI hates faces. And eyes. Faces with glasses. And hands. And limbs. And edge of clothes. And objects. And everything. It will try to mix everything, fuse clothing to body, deform limbs, etc. Jon was supposed to be snapping his fingers, but AI really wanted him to hold a glass... or hallucination object. Every new layer needed more painting to add cohesion. And it was not very predictable, I wanted, for example, to make Lois look less like a teenaged anime girl, but the model does have biases. After that, make the whole thing look not like a bunch of layer barely blended at the borders was more time intensive than difficult, per se. I couldn't just chuck the whole image on AI and hope it would do it for me because I don't have a server room full of GPUs, and I didn't want to compromise on resolution, so I did by hand. I will pretend the ghosting was an intentional artistic choice. It was my first time doing this, so... yeah, there are things that I didn't know I could do that would have helped in the beginning. My thoughts: Overall, it's very resource intensive process, for GPU, time and energy. It's not very modular, or more precisely, not easily modular. I did a test with an older Jon model and just the thought of spending even more hours made me decide it's not worth it. There's an advantage that it's free, but it's also single use unlike 3D stuff you can make or purchase or just find online. Aesthetically... I don't dislike how Clark or Jon came out, but Lois was not the best. It was fun, if exhausting.