July Development Update 1
Hello All,
Work on the next update of HEROES OF CELTICA is progressing, and I find I am pacing myself with this update, doing regular work every day in moderate amounts rather than pulling all-nighters, which I have done on way more updates of CELTICA than I would like to admit.
To make CELTICA, I use 3D models and assets purchased from online vendors at websites like DAZ3D and Renderosity. The concept is quite simple, someone once referred to it as "playing with digital dolls" and they aren't really wrong. The bulk of the work when making images is in setting up the scenes in a low-res preview window - finding the right assets you want to use and positioning them - and then loading up the characters and setting the lighting. When you have everything set, you render the image at high quality to get the final product, but not too high a quality! There is a balance between picture quality and file size that needs to be achieved (reminds me I need to convert the images in Celtica into webp format or something).
This way of creating 3d art for the home enthusiast has basically been around for 30 years. Of course the models and rendering engines have got better, but the workflow itself has essentially stayed the same at its core.
Until the coming of AI image and video models, I suppose. Like many, I have mixed feelings about AI. A lot of the 2D images generated by AI feels 'samey' and most of us now can spot AI a mile away. The days of six-fingered people is slowly fading into the rearview mirror as the newer models come out, but there's something soulless about it.
But really?While those of us that use 3D rendering engines express ourselves through the scenes, lighting and posing, we are all using the same pool of 3d enthusiast resources, so games built on things like DAZ3D are 'samey' as well. It pays to be self-aware!
But back to AI and damn...It's getting better. For the cutting edge AI image models that are publically available, you load up your reference images and learn the idiosyncrasies of how AI interprets your prompts and images into a video clip of no more than 10-15 seconds (current limitation, although the next models seem to stretch it out to 20 seconds). The fact you are using text to describe a moving visual scene is tangential to me - an imperfect input tool for the job. Of course, if you make your own AI rig at home, there are posing tools you can integrate into an image rendering machine, but unless you can chain together enough graphics cards to make a miniature server room, your processing time even on a high-end PC would still be about 30 minutes per 15 seconds of video at 720p.
Enter the dopamine slot-machine that is Seedance 2.0. I've been playing around with it and experimenting with how it works and the way my stupid brain functions, I started rendering clips that could basically be something from STELLAR INCOGNITA. Sometimes the clips look like a blockbuster Hollywood movie and sometimes they look like bad actors on a green screen, but I have to say, these bleeding edge video models have caught my attention. For a home enthusiast, to be able to pay for some data center computing power (R.I.P. local water supplies) in order to play with upgraded digital dolls, it makes me think about how this new tool can effectively be used, but not abused.