It was a slow clunky and awkward transition to polygons mostly because it was like starting from 'square one' again in resolution, sprites just hit there stride in technical resources around 32-bit consoles and whoop whole new standard introduced. Probably the most awkward place to be during this time was on platforms that did 3D graphics but only to a smaller degree like the 3DO. Jaguar, and 32X as you had polygons but you still had to use sprites to have any kind of frame rate.
2.5D sprites have always bugged me, they work but being such a brute force method they come with functional limitations. They're big bitmaps, they animate poorly, they can't do very many actions, they can only hold one item, and they struggle to approximate 3D.
There large image size is fairly unavoidable as detail is expected and extreme scale differences between units is usually proportional IE. Doom Grunt VS. Cyberdemon.
The poor keyframe rate is mostly related to the large size of every keyframe and having a corresponding variant from usually 8 cardinal directions per frame so 1:8 which is expensive resulting in a less than stellar 2-3 frames often. By necessity the playback speed is usually ramped up to compensate creating a fast jerky animation quality.
A byproduct of large image memory needs is a limited set of actions usually Walk, Attack, and Die. This results in a sprite that can do very little but combat so things like complex NPC behaviors go out the window.
Since the sprite is comprised of a 'whole' it can only carry one item resulting in a player avatar appearing to hold one default weapon even if the HUD shows differently.
The 8 cardinal directions a staple of 2.5D is adequate in itself but the main problem is that because of the high image size very few sprites actually get all 8 angles, NPC sprites partially but certainly not item sprites which are typically a static billboard.
2.5D FPS games still have active communities but new content is minimal and remastering old content is a non-unified field, some do high res sprites or 2D voxels or textured polygonal meshes. Because of this I think we need to build a better 3D sprite standard and I figured this forum would have fresh eyes rather than the rigidly defined camps of Doom or Duke3D forums.
I think the only way to "fix" animated 3D sprites is to go modular and billboard render each body part and then reassemble it into a rag doll skeleton rig. Only found one guy who took it pretty far which is Garry Whitton, he uses a quirky capture process but the premise will work with any 3D modeling method. If you're good at 3D math he could use your help. :)
https://www.youtube.com/watch?v=VcXb2P7BHAk
Visual quality takes some hits but what method doesn't lol, models have obvious seams like Virtua Fighter or my inspiration of full articulated action figures. To be honest it will probably look like Robot Chicken in practice but at least it will be more functional. One advantage sprites have over physical models is it can overlap and intersect rather than needing mechanical systems.
Attachment 9224
With body parts one can really take advantage to mix & match NPCs, swap out items, and share animation data. Facial animation should be possible like phoneme frames if not full animation and some angle clipping, no sense rendering a smile seen from the back of the head.
Gary set the view angle needs pretty high as I doubt prerendering will get much higher than 45 degree increments at most with lots of page flipping so left and right images are one. As a result a centered overhead light source is the obvious default.
There are some optimizations to be had with sprite rotation as in paper dolling some part movement rather than rendering redundant bitmap from different 2D orientations. Never quite understood the limits of sprite rotation and scaling hardware so getting some info on whether polygonal billboards is the better option would screen for likely platform standards.
Not much good keeping this rattling around in my head so might as well spread the idea around. :)
