As you model this thing, remember what will – and, will not – actually be visible in the frame. I would suggest that you should very soon set what you intend to be your “final” camera location and settings, and block-out such elements as trees and buildings, all to scale. Make sure that the distance to distant objects is actually realistic unless you intend to use “forced perspective” (and, so that you don’t unintentionally wind up doing so). At this point the buildings might be cubes, but they should all be there, at proper distance and scale.
I see the front of one street-car and the front-and-left-side of the other. If that be the case, don’t bother to model anything that you will not see. Also, don’t bother to model details that you realistically won’t notice, unless at some point in the future you determine that you “objectionably actually do.”
The proportionate dimensions of the model need to be correct: my blush about your current model is that it is not long enough. Get actual reference photos of San Francisco street cars, determine their dimensions and be sure that your models match them. Or, if your exclusive goal is to match the painting, ignore this admonition completely if you decide that it does not matter to you. “What’s best” depends solely on the project and your objectives for it.
Thanks for the feedback, I agree there is something wrong with the length of the cart. The painting reference is misleading with the foreshortening, I might have to fix this before I am too far in this project.
OK I tweaked the mesh a bit and scaled along the Y axis, adding a little bulge back and front, changing the bench model and adding windows on the rear side (as reference shows a cart from the rear).
Ive elongated interior area + added small driving lodge on the front
new UV unwrap + new texture painting
I’ve created different maps for the color, glossiness, bump and emit
and assembled them in a single shader
I want to add more details on the different materials to give some worn-out feel + decals for the text and ads
have started to model close-up buildings so Im putting a lot of details.
Im always torn between joining all meshes and merge all the vertices to have a continous non-manifold mesh and keeping several disjoined meshes overlapping each other. The latter option creates less polys but features hidden surfaces. For real-time this would be bad as the pixel shader would have to run on these hidden textures unless you do a visibility test first. But for ray tracing is it that bad? I would appreciate an opinion on that.
For ray tracing it barely matters. Texture lookups only occur when the surface is hit by a ray, by definition rendering it visible.
The main thing you might want to look out for ray tracing-wise is tons of super-close (nearly) parallel faces with intersecting bounding boxes.
If this is the case you may want to turn on Spatial Splits.
had a quick try with basic shaders. Obviously everything (the road) is way too shiny
Im having a hard time placing the building so that it matches the reference and it still makes sense in 3d…
I am having a lot of troubles with the general composition, I don’t think tweaking the focal length will change anything. What bugs me is the perspective in the reference:
So unsurprisingly, all lines converge to the horizon line which is at about 1/3 from the top. This means that the camera is tilted forward.
The problem is, there is no way the verticals on the side of the image would stay that straight with a real lens.
Basically the perspective has been corrected.
The initial choice I made in blender was to keep the camera straight to keep the verticals on the side aligned. But pursuing this way ruins the rest of the composition as it basically moves my horizon line to the center of the image.
So I will backtrack and tilt the camera, work with tilted verticals and correct the perspective in postprocess.
Do you know a blender node that would do that for me? I could also do it in photoshop but Im curious.
