The example of the cube is a simplified version of what I’m actually intending to do. My actual project has materials that use ray mirroring, which along with AO means it takes about 10mins to render one frame. Using this baking method cuts out the added time needed for computing AO.
I’ve used this baking method before and I thought it worked. However when I did it before, all my textures were made in photoshop and mapped directly to the UV coordinates, so no textures were baked from ‘orco’ map input settings on that occasion. Unlike this occasion.
I am aware that sepcular is camera dependent. That is why I have two scenes - one with original materials and procedural textures so that the specular information is obtainable, and a separate scene with the baked models. I’m not trying to obtain any specular information from the baked scene.
I’m beginning to suspect that I’ve actually just deluded myself into thinking this is feasible.
OK, I can see a need for the process you’re trying, if it will cut down on per-frame render time significantly. If you’ve done something similar in the past, I can’t see a reason for it not to work now as long as you allow for the possible differences in the result of baking certain materials features from models that are initially mapped in a different fashion. You certainly should be able to bake the AO for static objects, since I’m pretty sure it’s calculated only for actual geometry, not for normal effects like bump mapping.
If there is an image texture to be used for normal modulations (i.e., bump), my own inclination would be to convert that texture to a normal map (probably tangent), which AFAIK is a faster algorithm, and not bother trying to bake any portion of the normal mapping to a diffuse texture.
My only doubts about applying the normal map are that after I have baked the material with the AO, I need to set it to be ‘shadeless’. Will the normal map have any effect on a material that is set to shadeless?
Probably not, but for use in the specular pass, it will probably render a bit faster than a grayscale bump map. Not sure how to approach baking the “shadow” portion of the bump/normal pass in the Compositor, if the specular is actually done completely separately, it would take some experimentation.
The thought occurred to me last night that perhaps all I need to do is just bake the AO, as that is the main cause of increased render times, and obviously you can’t bake raytraced things such as specularity and mirroring.
So I’ve made a baked AO texture, turned off the AO in the world settings and applied the baked texture to the original material using the UV coordinates - mapped input is ‘flat’ and it’s mapped to ‘col’ using ‘add’ mixing method.
But of course I’ve obviously misunderstood some basic principles again as I’m not getting the correct results. The cube is too dark now.
Turning off the AO in the world settings and then applying a baked AO texture would still require the material to be set to shadeless… but then how will the bump map then work? And if the material is shadeless then I’d also need to bake the shadows too.
I can’t even see what the point in baking things is anymore. Seems to me there’s no way of preserving the true qualities of bump mapping when you start to bake things. I’m so confused.
Render times are marked in the image – pretty significant time savings. I think with some more tweaking you could probably get an even better match. The biggest diff is the floor texture, but I’m pretty sure that’s caused by baking a very large plane to a relatively small image size (1024).
Haven’t tried camera moves yet but it shouldn’t be a huge problem.
