Blender 2.8 baking improvement

So, does anybody know if there’s gonna be any improvement in cycles baking in 2.8? I have the newest 2.79a version, but if i want to bake any map like diffuse, roughness, metallic, bump etc. (lets say from a complex nodegroup, or a set of different materials, or a highpoly object or a group of objects or anything else) to make it simple and usable by other software it is terrrrribly slow. There is a huge discrepancy between rendering a final image from a complex set of nodes, objects, materials or textures and baking it onto a certain map in cycles. And yet baking a diffuse or roughnes map doesn’t need any lighting, reflection, pathtracing, volumetrics, caustics and all this stuff - it just has to copy the RGB or BW values from our set onto the output image. So thinking logically - it should be much faster than rendering, but in fact it is incredibly slower. There is also a huge difference between baking times in Cycles and Blender Internal. Baking only a diffuse map from a set of 172 highpoly models onto one lowpoly model in 4Kx4Kpx took over 3 hours in cycles compared to less than a minut in BI. So if we do care about a quick and effective production pipeline of if we’re working on deadlines - cycles baking totally sucks. But it still has more possibilities and available tools than good old BI and there are many cases in which you cannot just simply switch from Cycles to Internal. If you’re a quite skilled artist you can eliminate the need of using 3rd party paid software like Substance Painter or Substance Designer using just the nodes in Cycles.
I just wanna know if the devs are working on some improvement of Cycles baking performance. It would be wonderful if we had Cycles possibilities and BI speed in one single baking pipeline. Or if the new Eevee engine would allow us to bake PBR maps in a fast and comfortable way.

2 Likes

Actually they are not working in baking workflow except small improvements. At least I don’t remember any about it. But yes, baking needs a overhaul.

I am about to quit trying since i spend the past 2 days trying to get a decent displacement map in blender 2.79 with blender internal and even after 17 years in the game development field i cannot get a decent result!

I tried playing with distance and Bias without success

I also use Xnormal who do a very good job for displacement but i was trying to stay in one app and it seem i will have to forget about decent baking solution in blender for now.

The last thing i will try tomorrow is the textools baking and after that i will stop losing time and simply do it in Xnormal.

I agree, waaay to convoluted at the moment.

Figured out how to bake procedural maps as displacements (with emissive shader) but failed to get anything from highres to a lowres that’d work (displacements that is). Eventually went back to baking them in zbrush/xnormal.

I’m sure the pipeline is just work in progress, surely this will change with UDIMs coming up, hopefully some texturing improvements and the likes.

1 Like

Agree they will certainly improve it but it is always annoying to be force to export in another app just for one single task!

So far all my test comparing displacement in blender vs Xnormal let Xnormal as the clear winner for the quality of the maps.

Nowadays most artists i know who work on AAA games all make heavy use of geometry to low poly displacement for thing like scales, stones, deco, jewelry etc.

It’s an essential part of the pipeline!

2 Likes

To be fair, baking in max or maya isn’t that great either. It would be nice to keep everything in one program, but that’s quite difficult these days, especially for game artists. Seems like most people in games are using marmoset or substance. I’d rather see improvements to uv editing.

It is indeed! I just don’t understand why the devs don’t get it. In my humble opinion it is the most lacking feature to make Blender a real industry standard.

1 Like

A Baking system overhaul was proposed some time ago, and it seems like there has been some updates recently:

https://developer.blender.org/D3203