I want to quit 3d because I can't bake models properly.

piratek098

How many times I would not return to this topic and how many courses I would not watch about baking, this topic is simply too difficult for me to and/or these courses were not of the best quality. For a long time I have been looking at how clean baking with a simple topology I see in various professional artists and I also strive for it. However, I'm never able to achieve baking without artifacts, and I don't know what I can do about it. I can't spot where I am making a mistake. Today it can be said that I came closest to good baking. Maybe I'm exaggerating, and I shouldn't worry about these artifacts too much, but in the screenshot below could you tell me if baking is good?







My models:


I'm even willing to pay if someone shows me how to do it in a video. I've been struggling with this for too long. I'm sending the save with the files below:
https://drive.google.com/drive/folders/1KzeNDC4XdDzZxiqDdDbv4I9-Ul5xdvRj?usp=sharing
I'm starting to think that decimation is my only option to make a good bake. Thanks for the help in advance! 😭

Fabi_G

Hi! Not at computer right now, but a common mistake is UVs not being split at hard edges. Hard edges are useful at steep angles to control the lowpolys shading - a smooth curvature wouldn't qualify for hard edges (but can still have a UVs split). Your post is missing images of the lowpolys shading and UV layout, so hard to tell.

Fabi_G

UVs and shading should work. No hard edges, so what I wrote before doesn't apply here. With the lowpoly being constructed out of individual boards, I would try using Bake Groups (if you don't already) to prevent adjacent geometry being projected into the normal map. Highpoly seems pretty heavy and could be decimated it a fair bit without losing detail (and depending on the resolution of the texture that is baked to, fine details might not even be picked up). Keep it up!

piratek098

Fabi_G said:

UVs and shading should work. No hard edges, so what I wrote before doesn't apply here. With the lowpoly being constructed out of individual boards, I would try using Bake Groups (if you don't already) to prevent adjacent geometry being projected into the normal map. Highpoly seems pretty heavy and could be decimated it a fair bit without losing detail (and depending on the resolution of the texture that is baked to, fine details might not even be picked up). Keep it up!

Will I be able to separate the model into separate parts and reconnect them after baking, or will it significantly affect the normal map? I want to import this model into the game and I heard that each separate model that will not be connected additionally costs draw cells.

Fabi_G

Sure, you can join everything together before export (when using modifiers, apply first). In Unreal, you can also combine multiple meshes contained in an fbx by ticking "Combine Meshes" during import.

pior

- You high is needlessly dense. It's always a good idea to keep it at the minimum density needed, so that you can manipulate it freely in a regular 3D app. I ran a 20% and then a 50% decimate and the results were perfectly usable. That means that the processed high was merely a tenth of what you had.
- Also, its surface quality is not bold enough to transfer well. These little chips you have need to be more controlled and clear.
- Your UVs are too packed.
- Your low needs to be triangulated.
- The overlap of rays can be dealt with easily by animating both the low and high on the timeline, either by separating them into parts just for baking or (my preferred method) by assigning blendshapes. It does require to have a manageable high of course, which goes back to my first point.


I also widened these needlessly tight bevels.

It does bake perfectly fine - but I would say that indeed, the biggest issue here is the visual quality of the sculpt.



As you can see these are not some complicated rules - just basic/fundamentals, sprinkeld with a bit of common sense.

piratek098

Today I will sit down to this model again and let you know how it went. I feel that thanks to you, I have learned a lot. Thank you very much for your help.

EarthQuake

@pior's advice is generally very good. Except for this section:

"The overlap of rays can be dealt with easily by animating both the low and high on the timeline, either by separating them into parts just for baking or (my preferred method) by assigning blendshapes. It does require to have a manageable high of course, which goes back to my first point."

In Toolbag, you should use Bake Groups to separate intersecting or nearby elements to avoid ray intersections. This will give you more control over the bake and how AO is generated. Bake groups can be created manually or by naming your mesh objects with the corresponding naming conventions and using the quick loader. Bake groups do not require any manipulation or optimization of the high poly objects either, which is another plus.

See the Bake Groups and Quick Loader sections here: https://marmoset.co/posts/toolbag-baking-tutorial/#basics

Another approach would be to model these as watertight sections. If these objects do not need to animate or break apart, it's unnecessary and wasteful (both in terms of triangles and UV space) to model the low poly as individual staves, as the faces between each stave would never be seen.

Shrike

Dont forget to check out the understanding normals thread here on the forums, I think in technical talk

pior

@EarthQuake Hehe well, it sure may sound like a unusual approach but the blendshapes / keyframe animation method has a ton of interesting advantages - the main one of which being that it allows to prepare things non linearly, as opposed to having to setup everything at the end in the baker. And also no need for naming or manual sorting of any kind.

Now admiteddly this is better suited for cases when the high doesn't need to be crazy high, of course ; and sure enough, it also comes from having developped when having to bake directly in Blender/Max/Maya as opposed to TB.

EarthQuake

pior said:

@EarthQuake Hehe well, it sure may sound like a unusual approach but the blendshapes / keyframe animation method has a ton of interesting advantages - the main one of which being that it allows to prepare things non linearly, as opposed to having to setup everything at the end in the baker. And also no need for naming or manual sorting of any kind.

Now admiteddly this is better suited for cases when the high doesn't need to be crazy high, of course ; and sure enough, it also comes from having developped when having to bake directly in Blender/Max/Maya as opposed to TB.

I'm very familiar with this approach, I've used it and wrote about it back in the dark ages of baking - it has significantly more disadvantages than advantages. I would consider it if baking in Xnormal or some other baker that doesn't have modern features, but otherwise, you're adding steps only to produce worse quality bakes. With bake groups (or name matching in Substance), you have the organizational step (which tends to be trivial if your source files are reasonably organized), but clear benefits in quality and flexibility in configuration.

Benefits of bake groups:

1. Bake groups can be automatically configured by naming conventions (just use the Quick Loader), so there's no additional setup once you import the mesh into Toolbag. Groups are created/destroyed if you update the mesh with more or fewer parts (auto-reload), so it works effectively even mid-way through production of an asset
2. You can avoid ray intersection errors completely without animating or exploding the mesh, with meshes that have many intersecting parts setting up the animation is generally going to take more time than making sure your objects have sensible names
3. You can bake AO from one part to another without intersection errors - this is not possible to do with the animate/explode workflow without doing something like baking a secondary AO map from the low poly and compositing the two together, which takes a lot more time and will introduce low-quality vertex normal artifacts, double up AO details, and generally look rather bad
4. You can selectively choose which bake groups receive AO from other groups, which is great if you have static and dynamic parts, for instance, a bolt or magazine on a rifle - again, this is not possible without painful manual work, like creating a mask for a secondary AO pass baked from the low - you'll have to redo this sort of work if you rebake, change your UVs, etc
5. You can modify your cage per bake group - great if you have one or more complex parts that need extra work. You can make a custom cage with other workflows/apps but this is generally quite a bit easier (and you can use custom cages with bake groups too) 
6. Bake groups make working with complex assets easier, as you can toggle the visibility of the group to isolate sections and adjust skew correction, cage offset, etc
7. Lastly, without any of the multi-bake & composite steps, you get a real-time preview of the final result, which makes it easier to preview, adjust, correct problems, and finally commit to the bake knowing exactly what you're getting

Most of the unique baking features in Toolbag were designed specifically so people would no longer need to do the whole explode-the-mesh song and dance.

zachagreg

piratek098 said:

Fabi_G said:

UVs and shading should work. No hard edges, so what I wrote before doesn't apply here. With the lowpoly being constructed out of individual boards, I would try using Bake Groups (if you don't already) to prevent adjacent geometry being projected into the normal map. Highpoly seems pretty heavy and could be decimated it a fair bit without losing detail (and depending on the resolution of the texture that is baked to, fine details might not even be picked up). Keep it up!

Will I be able to separate the model into separate parts and reconnect them after baking, or will it significantly affect the normal map? I want to import this model into the game and I heard that each separate model that will not be connected additionally costs draw cells.

All of the above thread is incredibly sound advice. One thing to add based on this quote, while you are still figuring out the main pipeline of 3D please ignore this fear mongering about draw calls that so often gets blown out of proportion. 99% of the time in a non-production environment you will not need to worry about drawcalls. This horrid fear that is instilled upon new game artists that they need to make these PS2 level simplified and heavily optimized assets is so antithetical to the process of learning 3D and CG art in general.
The only reason I bring this up is because it appears to be bringing in some unnecessary stress into your learning process and you simply will not need to worry about it for most of your non-production lifespan.
When it comes to drawcalls, polycounts and material amounts ask the question of, "Is this sensible for how important this asset is?" That will clear up a lot of your worries and allow you more freedom to learn and explore your artistic sensibilities. The reality is optimization questions like that are so dependent on so many things that worrying about at every stage only hamstrings you from make quality art.