HOME | DD

Description

Why do they look bad?

It’s a really common case when you see an MMD model, you want to use it on SFM because its superior animation capabilities (Register Register Register Register). So you port it, do the texture conversions, make your VMTs and oh wow it looks really bad. You verify your port files, your processes and notice nothing out of the ordinary. So why does it look bad? Well, I have a theory.

One of the main reasons MMD models look bad is because they are, most of the time, low poly. I have a theory, don’t quote me on this even though I know you will, because of the low number of vertices, the interaction with lights are really poor, and if the interaction is poor, the result is going to be poor too, which in this case, is a really awful Ambient Occlusion.
This basically means, the lower the amount of vertices a model has, the less vertices light have to interact with, thus giving that sense of flatness.

It also doesn’t help one bit the fact that an Mmd model very rarely uses normal maps. They always come with a Diffuse texture, but not with a normal nor bump map.

That’s why only a handful of MMD models interact well with SFM, because only a handful are high poly.

The current solution: higher poly count.

This comes surprise to absolutely no one and if you didn’t thought about this solution while reading the above, you really need to pay more attention.

There’s one big issue though: a model having a high vertex count doesn’t make it a high poly model. Yes, I know what you are thinking, that it doesn’t make sense, but you are missing something important: every model have a topology. Topology is the most important part of a model. Basically, every model is built in a way that their Rigging will perform better if it follows its topology. Mess with the topology of a model and you will have a bad time. Just applying “Subdivide” to a model won’t make it high poly instantly.

Then how do we fix that? By applying Subdivision Surface.

Subdivision Surface (from now on, SubS) is a Mesh Modifier in Blender. I recommend you googling what it does for certain, because there’s a huuuge chance my explanation is wrong: What SubS does is, in absolutely basic terms, it applies a Subdivision and then it smooth the edges to give it a round look, all within the modifier. This modifier however, works different than manually applying Subdivide and Smooth on a Mesh.

By having a high poly count, the Ambient Occlusion is also boosted. This makes it for the solution we need.

Demonstration

This is where things get interesting. For this example, I decided to pick Tsunade.

This models ticks all of the regular MMD options (even though it's an XPS model): Diffuse maps only, low poly; Head poly count: 1330 (it contains face, ears and the cover of the mouth. It does not include eyes, teeth or tongue)

If we use Matcap (lovely tool), we won’t have too much information to work with, however, if we tick Ambient Occlusion, we are going to see some nasty things.

Around the eyes everything looks fine, but is nose, eyebrows and mouth where things start looking really rough.

We are going to use 5 Tsunade for the following Ambient Occlusion compare:

- No change.

- Just Subdivision

- Subdivision and smooth.

- Regular SubS

- Quad SubS

But before we continue, what is Regular SubS and Quad SubS?

Well, to not make things longer: Regular SubS is applying SubS to the mesh as it is.

Quad SubS is applying SubS to a model with quads instead of triangles.

You can see the results in the addendum.

These images gives us some really hard pointers that impossible to miss:

- Just subdividing a model not only won’t make it high poly, but arguably, can make the Ambient Occlusion go even worse.

- Subdivision + Smooth gave us very smooth Ambient Occlusion, but if we turn the texture view on, we can notice how hard are the changes in both her brows and rhombus.

- Using SubS in a the model without making any changes changed the Ambient Occlusion for a bit, but if we go into Wireframe view, we can see how absolutely disgusting the topology is. Absolutely horrible.

- If we convert our model by using Tris to Quads and then apply SubS, we notice right away the changes on the Ambient Occlusion. It looks really smooth. Since we applied SubS in our previous test, we then check the Topology and we see no major changes. Everything looks normal. So, now, we turn the Textures on, we can see visual changes as well, however, they are not AS aggressive as the Subdivision + Smooth option.

This, in my opinion, makes Tris to Quads into Subdivision Surface the most optimal setup for MMD/Anime/Low poly models.

However, before you close this and decide to apply this to literally every model out there, you need to check this out.

Big con: Use your better judgement. Don’t apply it to everything.

One big con about Subdivision Surface, is that it’s a Mesh Modifier, this means you cannot apply it to a model with Shape Keys. However, that is not the biggest con in my opinion. In my opinion, the biggest con of this model is bad judgement, because it can absolutely destroy a model on its own.

In the case you didn’t notice, this modifier messes heavily with details in a mesh. That’s why it’s more recommended to use it in MMD/Low poly/Anime models: because there’s no much detail to mess with.

For this example, I’m going to use Caulifla.

So here we have her... or her hair actually. Her hair is nice and cool, but is only 2923 vertices! Let’s fix that! (Main Addendum)

So we do the usual thing, Tris to Quads, then SubS and we get oh my god what the fuck is this… (Addendum, left)

The tips are open… that is not good no matter how you see it. So, in order to fix this, we need to Remove Doubles. For the experienced SFM user, this would have seen like a step to do before even try this, but it’s a layer I added for the last revelation, that comes right now. Since I didn’t apply the Modifier, I can just remove doubles and see the results before applying, so we do that and oh my god… (Addendum, middle)

Maybe we need to apply another SubS and maybe just stop. Don’t bother. This will get worse. (Addendum, right)

Thanks to applying SubS, her cool hair has now become a bunch of bananas.

Conclusion

I believe this is the best thing we can do in order to have good quality anime models in SFM. With no proper normal maps, and the option of still having Ambient Occlusion with just a flat normal map, our best bet is to use this in our favor.

This research is not done yet however. There is still a big dragon I think it needs to be tackled: the VMT setup. This might go easier though, because I have a couple of ideas and a couple of friends who can help me with this.

Terms worth googling:
- Topology (3D)

- Subdivision Surface (Modifier)

- Ambient Occlusion

I won’t leave any link. Search it by yourself. You are the one who should be interested, not me.

Description

Why do they look bad?

It’s a really common case when you see an MMD model, you want to use it on SFM because its superior animation capabilities (Register Register Register Register). So you port it, do the texture conversions, make your VMTs and oh wow it looks really bad. You verify your port files, your processes and notice nothing out of the ordinary. So why does it look bad? Well, I have a theory.

One of the main reasons MMD models look bad is because they are, most of the time, low poly. I have a theory, don’t quote me on this even though I know you will, because of the low number of vertices, the interaction with lights are really poor, and if the interaction is poor, the result is going to be poor too, which in this case, is a really awful Ambient Occlusion.
This basically means, the lower the amount of vertices a model has, the less vertices light have to interact with, thus giving that sense of flatness.

It also doesn’t help one bit the fact that an Mmd model very rarely uses normal maps. They always come with a Diffuse texture, but not with a normal nor bump map.

That’s why only a handful of MMD models interact well with SFM, because only a handful are high poly.

The current solution: higher poly count.

This comes surprise to absolutely no one and if you didn’t thought about this solution while reading the above, you really need to pay more attention.

There’s one big issue though: a model having a high vertex count doesn’t make it a high poly model. Yes, I know what you are thinking, that it doesn’t make sense, but you are missing something important: every model have a topology. Topology is the most important part of a model. Basically, every model is built in a way that their Rigging will perform better if it follows its topology. Mess with the topology of a model and you will have a bad time. Just applying “Subdivide” to a model won’t make it high poly instantly.

Then how do we fix that? By applying Subdivision Surface.

Subdivision Surface (from now on, SubS) is a Mesh Modifier in Blender. I recommend you googling what it does for certain, because there’s a huuuge chance my explanation is wrong: What SubS does is, in absolutely basic terms, it applies a Subdivision and then it smooth the edges to give it a round look, all within the modifier. This modifier however, works different than manually applying Subdivide and Smooth on a Mesh.

By having a high poly count, the Ambient Occlusion is also boosted. This makes it for the solution we need.

Demonstration

This is where things get interesting. For this example, I decided to pick Tsunade.

This models ticks all of the regular MMD options (even though it's an XPS model): Diffuse maps only, low poly; Head poly count: 1330 (it contains face, ears and the cover of the mouth. It does not include eyes, teeth or tongue)

If we use Matcap (lovely tool), we won’t have too much information to work with, however, if we tick Ambient Occlusion, we are going to see some nasty things.

Around the eyes everything looks fine, but is nose, eyebrows and mouth where things start looking really rough.

We are going to use 5 Tsunade for the following Ambient Occlusion compare:

- No change.

- Just Subdivision

- Subdivision and smooth.

- Regular SubS

- Quad SubS

But before we continue, what is Regular SubS and Quad SubS?

Well, to not make things longer: Regular SubS is applying SubS to the mesh as it is.

Quad SubS is applying SubS to a model with quads instead of triangles.

You can see the results in the addendum.

These images gives us some really hard pointers that impossible to miss:

- Just subdividing a model not only won’t make it high poly, but arguably, can make the Ambient Occlusion go even worse.

- Subdivision + Smooth gave us very smooth Ambient Occlusion, but if we turn the texture view on, we can notice how hard are the changes in both her brows and rhombus.

- Using SubS in a the model without making any changes changed the Ambient Occlusion for a bit, but if we go into Wireframe view, we can see how absolutely disgusting the topology is. Absolutely horrible.

- If we convert our model by using Tris to Quads and then apply SubS, we notice right away the changes on the Ambient Occlusion. It looks really smooth. Since we applied SubS in our previous test, we then check the Topology and we see no major changes. Everything looks normal. So, now, we turn the Textures on, we can see visual changes as well, however, they are not AS aggressive as the Subdivision + Smooth option.

This, in my opinion, makes Tris to Quads into Subdivision Surface the most optimal setup for MMD/Anime/Low poly models.

However, before you close this and decide to apply this to literally every model out there, you need to check this out.

Big con: Use your better judgement. Don’t apply it to everything.

One big con about Subdivision Surface, is that it’s a Mesh Modifier, this means you cannot apply it to a model with Shape Keys. However, that is not the biggest con in my opinion. In my opinion, the biggest con of this model is bad judgement, because it can absolutely destroy a model on its own.

In the case you didn’t notice, this modifier messes heavily with details in a mesh. That’s why it’s more recommended to use it in MMD/Low poly/Anime models: because there’s no much detail to mess with.

For this example, I’m going to use Caulifla.

So here we have her... or her hair actually. Her hair is nice and cool, but is only 2923 vertices! Let’s fix that! (Main Addendum)

So we do the usual thing, Tris to Quads, then SubS and we get oh my god what the fuck is this… (Addendum, left)

The tips are open… that is not good no matter how you see it. So, in order to fix this, we need to Remove Doubles. For the experienced SFM user, this would have seen like a step to do before even try this, but it’s a layer I added for the last revelation, that comes right now. Since I didn’t apply the Modifier, I can just remove doubles and see the results before applying, so we do that and oh my god… (Addendum, middle)

Maybe we need to apply another SubS and maybe just stop. Don’t bother. This will get worse. (Addendum, right)

Thanks to applying SubS, her cool hair has now become a bunch of bananas.

Conclusion

I believe this is the best thing we can do in order to have good quality anime models in SFM. With no proper normal maps, and the option of still having Ambient Occlusion with just a flat normal map, our best bet is to use this in our favor.

This research is not done yet however. There is still a big dragon I think it needs to be tackled: the VMT setup. This might go easier though, because I have a couple of ideas and a couple of friends who can help me with this.

Terms worth googling:
- Topology (3D)

- Subdivision Surface (Modifier)

- Ambient Occlusion

I won’t leave any link. Search it by yourself. You are the one who should be interested, not me.