Polycount reduction with all-quads model

So I have this little guy:

I tried to reduce the number of polygons from the neck down. The head needs this level of detail, but that was too many for the body. If you wonder why the face topology is pretty much featureless, it’s because facial features are to be painted on him. Also, he’s all quads, and hopefully can remain this way because he must be animated. So I searched bit how to do this reduction and came across this fantastic infographic which you can see here: http://s153.photobucket.com/user/animatics/media/Lightwave/Polygon-Table_v04.png.html

I tried applying the 4-to-2 method at the neck, like so:

However, it’s causing horrible artifacts with smooth shading, as you can see here:

I have tried adding edges loops around the area but to no avail. What I am doing wrong? Wrong method? Wrong area on the mesh? Both? Any help is much appreciated.


I trust you when you say that you need that level of details in the head, but from what you’ve written it’s not clear to me why TBH.
Told this, I think those normals’ issues could be generated by the sets of 3 vertices that are not consistently positioned with the loops flow. I think it’s too much difficoult to eliminate this issue with that topology.
Probably you could reduce it in some ways:

  • Smooting vertices (in edit mode, select vertices, W key->Smooth)
  • Smooting surface in sculpt mode (shift key and paint).
  • Reducing the area of action dedicated to the new topology and eventually smooth. If you notice, a big normal issue is created by the long edge that connects the 2 level of detais.

I tried to be the more clear as possible, but wait for people who knows the topology and normals’ matter better than me.

I need it because of the “hair” or spikes or whatever you want to call it. If the mesh is less dense there, they will really look bad. Unless they were modeled the wrong way to begin with, which is absolutely possible given the very little experience I have with organic modeling.

If that can help anyone help me, here is the evolution of the model (disregard the fact that the last one has only half a body). As you can see, the mesh density was increased in order to achieve the wanted look for the hair. I am aware that the face itself doesn’t need denser topology, however I’m not sure how I should have done it otherwise.


Just upload the .blend file and I’ll optimize the topology and mesh count.

Is a mesh with less polygons + subsurf modifier a suitable solution for your purpose?
It could be way more manageable.

I appreciate your help, but I want to stress that the point of me asking this question is to learn what I did wrong, what I should do/have done and why. I have attached the blend file in case it helps you showing me what I should do and understanding what I did, but I need to be able to understand and replicate the solution in order to do it by myself, otherwise you will have worked for nothing.

character.blend (831 KB)

Well that’s how it was modeled. The character is to be used in Unity however, therefore the subsurf had to be applied, and this is exactly what gives it such a dense topology.

Hi there, it’s me again. I’ve tinkered around with your characters topology so far, and I’ve managed to reduce it by quite a bit and I’ve also got rid off those ugly 6 sided poles. Let me know what you think, and don’t be shy about asking questions or for additional help with your characters topology.

.blend file


Thank you for your time. This is pretty much perfect topology indeed. Did you just retopologize over the original mesh? Is this a good way to go for this case or I should’ve been able to create this topology from scratch when I first modeled it? What I would really like to know is what is the proper workflow, especially for the hair. Do you start at the top of the spikes and then connect their bases together? I can see how one would go about about creating this shape (which is what I did after all), but I have a hard time to figure out how to make this shape AND keeping it all quads with the minimum amount of geometry.

You should always model your models as simple as possible and apply sub-surf modifiers for adding details in the latter process. And even if your model doesn’t have optimal topology or mesh flow you could always use it for retopology. What I did for the hair is I surface-snaped a subdivided quad onto it and then I proceeded to extrude from it. Polystrips would be useful for that kind of a thing. The way I reduced the topology from your HP model is that I selected an edge ring and then checker-deselected everything so that I could edge loop the remaining edges and then remove them.

There’s even a faster way to reduce poly count on almost any model and that would be the “Un-Subdivide” option which you access via Ctrl+E.

When you say you surface-snap a subdivided plane, do you mean you simply used surface snapping or shrinkwrap? I’m trying to start the retopo with an extruded circle shrinkwrapped around the mesh but I get this unwanted behavior of the modifier (first pic is with the mod off, second with mod on):

As you can see, the second edge loop reaches for the mouth at the front the mesh, even though there is a surface to wrap onto where the eyes are. Do you (or anyone) know any way around this?


To fix it just set the snapping to projection in the modifier and make sure x and y axis are turned on. And no that’s not what I meant when I said I subdivide a plane. This is, .blend.

What I do is try to make the base mesh with as few polys as possible. That way, I can make every single one of them count. Whenever I add a vertex to a mesh, it has a reason to be there and a specific detail it adds. That’s not the only way to do things, but it’s a good practice for organic meshes. The good thing about starting small is that you won’t be as overwhelmed, since there’s less you have to control. And since there’s less you have to control, you can have more control over it!

This is a little off-topic, but for the character’s design, look at the silhouette from every angle. You want the silhouette to make the character immediately recognizable. What worries me is how he looks from the side-- he has no neck. Or she. Or it. Whatever. English and it’s dumb grammatical gender. Anyways, what I do is draw the character with a pencil five or ten times, as fast as I can, doing some of the things he/she/it does in the game. Then I might draw a few more carefully, to plan out the colors and shading. (I’ve actually been designing some characters lately, and it’s by far the most difficult part of the process). This has two purposes- obviously, it’s easier than just modelling it from scratch, but also it helps you to plan ahead for animation. Because if you go to the trouble of modelling a character but can’t get it to animate, you’ll be pretty mad at yourself. Do you want his legs to swing like noodles? What you have will work. Or do you want his legs to bend like knees and ankles? You may need to add some loops. Maybe the shoulders will need more freedom of movement? That’s what I mean, and maybe you’ve already done all that. But if you haven’t, it might be helpful.

A couple more thoughts: When modelling, make sure to have even texel density- that is (as I understand), try to make all of the faces about the same size and shape (squares are best!), and if the size of the faces must change (in case one area needs more detail than another, such as a face or hand), do it gradually if you can. This could be a problem in Ognjen’s retop- the body gets more detail than the face. If this character had a nose and eyes that’d definitely be a bad thing. As it is, I think the polycount could be reduced further-- you’re making a game character, so don’t be afraid to use a few triangles! (I’ve been looking at some of the meshes on The VG Resource, they use triangles cleverly.) So what if it subsurfs ugly? Bake a high-res model’s normals onto the low-res model. You can use the same high-res, baked normal map on several different detail levels. That solves both problems at the same time. I hope this is helpful.

Did you mean “turned off” because I have tried multiple settings and the only way to make it work properly is to set it to projection, turn on both “postiive” and “negative”, and turn all the axis off, otherwise everything goes all over the place.

Just a quick update on this. I finally made it. I had a friend who is a professional 3D artist come and answer a few of my questions. He never used blender but he sure knows a lot about topology. The best thing I’ve learned from what he told me is that when you’re a beginner, you will have to try multiple times: when your topology is all over the place, don’t be afraid to scrap the whole thing and start anew with another method. This is what I did, had to do it 3 times but in the process I definitely learned a lot.

If anyone who is reading this is curious as to how I proceeded: I found the best way for this mesh was to shrinkwrap a subsurfed cube around the head, position it for optimal density, apply the shrinkwrap, delete the faces in the areas where the topology is conical (the hair spikes), reshape these holes so they fit the base of the spikes, snap a grid on the tips with the same amount of vertices in the outer edges as there are vertices around the base and then proceed to bridge them by extrusion.

The body was made the same way but with a normally subdivided cylinder instead of a cube for shrinkwraping, and the arms and legs were treated like the hairs on the head, and I also did only half the body with a mirror modifier for symmetry. I then applied the mirror and bridged the head with the body.

At first I tried it by starting with the tip of the hair and manually join the bases of the cones, taking care not to make any 5 sided poles or triangles or ngons, but the result made it really hard to keep an even density on the flatter areas, and the vertices in the loops were not spaced equally. It also made it really hard to follow edge loops, some of them were following a random path all around the mesh and I was told this was no good. See examples here:

(Final result in next post)

Here is the result (1 is the original base mesh, 2 is my retopology, 3 is Ognjen’s retopology):

I also re-positioned the arms into a more neutral position.

@Ognjen Thank you for your mesh! I referred to it often during my learning process so it has been quite useful.

if anyone has any comments on what should be improved at this point, don’t hesitate to let me know! Otherwise thanks to everybody who assisted!

You’re welcome, man. Your mesh looks even better than mine.