Mixamo retarget workflow for character whose poly count and textures are too much for Mixamo upload


I’ve read up as much as I can on Mixamo issues, and it seems there are now file size and poly count limits on the uploads among other limitations.

Ultimately, I’m trying to apply the animations to a character that has a higher poly count (900k as opposed to their limit somewhere around 200k), and a number of tattoos, uv’s for the face, etc which increase the file size… and are problematic in their own right with Mixamo.

The character at the moment is unrigged and has no bones. I was planning on using the auto-rigger in Mixamo. The arms and legs are single meshes, so no separae objects for hand, fingers, etc.

Since I can’t upload the character as is, but want it to render as is, I’m trying to work out the best way to upload the exact geometry to mixamo in a format it can handle, then retarget and retexture as quickly and easily as possible.

What’s the shortest path here? Can I just ignore all the textures for now, drop a massive poly reduction on the whole thing, export as obj, reimport the animated version, and retarget?

I know how to retarget bones, but not exactly sure how that would work in this case since there are no bones, and I have single objects that are being retargeted to entire chains of bones. Can I just retarget the mesh itself?

Or, would it be better to create some bones, rig it myself first, then just upload the bones to mixamo, reimport them, and retarget the bones? I’ve worked with half a dozen or so rigs, but haven’t created skeleton from scratch, so wary that might take more time than I’ve got.
What about a hybrid approach like exporting a simplified character, auto-rigging in mixamo, then outputting just the T-pose, then retargeting/fitting my character in c4d to the Mixamo rig exactly, saving the rigged character as the new base character, then from then on just upolading the skeleton to mixamo whenever I want a new animation, reimporting it, and… not quite sure from there… would I replace the static skeleton with the animated one, and the fleshy bits would just follow since all the names are the same, or am I missing something there like the ability to just import the animation without actually reimporting the skeleton.

Any tips / tricks appreciated.



The approach here is very standard: you really don’t want to have a rigged character that has 900K polys. It would be painful to use. So you have a low poly count character, and drive the detail with render-time subdivision and displacement.

So, you usually need to use a tool like ZBrush or 3DCoat to model/sculpt your character based on subdivisions on top of a low poly animation optimized mesh. You then export the low poly character and displacement maps that will restore the detail in the rendering process with subdivision and displacement. The low-poly character can be rigged and animated (e.g. via Mixamo), and will actually perform well in the viewport because you are within a low-poly budget.

The alternative is to use a proxy mesh to get rigged in Mixamo, and use some form of retargeting, but that requires you to rig and weight paint your character in C4D yourself. And it will be slow.


Makes sense, but in this case, I’m coming from the opposite side of the coin in that he was put together using a high poly head (and separate high poly hair and eyebrows), and high poly shoes that I found online, then I made a (lower poly) shirt, pants, arms, and chest, and Frankenstened him together… so the head and shoes are already high poly meshes.

I’ve only messed with re-weighting once, and it wasn’t pretty, but since the high poly items are all in the head and shoes, would it be possible to just create basic proxy shapes for them, send it off to Mixamo, and replace on re-import without re-weighting (except maybe joint in foot)… or does it not matter that they are at the ends of the skeletal chain?

I’ve done lots of 2d animation, and used c4d for other (non character animation) things before, but some of it’s new, so still looking into optimization, etc. Not sure if I need to re-topo some of my items,.


If you have access to ZBrush or 3DCoat, you could still retain the detail in high-poly model. The process goes something like this:

  • import high-poly model
  • duplicate it
  • retopologize the model copy to ~50-100k polys
  • generate new UV coordinates
  • subdivide new retopoed low-poly model
  • project detail from original mesh onto subdivided retopoed mesh
  • generate and export displacement map (and other maps) from the high-poly retopoed mesh
  • export low-poly version of retopoed mesh
  • upload it to mixamo for rigging/animation
  • import the lowpoly object/rig in C4D, and
  • set up a material with displacement; all the detail shows up in the render

You can actually do most of this in C4D using the sculpt toolset, but the retopology part will not produce a nicely animatable mesh. Google “character edge loops” to see the kinds of topology you are looking for, which ZBrush and other tools will be much better at giving you.


Lots to dig into. Thx. That’s exactly the kind of overview that’s hard to get when learning from piecing together hundreds of individual searches.

If I follow correctly, you’re not just stepping it down for proxy purposes, but actually using the low poly subdivided model in the final render. Should work fine for most of the base shapes, but the reason I used a separate head is because it’s a very detailed (wrinkled) face of a celebrity where the whole project doesn’t work if it’s not immediately recognizable. I did relatively little painting (just lip color, eyes, and some basics. Almost all the facial detail comes from the actual hyperdetailed topology. Won’t some of that fine detail be lost in the face via this process?

EDIT: Wanted to add that I checked each object, and the head and shoes (both of which need to deform since the shoes are huge and need to bend, and head is attached to thick Jaba the Hut type endless double chin… both are made of tris. Everything else is quads. Reading up on quadrangulation options in C4d, but not promising on that end from what I’ve read so far. Too many to do by hand. One more reason for z-brush workflow?


You should have no issue retaining high level details with a lower level model and a high quality displacement map. The trick is to make sure, ahead of time, that your displacement map is doing the trick. The basic settings in Zbrush will get you half the way there, but they do require some fine tuning to really capture your results.


Looking into several details now, and have watched a bunch of tuts on edge loops, quadrangulation, etc.

Problem is I’m rapidly running out of time, and without a decent auto-retopology option, learning a bunch of new tools may not be my fastest route to get the music video done. It’s a cartoonish character, so a bit of unreality isn’t a deal breaker.

Looks like the forum is letting me post links now, so here’s the actual file. I know I’ve stumbled backwards into this, but I have people counting on me to finish it, so need to find the absolute fastest way to get this character doing strung together mixamo animations.

Trump post symmetry.c4d (31.5 MB)

I’ve gotten him down into the 500k range without losing anything noticeable. I’d like to blend the neck into the chest, but the several tools I’ve tried thus far didn’t fare well. Otherwise, I’m happy with him.

I have a good road map for next time and I’m sure I’ll learn zbrush when I have the time to do so, but at this moment, not knowing zbrush, not having done a retopo before, etc, might it be faster to maybe add a few rings around the knees and elbows somehow (I think I saw a tool in the magic plugin series that does this), then insert a proxy head and shoes, then retarget and hopefully weighting issues minor since only head and shoes?

My gut’s telling me it’s time to reign in rather than expand the workflow due to time constraints. Good stuff, though. I’m learning quickly.



Its all been covered here already. I had a look at the mesh and its of course way too dence, its just not practical for use in animation. Now if there is going to be no facial animation then a auto retopo in Zbrush will be ok, you can then get the displacements to do the rest but its not going to get you the best results.

A proper reopology is needed to get the best results because this way you can get away with a medium poly level holding most of the forms in place a auto retopo wont do this and will rely on the displacement to do this.

The legs are also too dence, but you can just get away with it, but there is no need for such high polygon count for such a simple formed character.

For the body the auto reopo can do quite well and you can use this within C4D as part of the topology. What you can do is do a zremesh of the legs, send that over to C4D and delete the areas that are not good and use the polygon pen to finish off the topology. Send this back to Zbrush, store morph target, sub divide the mesh to your chosen level one stage at a time and project from high polygon model for each sub d level.

Once your done your have a nice mesh in which you can make decent Uv maps for. I use Unfold3D for that process. Then make your maps from Zbrush and your ready for Mixamo. It should take upto a day to retopo the whole character, make your maps, and rig it in Mixamo.

If I get time I will make you a quick video to get you on the way.



OK. Watching some tuts now on zbrush, retopo, etc.

What about the overall shapes? Right now, the head/neck collides with, but is unattached to the chest , and wherever possible, nothing else touches because I was trying to avoid deformations where one bodypart breaks through another. Not sure if that’s really the BEST approach, but it seemed to work in some test runs.

Would I be better off with another approach? Certain body parts connected? Get rid of all polys you don’t see like back of chest or top of head under hair?

Tried Magic Merge plugin for head/chest yesterday, and it wasn’t getting rid of interior polys for some reason… maybe because head isn’t closed due to neck hole.

I was also wondering whether creating an internal skeleton of actual objects that matches the desired bone structure in Mixamo might get their auto rigger to bind it’s bones to the skeleton since the proxy head / foot test I ran came back with some weird issues where the auto rig put the shoulders much higher than the clavicle connection to spine, etc. Not sure if that’s due to body parts being disconnected internally.

Running a few stick figure tests today.

Oh, and I get most of what you said… but why the extra trip back and forth from zbrush to c4d? If there are issues with the auto zremesh, aren’t there tools right there in zbrush to dial it in rather than making another round trip to use polypen (which I’ve never used anyway)?


The best workflow is to only make parts that are visible. If the head neck and torso is visible then they should be combined as they are all part being flesh. If the body is then clothed then they will hide the fleshed body so you make or keep only what’s seen such as hands or hands and forearms. The clothed object’s become the body.

The topo tools in c4d are better than zbrushes ones. Use the best of each apps abilities.



Why not Instant Meshes for retopo?


I arrived at something similar by analyzing some characters I downloaded.

After running some tests through Mixamo even with a proxy head and shoes and keeping everything else at it’s full subdivided poly count, I’m still getting some nasty deformations. I watched a few vids on the poly reduction, subdivision, projection process, and I get it generally, but it’s a bit much on the learning curve end of things for my available time given the numver of other things that need to be done.

I do have some pretty significant topo issues, though, so at the very least, i need to do something about that.

Wasn’t impressed and moved on from the vid or two I came across of Instant Meshes, but that could have been the luck of the draw on the vids.

Out of curiosity, seems to be near universal agreement that I need to go out of c4d into something else like blender, zbrush, etc for a variety of things. I’m no stranger to composite workflows in general (I’ve donea lot of work with Premier / Photoshop / Illustrator, etc), but wondering specifically what is so GOOD about c4d that keeps the folks here coming back into it despite regularly needing to make round trips to get the tools they need.


Doing a crash course on zbrush right now.

2 questions pop up in the meantime:

  1. Ive only done minimal texturing and painting in c4d. Is there any reason why it would be better to redo this in zbrush?

  2. That’s for the existing character, but I also need to make a second character… an Uncle Sam character that’s muscular in a top hat and tails painted in flag colors… I don’t see a character like that available for purchase anywhere, so I guess I’ll have to put it together. I have an older but pretty powerful mac, and my PC couldn’t handle Character Creator, so…

I’d guess my best bet to get going quickly would be to buy a character, import into MD to make the clothes, then into zbrush for the face, and all the stuff already discussed here, except that he needs to have white hair and long white beard… which he undoubtedly won’t unless I add it.

I think maybe Daz can work on my machine, but a bit confused by their licensing and very limited budget. I just need non interactive 2d (music video)… and I’ve read a few things about eyelashes and such causing issues downstream from daz if memory serves.

What’s my shortest path to properly configured (Mixamo again) custom character for mac? (Also limited by hardware to osx 10.11.6, so some newest versions might not work.)


Every part of this outside of c4d is new to me, but giving it a shot.

Watched some zremesh and other tuts, and ready to dig in. The retopoflow vid did make it look a bit easier to use, and I keep coming across warnings about the zbrush UI being notoriously unintuitive, but enough info out there to piece it together.

The problem with piecing together such a workflow from tuts is that the bridges between steps are often not covered.

Right now, it’s a high poly head and shoes, and the rest is lower poly with symmetry objects and subdivision surrface deformers applied in c4d.

So… can I just export as is, or do I need to create separate limbs and such first?

What format? If fbx, for instance, there are many options, and some apps don’t like the newer versions, so are there details there like the version of output format?

If I’m making round trips back to c4d, am I sticking with that same format each way? Are there other issues there like scaling, include/exclude items for export/import format that can break the process?

Best practice going back and forth for the mtl, tex, or other related files? Setup issues for project folder to avoid such issues?

These are the kinds of details tuts tend too leave out.


EDIT: Spent a bit of time in zbrush now, and I definitely get why people say it’s anti-intuitive UI.

What about doing the retopo with retopoflow in blender, then doing the rest of the steps as described? File issues or other hidden problems with that workflow?

Also: I’ve cleaned it up a bit, erased polys that are unseen beneath clothing etc, but as for retopo… the shirt and pants are folded back at the edges where they meet skin to give a pseudo 3d appearance, and I made sure to keep a small gap between each part as creases between objects were occurring in mixamo deformations.

Should I keep it like that, or is it better to have them touch… or even to merge them?

Came across a website that was saying to do the import/export between c4d and zbrush as obj, but unless I’m missing something, that merges it all into a single object, right?


Still wrapping my head around the downres / upres workflow, but trying out some of the tools and taking a stab at the retopo in the process.

A few sticking points, though. The file format thing is not a minor stumbling block. I’ve tried dozens of permutations, and every one has had at least some minor issue with not reading textures properly or something along those lines.

Also, there are all kinds of things I just don’t see covered anywhere. Take the hair, for instance. It’s very high poly, and ful of waves, etc… but I don’t really have a clue how to go about doing the retopo for it. Just an automatic zremixh kind of thing? Trying to do the face properly by hand.

The shoes too… they have 3d laces, etc. If I just treat them like any other joint, and poot 3 edge loops near the point of flexion, keep it simple, and rely on displacement, that will be able to deal with the actual geometry of the laces?

I’ve tried a few things like magic merge to merge the eyebrows into the face, but none of them haave worked to delete interior polygons since both the brows and the face are open meshes. Is there another option there?

Also not happy with the job Bodypaint was able to do on the UV. Even the basic stuff I did on the face, no matter which unwrap method I usedd, it jumps around some and puts a few bits in the wrong places… like I’ll be painting on the lips and a bit of it will suddenly jump to the nose, so looking into Painter… unless there’s another suggestion.


I’ve retopo’d several of the pieces in retopoflow / blender.

Still having some assorted file type/version issues, but I’m assuming that can be worked out. The more vexing problem at the moment is that nothing fits together properly once I retop to a lower poly count. For instance, if I take a circle with 100 segments, then retopo it to have 20 segments and add a subdivision generator, the resulting sphere will be smaller.

So… I retopo the pants, and they’re too skinny, now the belly sticks through, they don’t fit into the shoes correctly, etc.

Mind you, I’ve done this so far as experiments minus the displacement, and perhaps that part helps, but is there some trick to avoiding having everything shrink and mismatch… or do I need to just bake the whole thing into a single object, and retopo everything in one go so relative shrinkage stops being such an issue?


I’ve retopo’d or rebuilt everything except the head and shoes based upon the basic low poly philosophy with an eye toward deformation behavior in retopoflow (blender)

I’d like to use retopoflow for the head / shoes too as I find it’s tools a bit more intuitive. This brings up a few q’s, though:

  1. Do I just ignore the actual geometry sticking out of the shoes like the laces and let the displacement handle that, or does that level of geometry go beyond the “detail” that the displacement can handle?

  2. If I’m doing the retopo in blender, then I just need that and c4d, right? In other words, is everything else like generating new UV coordinates, projecting detail, exxporting displacement map, etc being done from c4d, so no need for zbrush or anything else once I have the retopo meshes?

  3. At what point (if ever) do I deal with the symmetry objects? Do I send to Mixamo with them still in place (only one arm, for instance, in a symmetry object), and then it splits into two arms when I get back so they can be tied to the skeleton it generates?


Wow there are alot of questions since I last posted, dont have time to examine all of them so here is my workflow, maybe some if not all your questions will get a answer.

1: Design. If I have a set design and have images front, side, and some action poses, then Id model it within C4D pur polygon modelling getting all the edge flow in place. Assuming this is for animation id take care to have loops for all articulation points such as shoulders, elbow, knees e.c.t.

I would model the eyes and hair, teeth, and all accessories separate, and if the character has non dynamic clothes the clothed areas would replace the body under it. In that case id model what is to be seen only such as hands, feet. If its to have dynamic clothes then I model the body very basic, do some sculpting to get the base shape down then build clothes around that in MD.

2: Zbrush base level: Assuming I had made a good topolological model in c4d Zbrush would be used to refine the final shape, and then id sent it back to C4D. If The model was designed within Zbrush and has bad topology and high dense mesh then Id send the base level sculpt no detail just volume over into C4D, scale it as it needs to be for the world space I need it for, then send bring it into Zbrush as a new name. The matching of names will cause the previous version to update in Zbrush and this can mess the scale up of the model and the brush size, so to avoid that I import the base version scaled correct from C4D into Zbrush as a new model. Zbrush will see this as a new mesh and apply the scale difference to what it needs in meters with the correct export setting for C4D. Id them start to sculpt.

2: Zbrush sculpting: If its got its final good topology done from the beginning id store the morph target, sub divide the mesh and add detail using the layers. id keep larger deformations in one layer and small details in another layer this way I can capture only micro details into the normal map, and large volume changes into the displacement, however I do my best to make the topology hold its own volume.

In the case that the mesh is made in Zbrush has bad topo but is the correct scale for the game engine id not bother with store morph target as the level is still a work in progress but its got the right scale. Id work on volume for a while, then only add the sub d levels as and when I need to, this avoids lumpy uneven surfaces. I wont add high detail to this version, only main forms, no skin details. Id send this over to C4D and use the polygon pen to retopo it. The re topo in C4D is faster than within Zbrush. Modo is even better.

3: The retopology will be given priority to jointed areas for animation, secondary is topo for anatomy, and I will retopo manual the areas that are going to be rigged and animated such as body. Zbrush remesher for some cases can be used such as hair, if this is going to be a static moving object with the head then Zremehser is fine. If its going to be animated such as long hair I would have modelled this correct in the first place and do away with any proxy hair made in Zbrush that aided the design to start with.

With Topo done, the hands, feet will normally be modelled as well as hair teeth and accessories, retopo can take long for some things in which modelling is faster.

Id send over the retopo version back into Zbrush as an appended object into the sub tool of the high res mesh, then project the lowest level of the low level Zbrush version into the low level of the retopo version. Id move up one level at a time and project each level. If the retopo model is good to the point it holds all its own volume id store morph target as to retain the base cage mesh that matches what is already in C4D, but its not always the case id do this as the projection helps push them forms into cage a little more.

When I have the projected version I d clean up scene deleting the high res version, id append in the teeth, hair, and accessories into Zbrush and save this as the final version. Id export a obj cage mesh for all parts and do my Uv work in Unfold3D. With the new Uv version id import these one at a time back into zbrush which will update each sub tool with new Uvs. Id save the Ztool again.

4: All parts will also be taken into C4D ready for rigging. In the Zbrush version Id finish the details and painting.
The details will be stored in its own layer and if any volume has been added that will be for displacment, or pushed into the cage mesh if no displacements are wanted. Should I have already rigged the model and later found that I had a better cage mesh produced in Zbrush id use pose morphs to update the rigged model.

In Zbrush Id go to high sud d level and paint the parts. Id produce the maps without smooth Uvs and I work in linear Uvs. Id take the maps and enhance them in photoshop. I may use substance painter for some parts like clothes, but skin will be done in Zbrush and Photoshop.

5: All maps are brought into C4D applied then the surfaces will be made refined later on. Id rig the character if Im used to making the correct topology id be at this point, but If I was learning it id not make any maps or Uvs, and test my deformation for rigging as I may need to add or re do some topo and you may end up messing up the Uvs, and the maps are useless.

You got to find a workflow that suits you, and be open to improvise and adapt to new problems as they arrive.



No worries. Tons of great info in here.

It’s been nothing but improvising and adapting as it’s almost all new, but…

I’ve gotten very familiar with zremesher, and done a few rounds of manual retopo for the head in retopoflow, rebuilt all the clothes and bodyparts (in C4d) to be low poly / edge loop deformed / subdivided.

I’m not entirely sure I’ll need displacement / projection as I’ve got the head down to 45k polys and am happy with it just like that… Hair adds another 15k, but still working on lowering that. The hair’s a bit boring, so messing with zbrush at the moment to see if I can whip up something more stylized.

The only high poly item left to tackle is the shoes.

Not sure what to do about the 3d laces. Since the foot only bends at the ball and ankle, and the laces are between those two points, could I just do a very simple low poly retopo on the shoes without laces, make sure all 3 edge loops for foot /toe flex are below laces, then just parent the laces t the top half of the foot or something when it comes back from mixamo? If it’s just additional, doesn’t need to flex or deform, then there’s no weighting issue, right?

Or… can the displacement fully handle them even with just basic shoe geometry. (I’ve done a bit of painiting, and a couple test runs on uv unwrapping, but haven’t gotten to displacement or other maps yet, and wasn’t entirely sure I’d need to.)


Point of clarification: Is “base cage mesh” a reference to the individual projection cages (full res versions of each body part toward which the low poly subdivided versions projected) as a group, or are you somehow referring to a singular combined object?

Also, I haven’t located any tuts that break things down as you do in step 2 with different layers being used for different levels for deformations vs details. Are you aware of any resources that expand on this? In all my experiments, I’ve ended up with 3-4 levels of divison prior to projection, and while I think I understand the basic concept of what you are trying to achieve in this step, I’m not clear how that would unfold.

I guess I’m a bit confused by the step by step reprojection of lowest level of zbrush versiont o lowest level of retopo version, etc.

In my particular case with the head, for instance, there is a single high poly version (triangles), and a single retopo step in which it is reflowed, quadrangulated, and poly count reduced by over 95%. I understand what the lowest level of the retopo version means in that case (no divisions yet)… but there is no “lowest level” of the target. Is it still a multi stage process in such a case?