Been doing this practice model of Felicity Smoak from Arrow, but just can't seem to get her right. Has anyone got any suggestions / opinions / paintovers they can offer to help me get her looking more recognizable?
Here's some of the face photos I'm using for reference:
Here's the model sheet I did up:
Here are the full body shots in Max:
Here are some closer up shots of the face:
And a bunch of wires just in case people want to see them:
What is your goal with this? Honestly it looks kind of PS2 era, or mobile maybe. If that's what you're going for, cool.
The thin, high contrast stripes are rarely going to look good on a screen. If you look at the photo, the stripes seem a bit thicker, and also are only slightly different colors. Right now the colors are very flat. I think you could benefit from adding some more variation, and also baking in some lighting, considering this is so low poly.
The skin could really use some variation. Also, i think she is a tad more pale, with more makeup on the face. Just more variation overall.
I don't have a specific platform target for this character, but I do have a couple of little never ending hobby projects I'm playing with and one of them is a mobile game. So I guess I'm trying to make sure that workflow I'm practising has the ability to operate at that level.
I agree with the stripes, I only chose that dress because I wanted to play with cloth sim in max and baking it to a low poly mesh (all her other dresses are *very* tight). Maybe I'll change it for one of her more "well known" outfits.
Re the skin and makeup, yes that was what I was thinking of attacking next. I'll see how I go on the weekend and post some more screenshots.
I'm assuming this has a normal map because of the ear, however your mesh has a lot of gauroud shading artifacts so I'm guessing you didn't bake the normal map from a high poly model?
I'd suggest you decide where you're going with this because it's hard to give you feedback without that information. Your topology is really messy, but how to fix it really depends on what your needs and goals are.
I did start as a sculpt in 3D-Coat which I then retopologized and baked. Here's a big collage that shows her at different stages in my process:
The last row looks nicest because of the turbosmooth but the cost of course is the polycount jumping from 1748 to 6970. I know that doesn't sound like much, but I guess what I'm aiming for is a workflow that will let me make a base mesh that can be used out of the box for a mobile game or beefed up with subdivision if neccessary for a PC game.
Keep in mind too I'm just a lone developer, so I'm trying to juggle "good enough" with "quick enough"
You mentioned you thought my topology was really messy, I was a wee bit surprised by that. It's not perfect and the silhouette on the top of the head is pretty chunky (which I didn't worry about for this model because I knew I was putting hair over it).
But for the most part I thought I'd got it mostly fairly clean. If you had to give it (the topology) a score between 1 and 10 (really bad to really good) where do you reckon it would be? And was it just a few problem areas that caught your attention or did the whole mesh seem on the nose?
1) Your normals are obviously not displaying correctly in max, this could be any number of things but to me it looks as though you altered the mesh after you baked the normal map. I'm basing this on the the fact that your side profile looks completely different in 3DS Max than it does in 3DCoat:
Tangent space normal maps store normal data from the high poly model relative to the surface of the low poly model. Meaning if you alter the low poly mesh after you have baked your normal map, you will need to re-bake your normal map again. If you don't, your shading will be broken (exceptions for flat surfaces)
Since your low poly should match your high poly almost exactly, you'll probably need to go back and fix your sculpt then refit your topology to it.
2) You shouldn't be using turbosmooth, theres no reason to use it for a low poly mesh since the geometry needs to be added with artistic judgement in mind.
If you're using turbosmooth to smooth out the shading then don't, your problem is you don't have a clean normal map bake. Once properly baked a mesh should look exactly like its high poly counterpart except with a more facetted silhouette.
3) Never offline render (meaning scanline rendering, v-ray, etc) a low poly model. The AO will simply make your model look flat and bad.
Without knowing what it was for, I'd rate your topology a 4. The limb topology is alright. The problem is mostly in the face, you kept it all quads but you didn't maintain good edgeloops around the features so you have uneven polygon density, too much geometry in the nose and not nearly enough around the eyes.
There are lots of good examples of good facial topology around the forum but this is my favorite.
Hmmm... that's new information for me about the tangent normal maps. I had thought they would have been able to tolerate small changes but I can see now what you mean (damn).
Re the turbosmooth, I was just using it to show the mesh in different states in the max viewport. Though now that I think about, based on what you said about mesh changes breaking tangent normal maps, how come the turbosmooth made it look better instead of worse?
Re the offline rendering, I guess that extends to the in viewport progressive rendering of maxs realistic shading mode? But the normal shaded mode looks so bad
That's a beautiful example of face topology too, I'll be digging through that these holidays
And yes, everyone's feedback has been very helpful. Thanks heaps for all the good advice!
Just a thought too: if I wanted to adjust a low poly mesh without having to go all the way back to the sculpt, could I do it by converting the normal map to a world space normal map, making changes to the mesh then converting it back to a tangent normal map?
Re the turbosmooth, I was just using it to show the mesh in different states in the max viewport. Though now that I think about, based on what you said about mesh changes breaking tangent normal maps, how come the turbosmooth made it look better instead of worse?
The reason why the shading is improved by Turbosmooth is primarily because of the way gauroud shading works, particularly because it looks worse in two instances:
1) There's not much geometry. More geometry will produce better gouraud shading so turbosmooth will usually result in better shading
2) Extreme changes on irregular surfaces (without enough geometry). Since you don't have support edges, turbosmooth is smoothing out the crevices and creases in the mesh which makes the gouraud shading less noticable
When you bake a high poly to a low poly, the normal map corrects the shading and the gouraud artifacts disappear. However when your low poly is altered afterward the gouraud surface shading artifacts become visible again because the surface no longer matches that which the normals were baked to. Since you put a turbosmooth modifier on it though the additional geometry and smoothing makes it less noticable.
Re the offline rendering, I guess that extends to the in viewport progressive rendering of maxs realistic shading mode? But the normal shaded mode looks so bad
Don't worry, it only looks bad in the normal shaded view because of the gouraud shading artifacts.
Once you have a correctly baked normal map, you should not see any gouraud shading artifacts.
Hey there, sorry to resurrect a (recently) dead thread, but I just thought of a question following on from the discussion above regarding reshaping low poly meshes after baking normals into them:
How come animating doesn't make it go nuts?
Is there some special thing built into rig based deformation that adjusts normals on the fly?
Understand normal maps as instructions on how to reflect light back into cameras.
Instead of say Pixel at 0,0 reflect Red 255 at all times (Diffuse map), a TANGENT (this is the one that doesn't freak out based on world coordinates) normal map does something similar, but to light.
It tells light to bend this OTHER way (contrary to the mesh surface normal), if a camera sees it, at a particular pixel.
And that's how you falsify depth, by bending light.
Replies
The thin, high contrast stripes are rarely going to look good on a screen. If you look at the photo, the stripes seem a bit thicker, and also are only slightly different colors. Right now the colors are very flat. I think you could benefit from adding some more variation, and also baking in some lighting, considering this is so low poly.
The skin could really use some variation. Also, i think she is a tad more pale, with more makeup on the face. Just more variation overall.
Thanks for the feedback, lot of good points.
I don't have a specific platform target for this character, but I do have a couple of little never ending hobby projects I'm playing with and one of them is a mobile game. So I guess I'm trying to make sure that workflow I'm practising has the ability to operate at that level.
I agree with the stripes, I only chose that dress because I wanted to play with cloth sim in max and baking it to a low poly mesh (all her other dresses are *very* tight). Maybe I'll change it for one of her more "well known" outfits.
Re the skin and makeup, yes that was what I was thinking of attacking next. I'll see how I go on the weekend and post some more screenshots.
I'd suggest you decide where you're going with this because it's hard to give you feedback without that information. Your topology is really messy, but how to fix it really depends on what your needs and goals are.
I did start as a sculpt in 3D-Coat which I then retopologized and baked. Here's a big collage that shows her at different stages in my process:
The last row looks nicest because of the turbosmooth but the cost of course is the polycount jumping from 1748 to 6970. I know that doesn't sound like much, but I guess what I'm aiming for is a workflow that will let me make a base mesh that can be used out of the box for a mobile game or beefed up with subdivision if neccessary for a PC game.
Keep in mind too I'm just a lone developer, so I'm trying to juggle "good enough" with "quick enough"
You mentioned you thought my topology was really messy, I was a wee bit surprised by that. It's not perfect and the silhouette on the top of the head is pretty chunky (which I didn't worry about for this model because I knew I was putting hair over it).
But for the most part I thought I'd got it mostly fairly clean. If you had to give it (the topology) a score between 1 and 10 (really bad to really good) where do you reckon it would be? And was it just a few problem areas that caught your attention or did the whole mesh seem on the nose?
1) Your normals are obviously not displaying correctly in max, this could be any number of things but to me it looks as though you altered the mesh after you baked the normal map. I'm basing this on the the fact that your side profile looks completely different in 3DS Max than it does in 3DCoat:
Tangent space normal maps store normal data from the high poly model relative to the surface of the low poly model. Meaning if you alter the low poly mesh after you have baked your normal map, you will need to re-bake your normal map again. If you don't, your shading will be broken (exceptions for flat surfaces)
Since your low poly should match your high poly almost exactly, you'll probably need to go back and fix your sculpt then refit your topology to it.
2) You shouldn't be using turbosmooth, theres no reason to use it for a low poly mesh since the geometry needs to be added with artistic judgement in mind.
If you're using turbosmooth to smooth out the shading then don't, your problem is you don't have a clean normal map bake. Once properly baked a mesh should look exactly like its high poly counterpart except with a more facetted silhouette.
3) Never offline render (meaning scanline rendering, v-ray, etc) a low poly model. The AO will simply make your model look flat and bad.
Without knowing what it was for, I'd rate your topology a 4. The limb topology is alright. The problem is mostly in the face, you kept it all quads but you didn't maintain good edgeloops around the features so you have uneven polygon density, too much geometry in the nose and not nearly enough around the eyes.
There are lots of good examples of good facial topology around the forum but this is my favorite.
Hope this helped
Hmmm... that's new information for me about the tangent normal maps. I had thought they would have been able to tolerate small changes but I can see now what you mean (damn).
Re the turbosmooth, I was just using it to show the mesh in different states in the max viewport. Though now that I think about, based on what you said about mesh changes breaking tangent normal maps, how come the turbosmooth made it look better instead of worse?
Re the offline rendering, I guess that extends to the in viewport progressive rendering of maxs realistic shading mode? But the normal shaded mode looks so bad
That's a beautiful example of face topology too, I'll be digging through that these holidays
And yes, everyone's feedback has been very helpful. Thanks heaps for all the good advice!
I doubt it would work though because your WS normal map would still be based on a TS map that isn't correctly baked to your lowpoly.
I've never tried it though so you'd have to test it to be sure.
The reason why the shading is improved by Turbosmooth is primarily because of the way gauroud shading works, particularly because it looks worse in two instances:
1) There's not much geometry. More geometry will produce better gouraud shading so turbosmooth will usually result in better shading
2) Extreme changes on irregular surfaces (without enough geometry). Since you don't have support edges, turbosmooth is smoothing out the crevices and creases in the mesh which makes the gouraud shading less noticable
When you bake a high poly to a low poly, the normal map corrects the shading and the gouraud artifacts disappear. However when your low poly is altered afterward the gouraud surface shading artifacts become visible again because the surface no longer matches that which the normals were baked to. Since you put a turbosmooth modifier on it though the additional geometry and smoothing makes it less noticable.
Don't worry, it only looks bad in the normal shaded view because of the gouraud shading artifacts.
Once you have a correctly baked normal map, you should not see any gouraud shading artifacts.
How come animating doesn't make it go nuts?
Is there some special thing built into rig based deformation that adjusts normals on the fly?
Instead of say Pixel at 0,0 reflect Red 255 at all times (Diffuse map), a TANGENT (this is the one that doesn't freak out based on world coordinates) normal map does something similar, but to light.
It tells light to bend this OTHER way (contrary to the mesh surface normal), if a camera sees it, at a particular pixel.
And that's how you falsify depth, by bending light.