Frankie

V72 no CUDA, DP I'm having a lot of slowdowns when painting and I wanted to know if others had the same problem. Basically, I'm usually at around 90 fps on my current object, then every 2 minutes or so it will suddenly fall down to 14 fps which makes the painting difficult. I tried changing windows priority to background / application without luck. I tried disabling mutliple CPU optimisation in 3dcoat options and it didn't change either. Andrew, I was wondering if it had anything to do with UNDO buffer being swaped to disk from time to time or garbage collection? And if there's any way I can avoid that. I'm currently using a rather old rig for DP. I will test on my 'voxel friendly' PC later to see if there's any difference. XP32 P4 extreme 3.5 GHz, 4GB ram, 6800 256 MB vram. Franck.

Haha, or MS Paint is to Photoshop :p

V66 DP Problem with layers: After painting on layer1, I merged down with layer0 and lost the content of layer1 on 5 out of the 6 materials (I'm left with what was on layer0 on 5 materials). Undo brings back the layer1, but not the content. In other words merge-down doesn't work on objects with multiple materials. Also clone tool doesn't seem to work with DP. Maybe it's just not yet implemented? Franck.

Wonderful. PS: just sent you the file for the save/load problem.

As I said, it's the 'overlay' part I dislike. Ed: I think having the source texture material on one side and the 3d object on the other side is much clearer for me. A matter of taste and habits I suppose, but would be a nice addition nevertheless.

I know the focus is more on debuging right now, and it's obviously priority. Heck I still had to go back to maya 3d paint for my current production although I'd love to get rid of it once and for all. ( I mean the 3dpaint part of course). However as much as I think 3dcoat is starting to be THE painting app for lowpoly work, there is one thing that I'm missing over maya 3d paint. Screen space clone brush along with a display image window. This is the way we're used to work in photoshop for cloning texture from one image to another, and I managed to have a similar setup in maya. I have to say I'm not too comfortable with the 'material' stencil thingie in 3dcoat. Simply because you can't see both source and destination material at the same time. And the transparency system is so distracting. Here's how I did it in Maya. It's rather simple, I create a plane and attach it below the camera so it stays in camera view. From there I select both the plane and my object for painting. I use the clone brush in screen space and there you go, a photoshop-like clone in 3d. I think it would be quite simple for Andrew to display an image in a window (or several) and alow for screen-space cloning. That would be a great addition to the toolset. Cheers, Franck.

Still save/load problems with version 65. Imported obj for DP, painted on layer 1 (won't touch layer0) using some 'material' stencil. Added a 3rd layer in multiply mode and painted on it. I tried exporting the textures and it went fine. This part seems to be fixed and ok. Then I saved the .3b project. On reload, the original textures showed up again (the one from the original .obj), and I'm not able to see or paint layer 1 nor 2. I can send you the .3b file if you want. Franck. ED: I tried to export textures after having reloaded the .3b project just to check and I got again empty 2k textures. (the original were 128x128)

I saved the 3dcoat file, and reloaded it. The original textures (before 3dcoat painting) showed up. So 3dcoat didn't save the textures I reworked in the 3b format either (not that's a problem it was just a test). At that point I was not able to paint on the mesh anymore (even on new layers). When I tried the smooth brush it crashed. Just sent you the report. I'll send the file right away. Franck.

V64 Got a big issue when saving the textures here. I paint a rock object that has 6 material with a 128x128 texture on each material (so 6 textures in total). Painting goes fine, but when I export the textures (all at once or individually, with or without filling empty area) 3dcoat exports huge grey 2k textures with only a few lines of pixels containing the messed-up texture data. I guess there's a problem with the 128x128 export size which is apparently not recognized correctly. Andrew, let me know if you need the file to debug. Franck. ED: Also I just discovered that the smooth brush has no undo. A sharpen brush would be handy too (or smooth with negative values)

Well bleeding, padding and clamping are different things. Bleeding alows for realtime bleeding of your brush outside the UV island. Hierarchical bleeding is even better for mipmaping (the further away from the island, the larger the bloc and it blends with the previous color, like maya 3d paint does). Padding usually refers to the extention of the border color outside the UV island when exporting, usually after painting/baking. Clamping is a hardware feature at the texture sampler level where it will clamp UV values that are outside the 0,1 range to 1 (or 0 if it is negative). This as the effect to limit the interpolated value of the outer texel to 0,1 and avoid wrapping interpolation. It has no effect on UV island inside the 0,1 range (and doesn't change the content of the texture). To me, having the three options available is the best solution. Hope this helps, Franck. ed: image from maya 3d paint documentation added

V64, my UNDO problems are gone when working on multiple materials, great. Thanks Andrew. Auto-load texture is great too. Another small feature I'd like to see: The ability to set texture wrapping attributes per material: repeat, mirror, clamp (standard openGL/DX stuff) Clamp would be particularly useful to avoid the interpolation of adjacent texel when bilinear filtering of the outer texel on low-res textures and would avoid having to move UV coordinates 1 texel inward. I use that in our engine everytime. Cheers, and once again a great release. Franck.

Thanks. I was thinking about the alpha for transparency again. I know this would require per-polygon sorting if it was to be implemented the regular way. Unless (like most game engine) you enable alpha testing in the OpenGL/DX pipeline (with a 127 mid-point treshold by default). It would be absolutely cool to paint alpha cutouts and have the result in realtime. No painter has ever done this although it is actually easier to implement than having to sort polys to correctly display transparencies. Just a though. Franck.

Wonderful release. Couple of issues though: When loading the model in 3dcoat, the smallest allowed texture is 512. I'd like to be able to enter any size, down to a few pixels if I need to (as long as they are power of 2 i assume). Right now I need 128x128 for instance. Undo doesn't work for me. My test model has 5 materials (UV sets as you call them) and when I undo a stroke done on one material it fucks-up the texture in another material. BTW to me the use of 'multiple UV sets' term refers to the ability to store several UVs per vertex to be used in multi-texturing for instance. A diffrent notion from the multiple material applied to a model each with a single UVs set. Just semantics here, but I'd like to know if others think this is a bit misleading? View unshaded model doesn't work. Some small features I'm looking forward to: I'd like to have the ability to disable bilinear filtering. I'd like to have the ability to use the alpha channel for transparency instead of specularity, as it's widely used in games assets. Cheers, Franck. PS: The particular system for this test was WinXP32, P4 extreme 3.5GHz, 4GB ram, GeForce 6800 Ultra, 256MB Vram. Latest NVidia drivers.

I do like it too. Although I always wondered why UVs were required in the first place when it's microverts paint. You could as well import UVs later on, but that's another story. Both DP and microverts are usefull approaches, it's just that one is better suited at pure lowres stuff while the other is great for the hi-res baked to low-res approach. Just assign two shaders on both sides and flip bi-normal on one shader and off you go. Anyway for diffuse-only work (low-res specifically) this isn't a problem. I'm thinking of the many artists working on handheld-level resources here. Think DS or PSP for instance. That's just what I said. What if I want uneven texel ratio. Because the sole of the shoe of my character isn't as important as his face. And I want to see exactly how it looks while I'm painting it. Sure there are workarounds. And right now for me it is to skip 3dcoat all along and stick with maya 3dpaint for lowres work wich is a shame because it is nowhere near as complete as 3dcoat in term of tools. But it is wysiwyg. Of course you can spend lots of time putting details in a uber-higher resolution texture not knowing what's gonna be kept and what's not when you downsample. Thats an important point here, because I think the whole problem with the hi-to-low approach in general is often an immense waste of time spent on detail you'll not see in the end. And the DP in general might be a nice addition to the artist palette of tool for him to choose what's the fastest route to the end result. Because only the end result matters in game art. All I'm saying is when you're on tight budget and schedule on... say a PSP title, you're looking for the most direct approach to create your textures and I think DP can help. You're a dead man... just kidding Franck.

DP allows for: Overlaping and mirrored UVs. Very handy when you don't have much texture memory available for your target platform. Since it works in UV space, you can come up with various texel aspect ratio on the same model, independantly of the model topology, and save some more texture memory. Better accuracy as what you see is what you get and the texel won't go through multiple reprojection during painting and at export (like it is right now). You might not see it when working with 2k images, but try to do a character with a few 128x128 textures right now and it will show. It is important when doing 'down to the texel' work. All these things are important for game assets, especially low-res work. Hope this helps, Franck.

Right now when you import a low-poly mesh, 3dc always subdivide it (i.e. 4 million vertices if you choose a 2k map). It then paints at the vertex level and transfer the result in a texture map on the fly. This allows for generating normal maps on the fly based on the deformed geometry if you paint-sculpt the model. However if you paint color only, a direct-to-texture painting scheme (or UV paint) would provide better quality and control since it would not require the original model to be subdivided at all. Also UV paint would allow symetrical painting using overlaping UVs, which is not the case right now. Hope this helps. Franck.

Actually, Claytools uses a 8 bit scalar for masking, and that allows for smooth grandient masking and transforming. Data painting as you suggest might just be yet another step further. Much like Maya's blind data editor, where you can paint any type of data (boolean, scalar, vector, even strings) at vertex/face level. This would be usefull in combination with a scripting language associated with the brushes (custom brushes), or at the shader level (for special rendering based on those data, skinning, procedural stuff or whatever). Those should also be transferable (baked) to maps or verts (of low poly mesh). However I agree with previous posts that we need UV painting first for V3. Then I hope Andrew will have a go on voxel masking/painting and maybe at some point generalised data painting. Franck.

V47. In the tree view, an object that is hidden cannot be deleted.

About 460 KB/s so I would say average as I've got a 24Mb/s downstream line. As a reference, stuff from Nvidia website come down at around 1500KB/s. Franck.

Some notes about performances using version 47. (both DX64 and GL64) With CUDA, the brushes are damn fast, and quite usable up to 40 millions (the max I've been able to get so far, I've had a crash past that ammount with no apparent reason, had plenty of ram left. A report has been sent). However when tumbling the view, it takes a while before your brush reapears. Depending on the size of the zone you've been working on. I suppose it has something to do with the copy to the undo buffer. Also sometimes, the tumbling of the view goes to a halft for a few seconds, then back. It would be great if these could be optimized at some point as it slows down the workflow. Also I find the increase resolution operation to be really slow. Past a few million, it takes several minutes to perform. Speeding up this operation would be nice. Apart from these things, I'm very impressed with the performances overall. If these issues are to be tacled at some point it will provide an even smoother experience across the board. Which can only be a good thing. Another thing I'm looking forward to is the ability to paint/select parts and separate to another object (probably requires the voxel painting tools though). This in combination with the treeview would enable larger scene management. Split parts -> increase resolution of parts -> work on them separately -> fuse them back. A workflow I'm used to do with Claytools. Agree with some earlier post that moving sub objects around would be alot easier with a simple universal transform gizmo. At least the pivot should follow the transformed object. Can't wait for the quadrangulation. I'm sick of doing manual retopo, even if it's the best retopo tool available lol. Keep up the very good work, Franck.

This is a bit off topic, but I wanted to wish all of you a very Merry Christmas, and a Happy and Prosperous New Year. And especially to Andrew and his family, and thank him for the astounding work he has been doing so far. Cheers, Franck.

Hi andrew. As you'll be starting the quadrangulation soon, I wanted to share some idea here. I started with this reflection: Mesh reduction generally doesn't give good results on hi-density, hi-detail source material, but works well for smooth surfaces. On the other hand, retopo tools enables to subdivide a simple quad mesh to better follow a complex underlying geometry. Then why not make best of both world: 1. Quadrangulate based on a smoothed version of the voxel mesh. At this stage the user should be able to choose the smoothing level and the number of quads (as few as possible) 2. Then bring the low quad mesh to the retopo tool and subdivide the low quad mesh to snap the result to the original voxel mesh. Additionnaly, this pipeline has the benefit of adding the possibility of fine tweaking by hand with the retopo tools afterward if needed. Hope this helps. Cheers, Franck.

Still no multiple undo on snakes in v36. Sorry if this is known or has been discussed already. I'm missing multiple undo because I relly a lot on snakes for shaping things up.

Hehe. Absolutely right about the smooth brush too. They put it as a separate tool in mudbox for a reason.

Yes but what I would like is the tools to remember which preset to use. For instance I might want to carve with a certain preset profile and size and pressure etc... and extrude with another preset, another size etc... I would like 3dcoat to be able to remember all those parameters when I switch from carve to extrude.