Page MenuHome

Cycles Motion Blur Intermittently Renders Rigged Mesh With Severely Stretched Triangles
Closed, InvalidPublic


System Information
Operating system: Windows-7-6.1.7601-SP1 64 Bits
Graphics card: GeForce RTX 2070/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 430.64

Blender Version
Broken: version: 2.80 (sub 74), branch: master, commit date: 2019-07-18 10:55, hash: rB687f1df5dbe2
Worked: (optional)

Short description of error
With Cycles Motion Blur enabled, intermittent frames on a rigged mesh are rendered with stretched/ripped triangles.

This is probably related to other Cycles Motion Blur reports, but I've spent about 6 hours troubleshooting this and am hoping that there is some information here that can help.

I would give a list of "bad frames" but I've found that sometimes the same frames will render OK, and
other times, they are broken. Additionally, a frame may render with artifacts - I could advance the timeline render another frame, then go back to the "bad" frame without making any changes, and the frame that was previously "bad" would render correctly - i.e. a frame that renders with artifacts may NOT always render with artifacts - something else is in play here.

Some things that have helped "fix" frames:

Turning off motion blur fixed every frame I checked

Turning off the Multi Modifier Armature "Preserve Volume" - this does not always work
Triangulating the entire mesh - this helped a lot of frames, but didn't always work
Adding or deleting geometry may fix a particular frame, bud didn't always work
Changing the motion blur duration may fix a particular frame, but didn't always work'
Changing the motion blur position (start, center, end) didn't always work

Additionally, triangulating the mesh improved shading on quite a few frames.

Some things tried that had no effect:

Changing materials didn't fix any frames I checked.
Changing lighting or HDR environment didn't fix any frames I checked
Rendering from a different camera on a frame with anomalies, the image still had anomalies
Moving the camera had no effect.

Exact steps for others to reproduce the error

Open the attached project (sorry for the size). Render out maybe 20-50 frames - you will get at least one frame with issues.

Event Timeline

i have same issues with RC2. i presume that issues going from animated meshes with non manifold edge and other modelling issue

That would have been nice if that's what was causing it ... but it looks like that's not the (only) issue since after cleaning up the model with the 3D print tool, I still get the same rendering artifacts:

dark999 (dark999) added a comment.EditedJul 22 2019, 1:18 AM

can you upload a more simple .blend file with same issue or your cleaned meshes file?

thank you

Roger that ...

Sebastian Parborg (zeddb) lowered the priority of this task from 90 to 50.
Sergey Sharybin (sergey) changed the task status from Unknown Status to Unknown Status.Jul 24 2019, 3:21 PM

This is happening due to autosmooth, which changes topology because deformation changes angles between faces.
Changing topology is not supported by Cycles motion blur.

Solution is to disable autosmooth and use extra edge loop or mark edges sharp+edge split modifier which will only use Sharp Edges.

Thanks for the report, but it's just a unsupported configuration.

Thanks for the reply, Sergey. This is good information ... that I've never run across. I went ahead and did some tests to see what worked and what didn't, and I've found that it's not necessary to disable AutoSmooth - but you DO have to click the "Add Custom Split Normals Data" button under "Geometry Data". This is a detail that I don't think I've ever run across (not that I'm a Blender Guru), but it seems important enough that there should be a pop-up warning when someone goes to render an animation with Autosmooth enabled, without Custom Split Normals Data added to a mesh. The reason behind this is that anyone coming into Blender/Cycles who doesn't know this requirement is going to write off Cycles as a renderer almost immediately - *I* was getting close to giving up on it. The difference in blur quality is dramatic:

I re-rendered the scene and the motion blur is much improved on the main rigged mesh - but I'm seeing some bizarre anomalies on the headress wings - specifically some frames where they seemed to "spin out" for no apparent reason. Do you have any idea why the headress would kind of wig out on these frames: 242, 315, 359, 451, 546, 553, 569, 643, 677, 813, etc ? I've attached another project with just those parts. (just open the project and render any of these frames).

Add Custom Split Normals Data

Interesting find, didn't know about that. Kudos to @Bastien Montagne (mont29) :)

It seems important enough that there should be a pop-up warning when someone goes to render an animation with Autosmooth enabled

It's not that simple. For example, imaging situation when mug is rigged by a single bone, with all weights set to 1. This allows to share set across multiple shots as-is, in a linked manner (so changes to set itself are properly propagated to all shots), but still allow animators to animate that mug.
There is nothing "wrong" here in respect with autosmooth: all the vertices stay at the exact location relative to each other, so autosmooth behaves exactly the same on any frame. Having a warning in such case will be misleading and annoying.

I've attached another project with just those parts. (just open the project and render any of these frames).

Ok, there is another tricky non-intuitive thing going on here. In the configuration form your file the motion blur uses current frame, current frame + 0.25, current frame + 0.5 (motion blur uses non-integer frames, to avoid possibly too much linear trails of objects).

There is something in the animation in the frames you've listed, which makes armature to evaluate to some weird shape. You can see that if you enable subframes in the Playback menu and set current frame to 315.0, then to 315.25 and then 315.5.

I am not really sure how the file is organized, but usually this happens when you have animation on constraints enabled state. Such transitions are to not use ease fade-out and use constant change, so they happen instantaneously.

You can also try setting shutter time to 2.0, so motion blur samples armature at integer frames, at which you've created keyframes and compensate this by altering a motion blur curve to make it less narrow bell-shaped thingie.

Thanks again for the information, Sergey. I had no idea that you could play back subframes - this does show that there is some weird mojo happening in this animation, sometimes with the armature doing a full 360 degree rotation in one frame. I'm guessing this is due to something breaking in the FBX import of the animation, as once you look at the curves (focusing on the Hips bone) it's obvious what's going on.

As far as the pop-up warning is concerned, I see your point. Perhaps just adding a static warning below AutoSmooth in the panel? Something to the effect that if your mesh is deforming, to be sure to lock down your normals (using Add Custom Split Normals) before rendering to avoid artifacts? - all in the interest of improving the user experience and getting the best results ;) ... and to help people like me remember to do it - lol.

I'm guessing this is due to something breaking in the FBX import of the animation

Don't really think this is FBX related. Depending on the rotation model you can have this happen quite easily in "regular" animation.

Perhaps just adding a static warning below AutoSmooth in the panel?

Can not be fully static. If the object is not deforming or doesn't have modifier like copy normals or so the warning will be redundant. Doing a complete check is expensive and is not something you want to be doing on interface redraw.

Also, there are a lot of settings which could break some other feature. I don't think we should be adding all those possible warnings in the interface.

To me better thing to do would be to make Cycles better on detecting such topology changes and actually report them to user (currently check is rather simple, could miss a lot of topology changed, and will only do a debug print which such case is detected).