I'm rendering a patch of Speedtree trees as part of a larger scene for a demo reel breakdown. Many of the rendered frames have weird glitch in the trees. There's also at least one frame where the terrain geo is visible, even though I specified it as a holdout in the render layers.
Any suggestions on how to eliminate this problem? This is also happening, to a lesser degree to the full scene render.
Thanks,
Niel
Hi Niel,
By any chance, do you have access to the render logs for each frame? For example, if you happen to be using Deadline as a render controller, you can often look at the render logs to see where a problem might be occurring. In this case, we're curious if the same version of Terrgen is being used by the render nodes rendering the incorrect frames, as the version being used by render nodes rendering frames correctly. Also, the render logs may show errors when trying to load the files.
Would it be possible for you to send us a couple of consecutive frames, like 72 and 73, in which one frame renders fine and the next does not, so we can compare the two? This may help us determine where the problem and what's changing between frames.
Does the render always fail on the same frames, or is it different each time the shot is rendered?
Are you using the Colour Management feature, and if so, is the render nodes which are failing to render the image correctly, finding the colour management config files? If the OCIO environment variable is being used at your studio, has it been set on the render node?
Niel,
To clarify, can we see frames that rendered correctly from the "full scene render" that you mentioned, to compare them with the holdout version that rendered incorrectly. Thanks!
Hi Kevin,
I've attached the two Deadline render logs you requested. I've asked my TD about the version of Terragen we're using on our farm. I'll update you when he responds.
The render does not fail on the same frames. I have been re-queuing the bad frames, and in most cases, they will render correctly. Also, it doesn't appear that specific nodes are the issue. Any node will render good and bad frames.
I am not using the OCIO color feature for my renders.
The "full scene render" I was referring to was from a different camera. The shot in question is one we've created specifically as a demo reel shot. So, I haven't done a full scene render with it.
Hi Niel,
Thanks for the logs and additional information. From the logs I can see that both frames rendered using Terragen 4.6.31. By any chance do you know if these frames were rendered on different render nodes?
Yes, each was rendered on different render nodes.
Hi Niel,
By any chance do you have two consecutive frames from the "full scene render" in which one of the frames has the error and the other frame rendered properly? Thanks.
I don't. With the previous renders, I re-qued all the bad frames. I can re-render one of the full scene renders, and get what you're asking for that way. I should also clarify. I should not have called it a "full scene render". I rendered all the foliage together, but as a separate layer. I didn't render out the entire scene, terrain and all. Apologies for the confusion.
That makes sense about re-rendering the bad frames in order to have the complete sequence. Ultimately, what we're after is a frame that's "bad" followed by a frame that rendered properly, so we can compare them. We want to use the same render settings that generate the condition in the first place, so it's okay if it was the foliage sequence as a separate layer...just so we're comparing the same thing.
Hi Kevin,
I've done another render with a larger patch of trees. The problem appears to be worse on this one. I've attached several deadline render logs, showing the output of good frames next to bad. Let me know if you need anything else.
I just did a quick count. My render is 280 frames, 128 of those are bad.
Could it be something with the render cache?
Hi Niel,
Would it be possible to see the rendered frames for a couple of those images, like frame 75 and 76? Which ever pair of frames that best illustrate the problem. You can send the frames to support@planetside.co.uk if you want.
It also appears from the logs that you've cached the GI for every frame, and that Terragen is blending the GI cache from the 5 nearest frames to the one being rendered. Is there a specific reason for caching every frame of the GI? I've always understood that for an animated sequence it's best to use a sparse GI cache, for example, rendering the GI for every fifth frame and then blending between the 5 nearest frames to the one being rendered. For example, when rendering frame 75, a sparse GI cache would use frame 65, 70, 75, 80, and 85, and this results in a smoother GI solution without flickering.
This is what the log file reports for the GI cache for frame 75:
2023-05-31 13:12:25: 0: STDOUT: drsp_101_001_150_trees_01_v01_gicache_0073.gic
2023-05-31 13:12:25: 0: STDOUT: drsp_101_001_150_trees_01_v01_gicache_0074.gic
2023-05-31 13:12:25: 0: STDOUT: drsp_101_001_150_trees_01_v01_gicache_0075.gic
2023-05-31 13:12:25: 0: STDOUT: drsp_101_001_150_trees_01_v01_gicache_0076.gic
2023-05-31 13:12:25: 0: STDOUT: drsp_101_001_150_trees_01_v01_gicache_0077.gic
Kevin,
I went ahead and sent you all the rendered frames from that scene via Hightail. Please let me know if you didn't get it.
Sorry, I forgot to answer your question about my GI cache. With a camera move, I thought saving a cache for every frame would provide more info for the final render, reducing/eliminating the render flicker.
That's understandable with a camera move. If you've seen the shot we did for the Terragen for VFX series, of the planes coming out of the clouds, we generated fewer cache frames at the beginning and end of the shot, like every 10th frame, but when the camera circles around the one aircraft up close, we had to render more cache frames, something like every third or fourth frame. So our cached frame numbers was something like 1, 11, 21, 31 ... , 151, 154, 157, 160... 200, 210, 220 .. etc.
I'll give that a try on my next render. Do you think that could be one of the causes for the flicker?
Now that I see the frames, I'm thinking it's not the GI. It looks likes a population is not getting loaded for a given frame. I'll run this by Matt!
Neil, one thing that I do see from the logs, is that some render nodes are using Terragen 4.4.67 and others are using Terragen 4.6.31. For example, frame 80 was rendered in 4.6.31 on the render node Falcon04, while frame 81 was rendered in 4.4.67 on render node WS02.
Uh oh, I thought my TD fixed that. I'll make sure we update all nodes on our farm to 4.6.31.
I'm not a programmer, so maybe I shouldn't even be asking this question. But in several of the logs I'm noticing this line to text:
WARNING: Python version for 'Terragen' plugin is not specified! Defaulting to Python 2.
Does TG require a certain version of Python?
Terragen doesn't require Python, so that's probably a warning notice from Deadline.
Do the logs you provided above go with the image sequence you made available to us? We just want to be sure we're comparing the same things.
Also, what if any, are the holdout items in the project? Are some tree populations being used as holdouts?
Yes, the render logs go with the image sequence I sent you.
The only holdout item is the terrain. Which you catch a glimpse of at frame 183.
I just sent you the scene file to the support email.
Hi Kevin, I'm the TD working with rolland1013. Are there any flags or env vars we can add to provide more verbose logs?
Thanks,
Tom
Hi Tom,
No, unfortunately there's no flags or env vars that provide additional detail and information to the log file.
Also, just following up on our request from Friday via our support system, would it be possible for you to render a pass with the trees and ground together; without any holdouts? Basically, a beauty pass of the trees and ground together. We don't necessarily need to see the entire sequence. Ideally, two frames should suffice, where we can see a frame that renders correctly, followed by a frame that has the problem.
Hey Kevin,
I just emailed you a mov of the rendered shot with visible terrain. As you can see, the problem is gone?? Any ideas?
I'm running into another type of rendering glitch. I figured I'd add it to this thread instead of starting a new one. I've run into this issue before, but never to this extreme. What is causing the little black squares to pop up? Any suggestions on how to rid my render of them? In the past, when they would show up, they would usually be in only a few frames. I'd simply recue those frames, but there are in most of the frames on this one.
I don't know what would be causing this, but it looks as though there are NaNs in some of the samples which are then spread out over neighbouring pixels.
If I were troubleshooting this I would try to narrow down which parts of the scene are causing it. First I would disable all the populations and any other objects in the scene besides the landscape and sky, and render some frames of that. If the problem remains, then I would try to narrow down which surface layer or shader is responsible. If the problem is gone, then I would introduce some of the other objects to see what's causing it.
If you're willing to share the project with us at support (and all assets it's referencing), then we will look at in detail. We might notice something unusual that we can focus our tests on.