Dude, that brings back old traumas from my studies. Because that's exactly what the math lectures were like in the first semester. The professor wrote on the board at the same speed as he spoke. He was like a machine gun: talk, write, erase the board, talk, write... 😂😂😂
There's a case to be made about INCREASING the size of the final image rather than reducing it. If you render at 200% there's 4 times the pixels, so you reduce the samples by a factor of 4. What this does is keep constant the amount of "samples per image", so in theory it should take the same amount of time; but the additional render passes used by the denoiser are almost free since they take just one sample for the most part. This to say that the denoiser has more data for the same amount of render time, so it performs better and doesn't take longer. Also, always use multi-pass denoising
this is amazing info - adds some minor extra seconds with layers of compositing but compared to the quality increase its worth it. Adding that adding grain/noise in post diffuses pixels and means that you can get away with real low samples and denoising, then add that grain back in at a higher res.
one of the best cycles optimization vids but you actually never want to use random noise seed plus denoiser since it ends up with more flickering on your final animation
@@br11bronc34 not sure tbh, don't use default denoisers usually. It has some sense tho, even small sampling differences from frame to frame causing flickering issue. According to that, I'd rather have this problem in area with deforming mesh than in my whole frame. Use temporal denoiser, combined with relatively clean image it gives best results from what I've seen
@@helloworld_2472I think they mean “can you talk more about why rendering via terminal is good for computers with low amounts of RAM? What specifically makes that technique so good?”
@@qulzam685 When you render via terminal, it doesn't have to open the project in a blender window with the ui, editor, tools, etc. Which take up memory It just opens the renderer and renders with the project's render settings, it also doesn't have to show you the image preview, which takes memory too.
Three things I really appreciate; the very useful content, the speed of delivery, and the advertisement at the end which I watched because it was all worth it.
It's a pretty good tutorial, but I think that you should render like a 2 or 3 different settings and show us a difference between them at the end. Also maybe show us how to optimize render settings for scene with volumetrics, as volumetrics are very heavy to render. Thanks for the video tho.
Agree! These are great tips yes and I've used most of them but i usually get way worse quality especially when rendering something organic. He seems like he's very confident in these methods (ik ik he said at the end that "this is an extreme example") But the confidence portrays that the quality is not affected that much so i wanna know what's the catch? How much is the quality reduced... Etc..
All of the computation power went into making the fastest Render for this Tutorial, that almost nothing was left for the microphone to export No, but awesome tutorial as always! A gem every time
It’s incredibly helpful! I’m cutting down my frame renders huge amounts!. I don’t need super duper renders, while I’m learning and building showreels of text animations , most people wouldn’t know the difference if you are putting up 5 sec samples on social media . It gets around hardware limits
ill never forget the day i heard a movie effects guy reference you in an interview they did. I cant remember where i saw it but i see why they did lol.
@wydua2049 I have no idea searching cgmatter referenced in interview doesn't bring up much. I just remember he was working on a animated movie and was saying how a video he did really helped him solve a issue he had
I always thought, there should be an option that the object should not be rendered which are out of camera boundary, but today i got to know it already exists... I feel stoopid!!
If you have finer details on your textures and if your camera is further away from your models, I don't recommend downscaling the render resolution because the denoiser will get rid of the details as it has less pixels to work with. Just a heads up.
It can all be done faster, but if you want to make a photo realistic rendering, you still want to keep the quality high. It would be ideal if they used EEVEE rendering as a template for Cycles rendering. Calculations are made in the background and put into a database. If there are any changes to your visual, it will be updated in the background.
I actually can't believe I could've been using cycles this whole time I thought my PC was just not good enough but I realized that the one thing that was making it so slow in cycles was that I hadn't setup my GPU render settings correctly
I guess persistent data does not work as I hoped with the current hair curves. I'm making characters with fur; it takes like 1.5 minutes of preprocessing/applying modifiers before anything gets sent to the GPU for rendering. Which makes sense, most of the time character is animated so the hair system needs to be built from scratch for the current mesh deformation.
Applying the subdivision and decimate modifiers makes the build faster in some scenes in my experience, despite activating Persistent Data, it consumes less ram and it's useful if you're testing the settings for your scene.
Or you know for this kind of scene you could just bake all the lighting in textures and then render almost instantaneously at full resolution with workbench
If you have a realistic scene like this... use camera calling is not a good thing, because if you have reflective materials, but behind the camera you have nothing to reflect, the result will be ugly... Or even worse, maybe by rotating or moving the camera objects appear that were previously hidden. So things appear out of nowhere. In an interior it is right to keep the things that are needed turned on and turn off the things in the other rooms that cannot be seen for example And be careful because if you lower the samples too much and use denoise you can have artifacts... it's a guaranteed click video, to do even better you could render with CPU at 10000 samples, and then switch to GPU 128 samples with denoise. so you can change the title of the video to render at 10000x faster :) I've been following you for a long time and I respect you, and I know that you are a master and know all these things, in fact I wonder why you made a video like this, which in my opinion is clearly useless and without explanation... If an entry level who wants to learn blender for the first time followed all the steps in your video he would want to never use this software again, because the result would not be nice to look at. then oh.. do as you want
For sampling and tiling I'd suggest using a number lower than the number of your cores in the gpu that way no all the cores can work as hard as they can. This only applies to the older models as you'd have to use a ridicilous amount of samples and huge resolutions
Hey, i was wondering if you could put out a tutorial about tracking objects with markers that occasionally go off screen, for example a hand, when it rotates a lot of the markers cant be tracked, is there an efficient way to track an object like that?
Before AI was all about stealing art, I took a video I rendered of a spinning Suzanne coin at 15 fps and... 480... and used Topaz to upscale to 60 fps and 1080. Over 90% of the end product was extrapolated. I then rendered out the same video. There were very remarkable differences - the AI botched the reflection something awful. But one took less than an hour, and the other an entire week. Definitely worth it if it's not meant to be a finished product. BUT you now have to contend with negative AI sentiment and AI hallucination will trigger that in a hurry.
with using Commnad line is longer than using BLENDER xD haha im on 4.2 , anyway differnece is About 2 seconds. from 16 to 17.5 second . Its looks like persistent data aint work under CMD. // EDIT: same speed i did more test.
I was always wondering if you would convert your meshes into simple shapes, and use parallax occlusion mapping on the faces how much different would it be, and how much faster would it be. (like on the default cube tutorial, just for every face)
