Home
Artists
Search Recent Random
Posts
Search DMs Popular Hash Lookup Tags Random
Importer
Import FAQ
Register
Partychan Telegram ThePornDude
Home Artists Posts Import Register

Cycles VS Eevee+Cycles (Patreon)

Published:
2024-02-23 22:02:39
Imported:
2024-02
Tags:

Content

Hey guys!

I have already rendered 2500 frames, there were errors, I fixed them. Total P4 ~25000 + the cum layers. These are just the introductory parts. In the meantime, I'm working on the simulations. I'm trying to understand why the settings that worked for P3 don't bring the result I want.
I'm testing the cum layer. I made a full Cycles render image for reference. I have noticed that the new machine is very fast in Cycles. I can render a 4K frame in 40 seconds. In the case of the combined method, the Eevee frame is 20 seconds + the cycles cum layer is 16 seconds. The difference is quite small. So I thought...

Cycles render: Pic1, Pic2 

Eevee+Cycles render: Pic1, Pic2 

These images have not yet been post-processed, they are raw render images.
Now I figured out that I will render the alternative version of P4 with Cycles. And so we can compare. I've rendered video with Cycles before. Halloween 2022. Because of the vacuum mask.

I hope that next time I can show well-functioning simulation WIPs.

Good weekend!

Files

Comments

Martin S

I think I have the answer, as it is something I've learnt to be aware of on a daily basis :) Semi-transparent alpha of the semen pass is most probably the culprit. It is either badly configured in the shader, or it is composited wrong. I could bet that if you throw just the semen layer over a 0,0,0 black background, it will look as intended. I have no idea what you use for your compositing stage, and if you use deep rendering, or mattes, to mix elements. Semi-transparent alpha channels(having non-integer values between 0 and 1, not counting the AA variation and MB/DOF, of course) are supposed to be a property of one and only type of objects in a scene - VOLUMES. Everything else has a clear boundary that a ray can hit with absolute certainty. Volumes on the other hand are for the most part just clouds of 3D pixels(voxels) with transparency value. Since they have no geometry, therefore no exact boundary, they make use of that mentioned transparency(or otherwise known as "density" attribute), which accumulates along the camera ray path through the voxel cloud. The moment that ray goes through all volumes it the scene and exits it has one of two options - to terminate into some geometry, or to terminate into infinity (empty scene space with nothing behind it). Therefore we have two options of termination: 1. Void, empty space, therefore - final pixel alpha = 0 + accumulated alpha for each voxel on the path (and/or AA, MB/DOF transparent samples) 2. Geometry, occupied space, therefore - final pixel alpha = 1 + any accumulated semi-transparent samples mentioned above In both cases the values are clamped in the 0-1 range, as this is alpha. (if for some reason they are not, they should be!) The spatial accumulation of alpha values along the path through the volume will give you semi-transparent alpha, which defines how that volume will mix with other elements inside it or behind it. So, now to link up that over-descriptive blob of info with the actual issue: The semen element is a liquid, therefore it has a clear boundary between its interior and outer void, which we define through geometry. That same boundary is also what we use to bend light rays, according to the real-world material density difference, which is described through the Index of Refraction (IOR). So, if we take what I said above about the types of objects and their alphas, we can conclude, that since the semen is defined by geometry, it has to have alpha=1. Many people would immediately say - yo, that's wrong, the material we're trying to replicate is semi-transparent, it cannot have the same solid alpha=1, as a concrete wall, for example! And they would be wrong! It's not semi-transparent, but it's actually semi-REFRACTIVE(or semi-transmissive, same thing, really) instead. "transparency" (the inverse of "opacity") defines how much less than 1.0 is the final pixel value of the alpha of a 3D object. That's all. And this is already defined by the nature of the 3D object. It's either geometry or volume. And here comes the issue with transparency - it affects all other lobes/components of a shader, just by mixing them by its value with the ones of the object behind. This is not how the real world works, though. Stuff behind a transmissive(refractive) object is visible through it, because actual photons (we can call them rays in the CG field) are going through all the different materials, and as a consequence their direction, color and intensity values are being affected at every step of the way. This is all calculated by the engine and resides in the transmission/refraction pass. Hence why if a scene object is made of geometry, then it must have an alpha of 1.0 The transparency thing, has no real, actual analogue in the real world. Even the air around us is not transparent, but is actually transmissive, as it bends, absorbs and colorizes the light. Just on a way less perceivable scale. If it was called "presence" instead of "transparency", it could've been way less of pitfall for many people. :) As a conclusion, we can say that due to the strong basis of raytracing engines on real-world physics, transparency (non-solid alpha) cannot be used as a reliable, physically-correct part of the rendering and compositing processes for any objects. Volumes included, as they have to use density, to properly define their variance in space. It's great for hacks, though! :) And definitely magnitudes cheaper than transmission! This is the sole reason why glassware in games has always looked like shit, and just now with the RT capabilities of modern cards, those old cheap tricks and hacks like "transparency" are on their way out. So to tie all this to your issue - Your shader is probably fine and renders with proper diffuse/reflection/refraction. But even then, if you allow an object to render with alpha values lower than 1.0, (or you multiply a proper 1.0 alpha or the RGB with a transparent one) you will have exactly the same issue as in the renders. Another important thing - make sure you render a shadow pass for the skin geo (the semen being the occluder). The diffused matter in the liquid should not just scatter light, but also absorb it. Good example are real-life clouds. They are super white and bright on a sunny day, as they internally scatter light like crazy, but still they cast shadows on the ground. Sorry for the absurdly long comment, but I decided to back up my guesses on the issue. I could easily be completely wrong, but I'll take my chances. :D Double check your alphas and how exactly the rendered stuff is put together. Compositing math can be a real bitch sometimes! :(

Martin S

Also, after sleeping on it, I took another look and noticed that you have fog on the Eevee-based renders. If the Eevee renders are just beauty passes of the whole scene and the semen passes are just the Cycles renders of the sim geo, then no, you cannot composite them correctly. If Blender had support for DeepEXR you can, but I checked and there isn't, without any plans for it in near future, as apparently the VFX ways are not the Blender ways... :) The remaining option is to have the Eevee fog thrown out on a dedicated pass, with a matte(holdout) by everything else in the scene. Then you layer passes as follows: 1. the Eevee beauty render pass of the whole scene. (excluding the semen object and fog) 2. the Cycles semen beauty render pass. 3. the Eevee fog render pass with holdouts. The Deep rendering tech is a godsend, in such cases. Hence why pretty much all big modern engines support it. Pity, that Blender is so great for asset creation and animation, and Cycles is so bad for some more advanced rendering. Please drop a comment what you think and if you have any revelations regarding the discrepancies. I'm genuinely curious.

wildeerstudio
>>130563661

Well, on a theoretical level, you're probably right. I'm just learning these. In practice, I see that there are two problems. One is that Eevee and Cycles have completely different shadows. It is clearly visible on Lara's neck and around her eyes. The other problem is volumetric light. This can perhaps be solved with compositing. If you think we can continue this on Discord. :)

Larry Hunt

I don't know what any of that means except Lara's tit's look amazing to me! Wish I had some insight , I could share with you....