dasfrodo

So for the last couple of days I've been trying to get really deep into digital color science and all the baggage that comes with it. This is all in preparation for upcoming projects and the desire to understand this topic once and for all, at least the basics. So far everything has been working out, from Input Transforms over ACES to Color Spaces etc. This all changed when I got the good old topic of Alpha (oh god help me please) As far as I understood now, and from a seemingly very knowledgeable source, there are basically two types of color encoding with Alpha: Premultitplied Alpha / Premultiplied Color / Associated Alpha Straight Alpha / Unmultiplied Alpha / Unassociated Alpha Before I start, we have to fundamentally clarify two things, important for terminology: RGB describes the amount of color emitted. Not the brightness, or how strong the color is, just the amount of color that is "emitted" from your screen, for each primary color. Alpha describes how much any given pixel occludes what is below it. tl;dr: RGB = Emission, Alpha = Occlusion Premultiplied Alpha ... probably has the dumbest name ever, because intuitively you'd think something is multiplied here, right? Well, that's WRONG. The formula for blending with Premultiplied Alpha looks like this, where FG is Foreground and BG is Background: FG.EMISSION + ((1.0 - FG.OCCLUSION) * BG.EMISSION) What this comes down to is that premultiplied basically saves the brightness of each color independently from the Alpha, and the Alpha just describes how much of the background this pixel will then cover. This means that you can have a very bright pixel and it's Alpha set to 0, so it will be invisible, but the information will STILL be there even though the Pixel is completely transparent. Blending works like this, where foreground is our "top" layer and background is our "bottom" layer that is being composited onto. Check if the current pixel has some kind of occlusion (Alpha <1) in the foreground Scale the background "brightness" or "emission" by the occlusion value (Alpha) (BG Color * FG Alpha pretty much) Add the emission from the current pixels foreground (BG Color from 2. + FG Color) Straight Alpha ... is considered to be a really dumb idea by industry veterans, and often not even called a real way to "encode color and Alpha". The formula looks like this: FG.EMISSION * FG.OCCLUSION) + ((1.0 - FG.OCCLUSION) * BG.EMISSION) What this means is that Straight Alpha multiplies the pixel emission by the occlusion (Alpha), as opposed to having the final emission of the pixel independently saved from the Alpha. If you've every opened a PNG in Photoshop this is pretty much exactly what Straight Alpha is. There is no Alpha channel if you open a PNG in PS, just a "transparency" baked into your layer. All the pixels that are not 0% transparent are not their true color, as Premultiplied Alpha would describe it. I have not read this terminology anywhere, but personally I would kinda call this a "lossy" form of Alpha, since the true color values are lost and are not independent from the Alpha, unlike Premultiplied Alpha. Why am I telling you all this? Fundamentally I just want to check if I understand this concept, because there is so much conflicting information on the internet it's not even funny. I am so deep in the rabbithole right now that I question if some softwares even use the terminology correctly, and C4D is one of them. You know how C4D has this nice little "Straight Alpha" tick box in the render settings? Well, according to the manual it does this: Am I completely crazy now or is this not EXACTLY what I, and the Blogpost I linked above, describes as Premultiplied Alpha? Because we have RGB and Alpha as separate, independent components? Another example, if you just search for "Straight Alpha" on the internet, you might find this image: This is the same story as above. Doesn't the Straight Alpha example look exactly like Premultiplied Alpha, and the example for Premultiplied Alpha like what Straight Alpha really is? I truly feel like I'm taking crazy pills here, and I hope someone more knowledgable in the whole Color Science / Compositing field with can tell me where the hell I am wrong. Did I misunderstand how these two concepts will actually look like in practice, did I miss some important detail, or is there just so much misinformation about this topic EVERYWHERE? If you've made it to here, thank you for listening to my ramblings. I hope I can be enlightened, otherwise this is going to keep me occupied for forever...

That is exactly what I thought I was doing wrong but I didn't know how to do it or fix it 🙂 thank you!

Hey bezo, thanks for the reply. I don't quite understand what your second approach is. Can you elaborate a bit more please?

Hi community, long time no see 😄 Due to circumstances (life happens ey?) I will probably be returning to C4D. Just to have a look at the new features I have missed out since R21 (which is a lot) I'm trying to do a bit of Scene Nodes. I have some stuff already working, but I'm currently stuck at iterating through multiple children. I have a very basic setup where I want to clone some spline onto the points of multiple objects that are children of the Nodes Spline. However, the spline only gets cloned onto the first cube. The second one is, as you can see, slightly offset. However, even if I change the order of the cubes nothing changes, so I suspect that I'm missing something to do with transforms / matrices / positions. Unfortunately my google skills either suck at this point or not many people are posting about Scene Nodes on the internet. Any pointers, please 🙂 ? Thank you!

This is why I love InstaLOD. It's not AS important for offline rendering but if you're going to create realtime ready models, like we do right now mostly, then it's your best friend. If I have a product, let's say a motor with a bunch of screws and hex nuts and whatnot, I will select the main body and set the max sag to the highest I can get it, and solely control the detailing by the max angle, which I set very low. Then for the smaller details like screws, I set the angle to something like 22 degrees and control mostly by lowering the max sag. This way I have precise control over which parts get how much detail. In my experience controlling with max degrees works better with cylindrical and "organic" shapes. What working with max angle as "main constraint" also helps with is that cylinders that are inside each other will get subdivided evenly, instead of unevenly. What I mean by that is this: Subdivision controlled by max angle, subdivision is even Subdivision controlled by max sag, subdivision is uneven Controlling the subdivision with max angle is scale independent since a cylinder will always be 360° no matter if it has a radius of 500m or 5mm Controlling the subdivision with max sag is scale dependant since it looks at the distance between the CAD model and the generated geometry So if you want perfect overlapping cylinders without the disgusting artifacting from above, use something that nicely divides 360° like 45, 22.5, 11.25, etc. for max angle and don't use max sag or turn it up so high that max angle is mostly the more aggressive setting that decides the geometry shape The cool thing about InstaLOD is that you can do this for every single part in your assembly, and get instant feedback. C4D just imports the entire thing in one detail setting wether you like it or not. So you end up with either reimporting and mixing and matching (which can take A LONG time depending on the size of the STEP) or living with not so great subdivision. It cannot be understated how much a tool like InstaLOD helps making great CAD conversions.

If you want REALLY good CAD conversion I can recommend InstaLOD. It allows you to have fine control over every single object in the STEP file, set detail settings for every single one, automate repairs, clean up the materials (so you don't end up with metal.01, metal.02, metal.03, metal.04, ...). etc. There is a free version currently, but you have to request a license manually. https://instalod.com/fsl/ We're using the tool for over a year now for all our CAD data needs and you wouldn't believe how much time and nerves this thing saved us.

I don't think people understand how futile this backlash is. This stuff is going to happen, and there is absolutely nothing we can do about it. I'm not even trying to be cynical here. Imho that's just how it is.

Uh uhm, yeah. I used goddamn once in a post. I am swearing so much. Please, think of the children! Speaking of "thread cops". You do realize that I'd have to arrest myself, do you?

I'm just going to pretend I did not see the last sentence, but I'm inviting you to reread what I wrote, and count how many times I "swore".

No, this is on Maxon. If it is that hard to find where to give feedback or report bugs, then they failed. Guess what pops up first when you Google for "c4d give feedback". This forum, this thread. With non personalized search engines like duckduckgo it's even worse. If you search for "C4D bug report" once again you find nothing usable. The first thing you findi s a goddamn link to Chaosgroup and how to report bugs for their C4D plugins. It should not be this hard to give free feedback to a company for a software that people pay money for. In a hypothetical scenario that some user wants to give feedback, what do you think, how long will they look for a page on Maxons site to give feedback? 95% will drop off immediately after they don't find it in the first 5 google results. It is absolutely beyond me how Maxon moved the entire Redshift community to this forum and completely ignored the overall C4D community. The situation is bad. Thing is, all Igor is doing is trying to keep this site alive. It's the only reason why we have the paywall.

I doubt they will do that, and if they do, it's not going to last long. Nobody in the Blender Ecosystem will pay this much for the tool. Well not nobody, but definitely not enough to make it worth the development cost.

Heh, seems like they're branching out. Smart decision I think. Them bringing their tools to Blender is huge, because it's still lacking a good particle system / tool.

Idk what's happening there, but it looks weird 😄

Something weird is happening with the teeth there I think?

That's... interesting. Last time I searched for it I either didn't find anything or I found something with it that makes it useless for what I needed it. All I can remember from last time I tried is that the transform tools were not sufficient for me.

I miss the transform tools from Photoshop. Especially Perspective and Freehand Transform. These are unfortunately dealbreakers for me in many cases.

https://parsec.app Stupidly easy to set up. works with a surprisingly shitty internet connection (~10mbit up on the sending end is already enough for decent quality). Almost no input lag. If you have a good connection you almost don't feel like you're using a remote PC. I've mixed up my local and remote PC by accident multiple times when I had Parsec open.

lol I still remember when I tried out Houdini a while back. Looking at a node, opening a dropdown, reading. "What the f*ck is a PolySoup?" Houdini has a couple of these really weird special words for things 😄

People haven't switched away from CC due to the simple fact that there are no alternatives. You can replace single softwares in that package but guess what, once you need more than just, for example, premiere (and you will need AE as well), you have to pay for the full package. Adobe is just shamelessly abusing its place on the market and they have been for ages. That's why people hate it. Their releases are god awful every time. Their development pace is pathetic for a company that big. And they gobble up good software. Substance / Allegorithmic used to be a great company. The software is still good, but the ecosystem and the forums went to absolute trash. And now recently they bought Figma. Might not mean much to many, but it's another thing that undoubtebly will be ruined by Adobes greasy hands. The difference between Adobe and Maxon is that Maxon does not have this luxury. They, unlike Adobe, do not own the market. I can decide to not pay their subscription prices and go to Autodesk, which is cheaper. Or I pay nothing and go to Blender. And in the Blender ecosystem I can get a hundred plugins for the price that would get me one big C4D plugin like X-Particles.

I was a C4D user for over 10 years and I ditched the software when I switched my job. I am pretty much a hobbyist 3D-Artist at this point and I had no issues whatsoever switching to Blender. I just cannot justify the price Maxon is asking for. If they had an Indie pricing, which people have been asking for the second they announced the subscriptions, then maybe I would have come back. But I haven't been using C4D for so long now that I really don't care anymore. Anything I want to do I can do in Blender, and more. The only thing Blender sucks at, for my usecase, is MoGraph. There is just no competition here compared to C4D. It's sad but it is what it is. I'm not even surprised they got rid of perpetuals. They all do, once they go subscriptions. It was just a matter of time. At least it took them longer than Substance, which was less than two years until they completely scrapped the perpetuals. I'm just glad I jumped ship when I did.

The biggest issue I have with geo nodes is that the average artist can't do much with it. It's just too complex and low-level. I'd argue even Houdini is easier because it has tons of Nodes that are just a collection of other low level nodes. Blender needs something like this. It needs easier, ready to use nodes that do the basic stuff like arrays. Yes, you can build them yourself (if you have the skill). And yes, there's already tons on the web. But you have to find them first, they are not updated automatically etc.

I'd look like this as well if someone jammed a spoon in my brain. Poor guy, didn't have to end like this. But I guess that's what you get for being delicious.

Thanks for the reminder. I want to take this opportunity to say f*ck Adobe. Thank you for coming to my TED Talk

What are you talking about? Especially with 3D printing you can easily use Booleans for pretty much everything, since topology doesn't matter at all. There is no shading errors to worry about. All that matters is the raw geometry. Unless your topology is uberscrewed with some reality breaking polygons it will print just fine. I've done this stuff for years.