I’m conducting some testing for a new project with new 4.0, actually I’m coming from 3.3LTS.
I don’t know if this a “feature” from previous version or some kind of bug, I can’t find any explanation yet or something similar in the documentation.
The thing is the point light radius is actually behaving like a sphere of light with a hard boundary, instead a smooth falloff transition as we have on 3.3. I guess something similar could happen to spot light maybe?
Anyway, I would like to know if this is something to be expected and how to control it, I like better the previous behavior of the area parameter and would like to keep or mimic what we had.
Left is 4.0 -------------------------------------------- Right is 3.3
Its expected. Its actually a Bug in previous blender versions. The Radius is defining the size of the actual source of the light. If you want a smooth soft light with a large radius, you could go for an area light and set it to a disc shape. It then has the same “roundness” and softness like in your example but doesn’t overlap with the geometry.
As I said, I come from 3.3. Don’t know if this is a feature or a bug, if it’s a feature or expected behavior I would like to have some documentation and know how to control it ( I really dislike this look). In the meantime I will report it as a bug, as @joseph mentioned.
It’s expected behavior. I registered bug yesterday too, there’s one or two every day. They changed how point lights work. Basically just don’t have it’s radius intersect mesh. They made it more physically acurate, so that whole radius is light source now and shouldn’t intersect mesh
This is opinion and not the intention of the topic.
To be honest I appreciate the effort on doing things more “physically accurate” but keep consistency and flexibility must be a top prior for 3D DCC. I’ve never see a light acting like that btw. Anyway, for me the problem its related with the lack of consistency with previous version files, the lack of control on hard/soft shadows with point & spot lights and the lack of some degree of control of the new falloff.
I don’t like to sell everything for “more realism”. I like the new Principle shader, good furnace test, preserving energy, MGGX…but it doesn’t have Thin film as Arnold, Internal tint for glass as Mtlx or Dispersion as VraMtl. Furthermore, It can’t represent physically effects as phosphorescence or iridescence. Does this make a worst shader? For me it’s a good option meanwhile I could keep consistency and I can have flexibility with it.
Something similar happening to new AgX gamut. I’m involve on some stylized rendering job lately and I love they keeping Filmic, because it gives more saturated look and that’s what I am aiming for. I’ve made some test and AgX doesn’t work well with color lights (and blackbody!), it’s shifting hue and killing the saturation. More realistic? I’ve seen RGB LEDs in real life and laser points with high power and keeping saturated look. Back in the days also I´ve filmed music concerts with Canon 5D Mark II and Sony FS7 and got saturated colors and complete different look one from each other brands. So, AgX its more realistic? It depends on how you define realness, of course. For me its a big step (and also a new mess in a ACEScg pipeline) for VFX, but not for NPR or Stylized.
This escalate in pursuit of realism doesn’t make sense to me, it gives me the vibe of those highly realistic games with a complete lack of fun and good gameplay. We are 3D artist, we need flexibility and tools, no realism on-off switches.
TL;DR
I appreciate the effort of Blender team to keep pushing their software, but I would like them to keep consistency between versions and a high level of flexibility, I wont sell this for more physical accuracy
Well 4.0 was for breaking consistency, that’s why it’s such a big jump, it’s a breaking release. I would advise, and I’m sure people at Blender too, that projects you started in 3.6 you finish in 3.6, especially since it’s LTS and will receive updates for another year or so. That’s a common practice and good practice for every software, and how industry works in general. As for new projects, you can start them in 4.0
One thing about realness is that there’s not really a definition for it, everybody things they have one, but the fact that multiple opinions exist and are in constant conflict proves that there’s not. But the changes Blender is making is necessary to make Cycles industry standard and work with MaterialX, USD and etc. So basically if every other softwares uses points lights certain way, Blender would risk falling out if they didn’t too.
But of course Blender is never afraid to be a loner if they think industry has it wrong. In this case, good thing is that while we gain something from this change, we’re not losing anything. You can get similar results by simply repositioning your light and adjusting power. Basically it forces you to use more realistic placements and powers, and as for falloff, it can be controlled with nodes, but since users seem kinda lost they’re thinking about bringing Falloff slider in sidepanel as well.
Back to the topic, in this scenario I think most of us see “radius” as other engines do, even some game engines in realtime, as a direct relation with softness of the shadow. For me, as rendering artist, this parameter is paramount since I use point light as a fill light for larger volumes keeping shadow consistency with other sources of light. Also is not always possible keep your light further enough of your mesh. The result of this new idea of make point light radius as sphere with harsh radius is limit your workflow without improving from what we have previously. I don’t see why having this behavior makes the light physically accurate.
I think I started with Blender 2013 (there’s post from this account since 2014), since then I’ve worked professionally in 3D 8 years and used a wide variety of softwares. And yes, Blender is innovative sometimes, likewise I said speaking about realness, this not might be a excuse for don’t get some constructive criticism. For example, something similar I can relate about Light/shadow linking. For me it’s the feature why I’m moving to 4.0, but sincerely, the UI is one of the worst I’ve seen. Coming from Karma in Solaris I don’t understand why your first iteration of this feature present on so many softwares need to be so bad implemented.
A reasonably formed picture will carry a wider breadth of picture information, with a larger continuum of purities. These can easily be expanded as required by a stylized or NPR shader. The same applies for someone seeking to increase the differentials and “push chroma”; it’s vastly easier to have a wider range of representations at the origin, and selectively crunch them out, than to pretend it is possible to go the other way.
What an NPR / stylized shader cannot do, is form a picture reasonably alone.
Think of a classic sphere with some sort of “picture” shading. If we “illuminate” the sphere with a pure BT.709 “Blue” lamp, all we have are varying intensities of the “Blue” primary. Cognitively, however, we expect some degree of form information which, in classical film and painted pictures, would manifest as a transition along purity.
An NPR / stylized shader, or a colourist working a classic picture, then has more information to draw from via an interstitial picture than they would were they to skip a discretized picture formation stage such as AgX. Want a series of stepped cel like shading? No problem. Want to posterize to a solid mixture? No problem.
TL;DR: There is a pernicious myth that reasonable picture formation is not required for NPR / Stylized work. It exists as nothing more than an ill-informed myth. Just as a colourist can push a wider volume of mixtures from a reasonably formed picture as an entry point, so too can NPR / Stylized shader authorship. It’s wiser to start with denser information in the picture signal, than sparser. And the picture information is in no way contained in the rendered colourimetric data.
It does the same for me, I don’t think it’s a bug, it’s the new behavior, but the point ligth only affect the mesh like that when the radius is intersecting with the mesh.
