I can replicate it in the sense that they happened to me in some frames of an animation but I couldn't pinpoint what angles point of view caused it. All that I can confirm is that they happened with AO+Bent normal + Subsurface. It was force to fix this using the Compositor to clamp the high values, erase the alpha and use the Inpaint filter.

Now that rBec0eeb918bac is implemented, objects with a modifier stack deform the textures correctly. All that's left is for objects with no modifier stack to work as well, those being deformed by shapekeys, which I think is the only non-modifier way of deforming meshes that's not being taken into account. Putting a disabled dummy modifier is my current workaround though

Not quite sure if I don't understand the controls or what's happening but I feel it's doing the exact opposite of what I want. It's sampling with max samples the already less noisy bright areas and using even less samples for the noisy dark ones. I'm trying to make the bright areas as noisy as the dark ones, and for the dark ones to remain the same as the base render without adaptative. Maybe it's the build I'm using, idk https://www.dropbox.com/s/s655x863joidbuu/adaptive-sampling.zip?dl=0

Thank you guys!

I agree. Even without an image I make annotations on the positions of UV islands, which needs canvas lock. So the idea of the warning is not ideal either, Annotations definitely have use without an Image open.

What seems to be truly happening is this;

The cross is just caused by a highlight with the Glare node. Either clamp that area of the render or adjust your glare nodes or use other nodes.

Every video editing software allows you to drag and drop entire image sequences at once and form a video strip

I too rely on viewport visibility for my rigs. Allowing me to use drivers to control from within the Armature things like outfits, hiding, showing different object parts

Line segments and Curve segments don't support Subsurface scattering while Triangles does.

You have different Subdivision count in render vs viewport. Set the same subdivisions for both render and viewport and they will look the same.

It was that all along? You guys are lifesavers! ❤

Thank you for the file. After seeing your model now I can definitely find it on mine, but to a lesser extent. It seems those flipped normal are areas where the combined normals go above 90 degrees away from the camera. I imported the model to unreal and it definitely happens there too, but less glossy, more similar to the rest of the material. It would be nice for Blender to do it that way but I'm not sure the developers would consider that a bug instead of a known Eevee limitation.

I can't replicate it with my own models and normal maps, some of which are directX, but all of them look perfect. That's why we need an example blend file with textures attached, to discard factors like the gamma and the kind of normal texture itself, the UV map used as tangent in the Normal map node for model with multiple UV maps, to help you if it's not a bug but just the setup, and ultimately to know if it's just a bug on that happens with one particular gpu driver or a more general bug.

Please provide a blend file with an example of the bug. The nodes are collapsed in that screenshot so it's hard to guess the settings of the math node and the normal map node

The default shadow settings are more on the side of speed than precision. Use ESM shadows with more exponent or use VSM shadows with high bitdepth and 0 bias, make sure walls are thick (double sided), and adjust the Clip Start of the shadows. You should not get any light leaks with this.

I love this! A couple of thoughts:

The noise is smaller than the size of a pixel in the cubemap, and thus the small noise disappears in the downscale process while the few ones that appear look big because that's how big the pixels are. It's like downsampling a picture of stars in photoshop, that's what happens, and no filtering algorithm would solve that 100%. It's definitely NOT procedural textures changing scale, because then it would change scale even with big textures like in the imaged I used above

What you see is a combination of a moire effect, due to small scale of the texture and the low cubemap resolution. Another way of putting it, the texture is so small than when it's captured on the cubemap filtered at a lower res, it just happens to look like a large scale version of the same texture, an optical illusion. But if you use any other procedural texture, or the texture at a larger size, you see that lowres cubemaps don't change their scale. Like in this image with the lowest cubemap resolution possible.

I know there are now some simple expressions that run without python and are considered safe so they don't require permission. Maybe try writting a more complex expression to eventually trigger the warning

It seems Separate Albedo gets forced on during renders. If you turn on Separate Albedo, the viewport and render look the same, but I think the fact that the render doesn't respect Separate Albedo being off is a bug

Count me as one of said users, but there's no hurry ☺ I even made a thread in DevTalk describing a really neat use case that my friend uses it for. Now that Local Cameras are back, this would be the other last feature needed

Thank you! I love it 😁

Yeah, this is due to the textures being 3D and the coordinates used. This doesn't happen with UV coordinates which are 2D or with Generated coordinates which stick to the surface, it's only when using Position or Object coordinates and the likes, the newly elevated surfaces intersect with the texture at a different point in space. I reccomend you using UV or Generated in the meantime.

Yeah, this seems like a bug, although specific to your graphics. Let's hope Clement sees this

I can't replicate it. Transparent clip shadows work fine for me, although at first it appeared mostly opaque because the texture is too bright above the clip threshold. By dissapear you mean the shadow becomes fully opaque or fully transparent?

Ideally Matcaps would be tagged with different color managements for each

Dupliframes is a real loss here, the array modifier and curves modifier don't cut it

Yeah, boolean doesn't work well with edges and faces that have 0 distance and are parallel. It's a known limitation

There's a big difference between times reported and actual times. It renders slower than it says. For example, after the first frame, not counting build time since im rendering a cube, the time reported goes down, but it it's actually the same. Regardless, these are the times on my machine when rendering, but I think they are slower

I see no bug. It works the same, except for the fact that it creates Ngons instead of Triangulation. Hide the cilinder or apply the modifier and you'll see that it works fine. Maybe you got confused because you are in Wireframe without Xray mode and you couldn't see the interior, press Alt+Z to toggle Xray. Also, that's a very old version of blender to compare it to. 2.79's boolean works much in the same way as in 2.8, with Ngons instead of triangulation

I can confirm too. win 10 64, nvidia 1060 6gb

Alright, so after some trial and error, I found a workaround, I still ask for this to be considered a bug though.

the problem that op points is that is not the default. not that it can't be done

That also is an option that was also there in 2.79 but now is a default. It's called Release Confirms and it's in Preferences>Input>Devices>Mouse

Not a bug. Those are the limitations of real time z depth shadows, since they are not raytraced. Very common in videogames.

Not a lie nor lost feature. Blender 2.8 still has Right click select, which you can change in the preferences of 2.8, and you don't have to deal with that "bug", you can select anything inside the circle of the gizmo with no problem. And Left click select existed in 2.79 as well, which you can change, again, in the user preferences of 2.79 and youll find a similar issue.

Another update. Turns out that plugging a texture into the Subsurface input of Principled BSDF throws another wrench into the mix. Not subsurface radius, which i know is not supported, but Subsurface. Meanwhile, using Subsurface just with a value doesn't break either the diffuse or the subsurface, at least during viewport.... sometimes... and using Separate Albedo in the render properties also can make things worse... sometimes...Let me explain:

I connected nothing in the second socket as a "simplest blend file" example. But in practice I was using it with emmissive shaders, and there it's a deal breaker

Now that the precision of the grid has been increased, can we have the Clip Start distance in all the Workspaces be 0.01? https://developer.blender.org/rBb5bc2158a07140e5463e209839bcf36d2ce2cddc

I believe this used to work fine a couple of 2.8 versions back. idk what happened, or why it works on lookdev but not on solid

Screen space effects like Screen space reflections and refractions, ambient occlusion and contact shadows, only work with what's available on screen and fade on the edges. They're not real raytraced effects. You can increase the virtual border during render by going to Render properties > Film > Overscan. Only works during render though

I like Alt 3. My main problem Current and Alt 1, and to an extent with all of them, is that the thin parts don't look realistic. Maybe an actual simulation with flip fluids would be nice

The shader does work, it has an Eevee fallback, maybe not the intended fallback but I do use it with hair quite often

can you provide a blend file with the hdr packed in it?

I hope they fix this. The Bump node basically causes bumps at half the resolution of the render. My understanding is that it uses the difference of value between adjacent pixels and that's why it's half res. It was also like that in 2.7's material viewport, but I think Eevee needs it if it's going to be used in final production. Maybe internally sampling the bump texture at double the render resolution, or with a jittered approach for each sample. I had to render at 4k so it's less noticeable 😂, luckily Eevee is fast. Btw, actual normal textures with a Normal map node, like tangent space normals, look fine and fullres.

Can I provide a blend file on their behalf?

In T59275#580219, @ronan ducluzeau (zeauro) wrote:

confimred on 21523b5 on ubuntu 16.04

Your .blend file contains a shapekey. Problem also exists with modifiers.

Proper handling of auto-texture space is probably a complicated todo.

Your gif doesn't convey well. Please attach a blend file with the simplest example. Is it possible you are referring to this?

It reminds me to some months back when matcaps, the HSV color picker, and the lookdev hdrs were all filmic. Now it's better for the matcaps and lookdev, but the color picker is hard wired to corrected filmic, regardless of the color management settings. There should be more options for the color picker, depending if you want the HSL and HEX to be corrected for srgb, filmic, your scene color management settings, etc