So far, there is a broad agreement for the following effects:
Clarity / Unsharpen filter
Sharpness
Vignette
Film Grain
Camera Shake
The next step is for artists from the Blender Studio to implement those. The node groups shared in this thread and others will be taken as reference. Depending on the implementation, we will decide whether existing nodes should be extended or if they should be their own assets.
Since these new assets won’t be ready before the 4.5 beta phase next week, the asset shelf as well as the new assets will be part of the 5.0 release.
Here are two interesting synthetic effects that I often use: 1. Simple synthesizer scene fog
2. Take pixelation as an example. Only perform the compositing effect on a specified object and its mask
There are a few more questions:
So far, does the fact that Cryptomatte masks are not supported for use in node groups to some extent prevent synthesizer node assets from having the ability to only take effect on the specified objects in the picture?
Will Cryptomatte nodes be allowed to be used within node groups in the future? Or is it that assets are not required to be in units of node groups, and multiple nodes can be directly loaded as assets?
2. From the two cases I provided, it can be seen that some compositing effects rely on the renderer to enable some channels that are not enabled by default, such as Cryptomatte. Can assets that rely on channels directly enable the corresponding channels when in use?
I’m not sure per object pixelation would be a good asset. In this case I think it’s better to have a pixelate node/asset and rely on cryptomatte as the “object selection” node.
This is one of the main reasons why Render Layers and Cryptomatte nodes are not allowed to be within node groups. We have plans to support that in the short term, and we are also considering supporting Bundles to simplify links. Meanwhile, assets should treat passes as separate inputs.
Yesterday I downloaded the asset and tried it but I have some doubts.
The first doubt, as has already been written, is the roundness feature that generates sharp and discontinuous shapes because a lens distortion is used to generate the vignette mask. The only use case that comes to mind is simulating obstacles in front of the lens.
The second doubt is generally linked to the too “artistic” approach used. This will require the user to either know how the asset works by going to see inside how it is made or simply move the parameters randomly to see what happens.
Artistic does not necessarily mean more inclusive and simple.
Could it be a good solution to try to create assets such as “camera effects” using a simplified but physics-based approach?
For example in the case of vignetting, this is generally created by 4 factors some of which have a precise falloff in light. These are: mechanical, optical, natural and digital.
Just like the Camera Shake node example, you could create a single parameter or a couple of global parameters and then add an advanced panel with subpanels that toggle each of these vignette layers with their respective (simplified) parameters.
This way beginners can use quick and understandable parameters and advanced users can use objective knowledge to layer their vignette without having to go in and see what’s going on.
I think for this concrete example, I would just expose Vignette as its own asset, because of how common the effect is and also for better discoverability. From a practical point of view, we don’t support menus (enums) in node groups yet, so a “camera effects” asset is not feasible currently.
A big advantage with assets is that advanced users can simply go inside the node group and change things as they please. Highly depends on the asset of course, but for example I wouldn’t add a translation input if you only need for say 10% of the cases. Instead, you could just add a translation node inside the node group and expose it as an input
I don’t mean creating a single node that does all the “camera effects”, but for the effects that are part of the “camera effects” category, such as vignetting, lens distortion, chromatic aberration, since there are several studies on the subject based on physics, why not use a more rigorous approach (with the necessary simplifications) and less artistic in the implementation?
I agree with exposing only the Vignette node
Sure, but as you say it depends from asset to asset but also from “advanced user to advanced user”. Physics should be simplified and functional for artistic purposes with the relative control parameters. The point is to try to use objective knowledge to avoid entering the node group and see what happens. An artistic approach is very arbitrary while a physical approach is more understandable by an expert user. For a beginner the only setting of factor would be enough in reality. The other settings (for example mechanical, optical, natural and digital vignetting in the case of a physically accurate implementation) should go in an “advanced” panel or similar.
Now that we have access to the image info we can finally generate a superellipse.
since there are several studies on the subject based on physics, why not use a more rigorous approach (with the necessary simplifications) and less artistic in the implementation?
Both seem useful for different situations. If you have a physically based model for vignetting, please do share! That would be an awesome option to have.
