- More samples and a bit of clamping (but high values so you don’t loose a whole lot of dynamic range) will get rid of fireflies. It’s a result of specular and diffuse objects with bright light sources being in close proximity to one another.
- Different calculations? Thin glass material in Arnold? I don’t know.
- Specular in principled shaders is more like an approximation of fresnel, not how specular a surface is. That is determined by roughness.
- You’re doing it by using the volume absorption input, but can be calculated differently in different renderers. Also, having the model built to scale goes a long way.
Another tip is that for a scene like this, you will rarely need 128 bounces of anything. It’s an open stage so there is a good chance that your diffuse bounces will expire at 1 or 2 bounces. The higher bounces, like around 6 to 12 should be used for glossy and transmission for this model. I don’t use the square samples, and rather type in a hard value. 1024 gave a good result, of you want you can use a bit less and activate denoising, but this can be problematic for materials like this.