Simple glass with whisky and caustics

[Martini]

Here is a 'simple' glass that I modelled (it's a Duralex Picardie model that you see frequently in cafés) and it started out in Cycles, but looked terrible so obviously I had to switch to LuxCore, and add some liquid, ice and caustics

I rendered with PhotonGI caustic cache because it adds a nice brightness and warmth to the interior parts of the glass and ice, and light tracing for the actual caustics. I had to re-render about 15% of the frames with a different seed value (I also had to modify one of the BlendLuxCore addon files to allow me to automate this) because sometimes the caustic brightness just goes crazy high, and causes intense flickering in an animation.

Render time was about 4 - 5 minutes per frame at 384 samples.

I plan to make a better (more fancy) glass for the whisky later, but right now I lack time and motivation

This is my first contribution to this topic, hope you like it! Feedback welcome.

Short animation: https://youtu.be/_eAn51ynjro

[Luximage]

It's looking good.

[Sharlybg]

It's looking good.

So good I shared it on youtube .

[Martini]

It's looking good.

So good I shared it on youtube .

Thank you!

[BruceXu]

Did you use denoiser?

[TAO]

Good job.

[Martini]

Did you use denoiser?

Yes I did, but unfortunately the caustics are calculated with the Metropolis sampler, and OIDN is trained for Sobol pattern noise, so it does not do a great job. Also, there is not yet a reliable way to extract the caustics and SDS cache paths as render passes to denoise separately.

I think I just need to render it with many more samples to try to get rid of it by brute force...

[Dade]

Also, there is not yet a reliable way to extract the caustics and SDS cache paths as render passes to denoise separately.

Latest LuxCore has a caustic AOV

[Martini]

Latest LuxCore has a caustic AOV

Thanks, I'll check it out!

[Martini]

Latest LuxCore has a caustic AOV

Thanks Dade, the caustic AOV works great! I can't seem to denoise it properly with OIDN in Blender, but at least it's there now. Is there a trick I can replicate to denoise them with OIDN? Or do I just always need more samples? Seems like 512 light path samples are barely enough for a still-slightly-noisy result, whereas I can denoise quite well even 64-128 eye path samples.

Now, all that's missing is an AOV for the PhotonGI Caustic Cache contributions. Unless I'm just stupid and can't find it. I enabled all the AOVs I could find in Blender and none seem to have the information there, so when added all back together, the result is still missing the SDS photons from the beauty output, and lacks some 'warmth' in comparison.