Kitchen ArchViz Update

In my spare time – which is in short supply currently – I have started creating some objects which are going to help fill out the kitchen scene I started a few weeks back:


They are only simple pots (without their wooden lids for now), and only feature very basic materials (a colour and roughness map), but they do seem to work quite well. These pots can be used in my other ArchViz scenes if required. I might even have a go at designing my own patterns and pot shapes.


Below now includes the lids. The material came from a reference photo I took with my phone camera so there is definitely room for improvement, but it’s a decent start:



I realised that Unreal Engine didn’t do a great job of adding Ambient Occlusion onto self shadowing meshes (such as the lids). This is hardly surprising since it is a real-time post-processing effect. In order to make the lids look believable, I baked out AO maps for the pot and the lid from 3DS Max and applied them to the AO slot of the corresponding materials. The result is looking quite nice for only an hours work:



I’ve decided not to use the pattern from the reference pot as I want as much work as possible to be originally created by myself. Plus I also fancied having a go at designing my own patterns for ceramics. I therefore decided to take the background images I created for my response to the first Adobe AGP Publishing assignment and re-purpose them for pattern application:



Whilst I was now happy with the geometry and the material of the pot (including tweaked AO), I was still bothered by how the pots sit in the scene. They don’t appear to be connected to the table as there is no contact shadowing. In order to achieve a believable effect I had to play around with the Global Lightmass settings. In Unreal Engine 4, these settings controlled things like the bounce count of indirect lighting, global AO, quality of lighting and shadow/lightmaps, etc.

I found that my current settings were allowing too many light bounces (100!) so everything was looking washed out, including any shadow detail. I therefore reduced the bounce count to 45 and the halved the amount of AO applied to the indirect lighting. There results were better as there is now more global shadowing but it’s not too dark:


Whilst the global AO is now almost perfect, there is still the issue of the shadowing. In order to achieve realistic results there are a number of variables you have to consider. Rather than list them all here I’ll provide this link which details how an ArchViz studio set up their scenes with lots of images of how the different settings affect the output.

For my scene however, I upped the shadow map size on the table surface (as that is obviously going to be receiving and displaying the shadow detail) and reduced the static lighting level scale (to 0.75) in the global lightmass settings. For the sake of creating a great looking render, I also cranked the static lighting quality up to the maximum of 10. ADJUSTING THESE SETTINGS WILL DRAMATICALLY INCREASE LIGHT BUILD TIMES. Sorry for the bold but they really do turn your real-time game engine into a vray rendering engine and you’re back to waiting for up to 5 minutes for the lighting to build. It’s worth it though:


I did one final render with the best settings/highest sized lightmaps within reason just to see how much more you can squeeze out of a game engine:


One thing that struck me while going over the article I posted above, was that the author insisted that the use of the “bounce card” technique inadequate for the results they desired. This puzzled me slightly as without this technique I have never been able to achieve enough brightness in my scenes from just the bounced light of the skylight and directional light. I understand the logic in that faking bounced light is too inflexible for architectural realisations as it is inefficient to have to set-up different lighting systems for each location and there would be no consistency if the end result was to be a seamless interaction between the outdoor and indoor lighting (especially in daylight).

I therefore decided to experiment with their proposed settings on the scene above so I had a baseline for comparison:


I was shocked at how different the scene looks and how much it has affected the materials, especially the roughness maps. Everything all of a sudden is much more reflective AND much more like the kind of results I was expecting when producing the roughness maps. I always thought I was making them way too shiny (in the greyscale image space which they work).

After some experiments with the lighting, I ended up with this:


I adjusted the levels of all the roughness maps in the scene and made them about 25% closer to white which feels a little more realistic and provides a much greater dynamic range. The whole scene is now not so bright which is a little more realistic, even if it doesn’t look as good as the bounce card version. For one final render (before moving on to something else), I upped the global brightness a little and decreased the indirect lighting level scale to produce my production render for this scene:


Although I still prefer the look of the approach that used bounce cards, for the sake of accuracy and flexibility, you can’t argue with the results or logic of the skylight/directional light only model. However, it should be noted that it took 25 minutes to build the lighting for the above scene.


I modelled a small watering can to help build up the collection of kitchen objects I am trying to establish. It’s going to have some branding and normal/roughness map detail added soon, but here is the result with a basic colour material and AO applied:


Before the close of the day I managed to create the material for the watering can that consists of colour, metallic, normal and blended roughness and grunge maps. The grunge map is a place-holder with really basic texture mapping in a third UV channel:


There needs to be quite a bit of optimisation because 3 UV channels is not acceptable on a small object. It’s difficult to add grunge without a tool like Substance Painter. Maybe it’s time I invested.


I became very interested in the differences between the bounce card technique and the sole dependence on skylight and a directional light to illuminate a scene so I posted a message in the Unreal Engine 4 – Archviz Facebook group to find out people’s preferences and why. It became instantly apparent that the bounce card approach is outdated offers no advantages. Most designers no rely on just a skylight and a directional light (as my previous research revealed) however, I need more convincing and guidance so I was very excited when a kind person posted a reply with a link to a tutorial he had written on such a setup. He also stated that version 4.11 of the engine was slightly optimised for speedier lightmass build times. With this new information, I set about starting my own scene from scratch.

After some playing around with the settings, I was eventually able to produce the following:



From now on, anything further I develop in relation to this scene will occupy a new post. Overall, I am very happy with this new setup. It’s easy, accurate and flexible. I look forward to developing more on this scene or new scenes using this new approach.