If you're interested in the custom engine I'm building for my game project, I've made it available for free on Itch at page below. It's still early days, but the purpose of all this work is to build a full-featured modding toolset directly into my next game. interkarma.itch.io/project-arro...

The scene editor in my engine is now functional enough to author complete scene assets. I'll run through some of what I've built over the last 6 months. First up is visualising collider shapes (mesh, capsule, sphere, convex hull, etc).

After a year of working full-time independent on my game, I've been fortunate enough to pick up some great part-time work. This provides me more financial security and some resources to help self-fund development into the future.

Asset manager can now import a mix of files and folders while preserving source hierarchy on import. Asset builders are multithreaded and run even while files are being copied, so whole process completes in about the same amount of time as file copy, including thumbnails for textures and models.

Deep Engine drag-and-drop several hundred FBX model files into asset manager

Setting up a modal progress dialog so editor can display progress after user drags and drops a large number of asset files into engine.

Early stage of Deep Engine's asset rebuild and hot-load in action. After editing and saving image file, the build system detects change to source file and rebuilds it into a texture asset based on current meta properties. Note editor thumbnails on left update after image is saved.

I haven't posted a blog update in several weeks now, but I have been working hard across several fronts! This blog post catches up on what I've been doing and where I'm headed next. dev.interkarma.net/2025/05/02/m...

Making an editor front-end for asset compiler. Asset builders are now multithreaded, so an expanding log reader is required to solve the problem of linear output becoming interleaved from different threads executing in parallel. I can also highlight any build errors directly in the editor.

Making some big improvements to asset browsing. In addition to thumbnail view, it will now support lists and trees and more filtering options. Also built a new tree view control that's fully virtual and can scroll hundreds of thousands of items without slowing down.

Just a heads-up that I'm changing hosting providers for dfworkshop.net and interkarma.net. They will go down over the few days as I move everything over.

I really want to become more proficient in Blender. Following some courses and immersing myself in practise each day. Right now learning how to model a detailed human figure and I'm slowly becoming more confident with the tools. Best way to learn is to make mistakes and fix them.

Animation retargeting now working well enough to throw all kinds of animations at all kinds of models. This whole process was really involved. I had to set up a retargeting pipeline to handle joint-to-joint mapping, pose correction, and proportion correction.

Revisiting and improving animation retargeting in my engine. Looking forward to supporting more animation inputs.

Really happy with my procedural sky so far. Shadows lengthen and shorten as the sun disc arcs from dawn to dusk. Clouds are lit by sun's direction as they drift and evolve throughout the day. The skybox colours light environment based on time of day. It's a good base to build on!

Spent the last two days in cyclone prep - trimming trees near house, cleaning gutters, securing what I can. Now feeling completely physically drained. Put in about two hours of gamedev today and I'm spent. Just going to crack a beer and watch the storm roll in from this arvo.

I forgot to mention - new sky system casts shadows over terrain that move as the sun tracks across the sky throughout the day. I'll do a proper pass on this once distant terrain is further along. It's not much, it just adds a little extra something.

Working on sky colour gradients that are a bit more "grounded" by reality, then I'll finish animation of day night skies. These skies aren't going to win any technical achievement awards, but they can sometimes make pretty background vistas, and I'm OK with that.

Working on game time and day/night cycles. An animated sun disc follows directional light as it rotates through the sky. Also animating environment light intensity on a curve. Now to animate sky colour gradients and a few other refinements, and I'm done with skies for a bit.

Fixed some visual bugs and improved performance a whole lot. Static HDR image-based lighting on left, dynamic sky shader HDR lighting on right - computing whole skybox + irradiance maps every frame at only +3ms.

Now using my dynamic skies as a source for HDR image-based lighting. Comparison between a static HDRI and dynamic sky pickup on spheres with roughness and metallic gradient.

Creating utterly lurid skies to get them out of my system.

A basic procedural sky with clouds, gradients, sun disc, and a bit of lighting on clouds based on their volume and sun direction. Lots of work to do on this in future, but it's a good enough starting point.

Just a tiny bit of ray marching through noise volume gives the clouds a better sense of light direction and volume. It's still fast too at ~1ms per redraw.

As much as I want to go nuts working on awesome clouds, I can't get bogged down. A quick compute shader to render noise into a texture cube will do for now.

Using colours to represent a vector in shaders is a neat way to "see" directions, sort-of like synesthesia for spatial orientation.

In Project Arroa: Revenwood, Dougal’s Glade is where the player starts their adventure. It’s also a sandbox where I can experiment with production, art styles, and gameplay. This post explores the concept along with a few additional details. dev.interkarma.net/2025/02/21/d...

Bright moon dipping below gum trees behind the house tonight.

First video of functioning terrain painter tool. Paint two textures at once with brush radius, UV scale, and slope threshold. The slope threshold controls how the horizontal and vertical textures are blended. UV scale defines texture repeat.

A downside of this terrain texturing method is the control map texels look blocky when adjacent to different settings. A solution is to resolve neighbouring control blocks and blend between their colours manually in fragment shader. And to carefully manage weights for more natural-looking blends.

Some random screens of my terrain system after a couple days tightening up shaders and fixing lighting bugs. Can paint with two textures at once (background+overlay) from a palette of 32 textures via a control map including params for slope threshold and UV scale. Now onto front-end tooling.

Setup a nice little decaling system today. Only need to use it with terrain brushes for the moment, but could use it for much more later (surface details, spell effects, etc.)

My first devlog for 2025 is now up! Read about my work over the last month preparing Deep Engine to render trees and foliage. After more than a year working on engine and tools, I'm finally ready to make some progress on the actual game. dev.interkarma.net/2025/02/03/t...

I'm happy with new foliage system. It needs a performance pass, and it will receive that as I integrate with terrain. As of now, I have all the engine systems I need for map building, so I'm going to say YOLO and start building the first open area so I can get into something that looks like a game.

Shovelling 262000 grass and flower models at my foliage system with literally zero optimisations yields about 20fps. Not great, but also kinda not bad for a system that only came alive a few hours ago. We'll see where this is after further iteration.

Fixed my placement bug and reproduced the stress test layout to compare like-for-like draw indirect vs scene instancing. GPU time is roughly the same between both (same models and instanced is instanced, I guess). CPU is doing less work now. Time to iterate and look for improvements.

Indirect drawing now supports models with submeshes and materials, so the trees have their leaves back. Unfortunately have some some bugs with layout system I need to fix before I can scale up new foliage pipeline. Almost there now, another day or two should do it.

Transforms now pass correctly through compute shader and out the other side of indirect drawing process. Next on to GPU culling and LOD selection.

Indexing bug fixed. Now for that transform so our indirect drawing pioneer can stand up.

After days work engineering new systems and enhancing old ones, I push one single tree trunk, a pioneer if you will, through my new foliage pipeline. This is my first opportunity to see anything on screen come out the other side. Well, sometimes things don't work perfectly first go.

It's fascinating how many changes I've made to my personal coding style while building a game engine compared to my last project. Some of the early decisions influenced this like enforcing nullable, good handling of unsafe interop, reference counting, prolife often, and errors as warnings.

Buncha trees at more interactive framerates. I'll keep chipping away at the problems and improving things until this is where I need it to be.

Listening to Hottest 100, snags and beer on the side. It's a pretty chill day.

One of my hobbies is baking bread. I'm often critical of my own efforts, but it helps to compare progress over time. There are 12 years of incremental refinements between the first two loaves and the second two.

Now that my engine can render trees, I need to work on new instancing techniques to draw them more quickly. Shown here, I'm simply drawing a bunch of trees individually for a total of 14k drawcalls including shadow cascades. Nonsense approach, but it's my baseline to measure improvements against.

Different tree types and materials, shadows fixed for trees. The dappled leaf shadows will eventually look great on the forest floor.

Made some upgrades to nature shaders and materials. Trees now render more nicely and interact better with my lighting system. I still need to fix tree shadows, my shadowmap renderer doesn't understand leaves yet.

More work on trees. Made a rough instancing test with 10k of them and engine didn't explode. Not bad, but have some work to do before I have proper foliage at an acceptable frame rate.

LOD groups now working in render pipeline, shown up close here with a bias to see the change in detail between each group. Now I can move on to foliage shaders and more serious tree action.

Building out requirement to render mesh LODs (level of detail). I'm using the approach where mesh screen area percentage decides which LOD to use, rather than distance alone. Blue box is the mesh bounds in world space, yellow rectangle is the mesh area in projected screen space.