Onwards to the weekend

Sparc3D vehicle outputs. I then decimated and cut the vehicles in half so they could be instanced with a mirror modifier. The next steps: better topology. This might be a case of cut and clean, such as making those windows a single polygon, right now they're part of the model's dense triangulated mesh. 

AI Sloppy? Maybe, but they are amazing starting points.

A quick terrain test. I'll never be able to stop this... Terrain addict, I am. I whipped this up in about half an hour, but then spent a couple of hours struggling to add some scatter. There were crashes!

 Next Steps

• Back to the environmental training course. Two lessons today.
• Tidy up assets and build some more:

        Lighting rig
        Cargo container 
        Oil Drums and crates

        Portable huts and a gatehouse.

A bad habit that I need to lose:

I iterate too slowly because I pump up settings too soon. The Card scene should have taken a couple of hours of work, but I must have spent more than that waiting for test renders, which I save but don't use much.

Industry disruption incoming

 Oh boy. This is significant

Combine generative technology like Midjourney with the mesh generation of Sparc 3d and what happens?

1. Sketch

ADS-77 Razorbill

2. Retexture in Midjourney

Input the above sketch with a simple description:

"Halo Pelican style VTOL dropship with pivot engine pods photorealistic on a white background"

I like this more, but the cockpit wasn't well-formed.

3.  Generate the mesh with Sparc3D

A few minutes later...

4. Import into Blender

So sketch to model in a few minutes. Granted, this is where the hard work will begin because those models are unreasonably dense (after a remesh, they're 5 million faces).

I need to watch some retopology tutorials to take one of these models to the next step: make it usable.

Most impressive!

 I have dabbled with photogrammetry. It's a powerful and relatively easy way to generate 3D assets for a scene. I always intended to go deeper into the field, maybe trying my hand at a full workflow instead of using my phone app, which admittedly works pretty well.

So there's a new method. Instead of scanning an object with a lidar rangefinder-equipped phone, you input a single image and allow a machine-learning algorithm to do the hard work. That doesn't sound like it's going to compete with a photogrammetry capture, but you know what, a bit bloody does.

See  https://huggingface.co/spaces/ilcve21/Sparc3D