itch.io

Thanks for the heads up, I've updated the code and releases with additional variables to control the winding order of the faces as well as the culling mode.

I appreciate the feedback!
I implemented some additional variables to control the winding order of the faces as well as the culling mode. I think these images should illustrate the fixes:

Hosting via WebXR is a little tricky if one domain hosts the game environment (say itch.io) and another is meant to be providing the rendering. I might just need to play more with CORS and other settings next time (where available).

Regards speed I was able to get a hold of a 3090 (24 GB of VRAM) instead of a 3060 (12 GB VRAM) for the remainder of the game jam which makes a massive difference (safetensors model is 16 GB, anything less and processing spills over to system RAM and becomes CPU bound).

Long term I intend to roll the work over into another Open Source project so will keep expanding and improving as better models become available (generating 3D objects is still pretty new).

This was nice, I enjoyed the use of spatial to scan the room environment and how it was used with the growth spell and the way the fireballs collided with objects.

This was fun. Simple but carried out well and felt completed. Nice job!

Well done, really well-rounded and polished experience. The visual designs were clear and the mechanic was straight-forward to work out and consistent to use once understood. The names on the books were just right to figure out which was needed without giving anything away. Excellent audio effects too.

Gorgeous environment, I really enjoyed just wandering around. The "puzzles" were well designed, especially if one didn't read the intro text and was just experimenting blind. Nice job!

Fun mechanic, really well implemented! I could see spending more time playing this. Things like navigating through non-destructible canyons, fighting other dragons, maybe picking up a "rider" (or saving a princess) by flying over them with mouth closed and not being able to attack while transporting them, and so on could add complexity. But meanwhile it looks great, plays well, is genuinely fun, and is really impressive.

Thanks to a rather handy WebXR tutorial, I've added WebXR support alongside the temporary render host. I'll keep things running for the rest of the weekend for anyone who wants to try out Genchanted without any location installation required:

https://jigsaw.studio/genchanted/

Ha ha, (w) oof.

While the training set for the model has not been released, my understanding is they started with the Objaverse dataset, converted GLBs to vector and face object strings (the same thing the model generates as output and we convert back to display), and then filtered out any objects with more than 500 faces. Whatever was left they used for training their model. Therefore any complex objects wouldn't make it in, only simple ones.

I'm still experimenting to discover what remains, because so long as you stick within the training data you'll usually get decent results, anything outside looks like it... kinda melted in the oven.

As mentioned above, I've made some minor enhancements and posted details in the Devlog. I will continue posting update details through the rest of the Game Jam and in the future (so there's just one location going forward).

Meanwhile Ihave a better temporary server set up which can load the full model into VRAM, so execution speed should be more reasonable.

I've made some minor enhancements and posted details in the Devlog. I will continue posting update details through the rest of the Game Jam and in the future.

The hope is that someone who has PCVR or a Apple Silicon system with enough VRAM (or at least RAM) might be able to host the model entirely themselves, and not have to pay for a service. (Apple Silicon shares CPU RAM with the GPU so while it won't run nearly as fast as NVIDIA hardware it is better than relying on only the CPU.)

The current `llama-mesh` model isn't ideal in size (~16 GB) and may lack enough  variety of training data for the long term, but it does work now, and it  should be possible to swap out for other  models as they come along without too much extra work. Specifically the Ollama quantizations don't produce as reliable results as the original safetensors now, but it would be a single Ollama command to swap out for a new model later, without change to Genchanted's code since it already supports the API.

Thanks for checking out the project and the helpful feedback!

I just set up a temporary host running the original `llama-mesh` model under `generate-glb` (a separate Open Source project I wrote last week to help support this Game Jam). There's another post with details (and some caveats).

Regards the UI, yes I need (and intend) to make some tweaks to better show connection and activity, especially as it may take a couple minutes for the preview of the vectors and faces of the model to start appearing, depending on whether or not hardware acceleration is available on the server doing the rendering. I honestly didn't have much time to devote to the UI since most of the week was dedicated to getting generation dialed in (long story short the quantized models used with Ollama don't produce nearly as good results as the original model, but Ollama is the easiest to set up). Meanwhile the easiest way to check if Ollama is doing its thing is if your GPU usage (or at least CPU) starts to spike after clicking "Generate" - not ideal I know!

I will post a follow-up after cleaning up the UI, either late tonight or tomorrow most likely.

Thanks for checking out the project. I've created a new set of releases (which include "gamejam" in the filename. These point to an internet host, which through a series of SSH tunnels and Docker containers arrives at temporary server running `generate-glb` (a separate Open Source project I wrote last week to help support this Game Jam). This host is running the origin non-quantized `llama-mesh` model, which I've found has better results than Ollama.

There are a few caveats however:

• The original `llama-mesh` model is ~16 GB and the GPU on my temporary server only has 12 GB, so some of the model is offloaded to the system RAM. This makes everything run extremely slow.
• The UI of Genchanted needs some improvements to indicate when a connection has been established, progress during generation, and so on. I'm planning to add that before the voting ends, but "real life" work has caught up this week after devoting time to the Game Jam last week.
• It may literally take 2-3 minutes before you start to see the vectors and faces preview in the UI. The original effect was the semi-transparent cube would appear in the scene and slide into place as it was generated. But since generation is so slow on this host you might have to look above or behind you to spot it when it first appears