WEB GAMING IS BROKEN
- Even today, web games that look like 2001. Low-poly models, janky animations, zero art direction
- Ads that last 30 seconds and force you to the app store. Interstitials and video interruptions that respect no one
- Freemium dark patterns. Pay walls, loot boxes, energy timers designed to exploit
- Unity web that underperforms. Unreal is over kill for some experiences
- No world builder tooling. Capable engines exist — the artist tools don't
- A generation of players who gave up on browser games and never came back
- There are so many games that don't need be told in Unreal Engine. I love AAA games taking advantage of RTX Pathtracing but not all stories need ot be told in that medium. Some stores are made to be toon shaded, or have a different art direction. Web can actually service this space without feeling visually compromised.
Web games should look as good as the stories they tell. Right now, almost none of them do."
NOT EVERY GAME
NEEDS UNREAL ENGINE
The problem with web games isn't that they can't run ray tracing. The problem is they don't try to look like anything at all.
Okami looked like a Japanese ink painting. Cuphead looked like a 1930s cartoon. Gris looked like a watercolor coming to life. Jet Set Radio invented cel shading because it had a vision worth fighting for. None of them needed a $3,000 GPU. All of them are unforgettable.
Okami looked like a Japanese ink painting. Cuphead looked like a 1930s cartoon. Gris looked like a watercolor coming to life. Jet Set Radio invented cel shading because it had a vision worth fighting for. None of them needed a $3,000 GPU. All of them are unforgettable.
Web games don't need to look like Unreal. They need to look like something. Toon shading. Anime. Concept art come to life. Ghibli warmth. Brutalist geometry. The technology exists in Babylon.js and PlayCanvas via WebGPU today. What's missing is the pipeline to build it all. That is what Weave provides."
IMAGE IN.
GAME ASSET OUT.
A fully automated pipeline that converts a single reference image into a compressed, game-ready 3D asset. What took a 3D artist days now takes minutes. Already running on an RTX 4090.
Comfy UI - Image to 3D Modeling pipeline
Reference image → Trellis2 AI → Blender auto-process → Draco compression → Web-ready GLB
|
Step 01
AI Generation Trellis2 infers a complete 3D mesh from a single image. Back-face geometry inferred automatically — no manual clean-up needed. |
Step 02
Auto Post-Process Automated Blender pipeline handles decimation, UV optimization, and texture baking. Zero manual intervention. |
Step 03
Compression Draco mesh compression delivers fast web load times without sacrificing visual quality. |
Step 04
World Builder Assets load directly into the Babylon.js world builder editor — placed, lit, and animated in the browser. |
As you know I was a 3D modeler at Disney Interactive a life time ago. I can still poly model but lets see what this AI image to 3D can do for a DIY pipeline. I was using Comfy UI for several video generation models last year in 2025 and there are so many tuned up models out there, all it was doing was making inferior videos while stressing out my 4090. This year we are going to see how well these low poly models can do and if I am really replaceable. Warning: It can do somethings decently but there are plenty it can't do.
Problems with Models - Hunyuan has problems. Trellis does some really strange stuff but it can be used at the cost of high poly to result ratio.
|
|
|
|
|
Trellis seems to handle text poorly but look at the image import and how it generates the back. It made some problems describing the barrel behind it but it is promising. These are low poly direct to GLB file. I think I can get better results from the high poly to Draco compression better. I will push this out.
Problems with Geometry
|
|
Trellis has a problem if you try to crank the geometry down but have complex details. Its AI decision isn't really for optimization. A good games model can look good with low polycount by having flat box like shapes but then added higher poly count in areas of rounding to achieve better rounding and silhouette while still having a relatively low count. The worst offender of understanding the geometry was Hunyuan so I abandoned it for Trellis 2. I'm sure it will only get better until it replaces my work flow and I add a new model.
Dos, Don'ts and Fixes
Solution with Objects: ChatGPT prompting for the right images for 3D Generation
|
|
Problems with Tunnels
|
|
it doesn't seem to understand shapes were i am generating an image from the inside like a room but it can make props well. There are problems with details and complexity but it seem to want to add detail on the back perspective that's true to the design and I really like that.
Problems with People
|
|
Look how Trellis can solve ambiguity. ChatGPT generated the source which is a woman. This was an early test and you really need to ask for a relaxed or T pose. Look at the face. Its really looks like a cartoon or game stylization. its not just low poly but the AI is solving the limitation by morphing the shape of the face and possible doing it to satisfy its model.
Problems with really large buildings even with a huge polycount
|
|
Picking a Game Engine
Engines |
3js |
Babylon js |
Playcanvas |
Description |
Okay its not really an engine, its a framework but it has a huge support Library
|
A true game engine,
|
Another true web game engine
|
Pros |
|
|
|
Cons |
|
|
|
WebGPU |
Yes
|
Yes
|
Yes
|
Cost |
Free
|
Open Source Free
|
Free engine, Pay tiers for editor. Doesn't matter to me since I am making my own world builder
|
WINNER? Babylon or Playcanvas
THREE DOGS.
ONE DYING CITY.
A STORY WORTH TELLING.
As a father of 3 dogs, I wanted to tell the story of my children wandering a beautiful, deadly, post apocalyptic ruins of of a decaying city. This district is the slums where a cyberpunk metropolis is built over an older city still inhabited by the impoverished.
CONCEPT ART
I could paint each of these by hand but time is short. References, assets and mood boards can be done in AI as a one man artist. The art is the end product. I may spend time to animate, model and rig a few key main characters but as a one mane show, AI on assets and back drops is a must.
Rigging
Already running.
Not a concept.
WIP Things I vibe coded with Codex
|
Coming Next:
|
What you get for backing Weave
Every tier includes the finished game. Your backing funds 6 months of development, a playable demo, and open-source pipeline documentation for the developer community.
$5 |
Believer
Your name in the game credits. You believed before anyone else did. |
$15 |
Early Access
Credits + early access when the game launches + monthly dev update dispatches showing real progress. |
$35 |
Most popular
Builder Tier Everything above + exclusive in-game cosmetic item only backers receive + access to the pipeline documentation when published. |
$75 |
Studio Backer
Everything above + your name on a dedicated Backers wall inside the game world + exclusive Discord access with direct development updates |
$250 |
Founding Developer
Everything above + one-hour call with Mark to discuss your own web 3D project + prominent credit placement. |
$1,000
|
Immortalized
Your dog — or you — becomes a named NPC in the game world. Permanent. Designed with your input. Only 3 slots available. |
Stretch Goals Subject to Change
- $25k Campaign funded — 6 months runway, game ships on the open web
- $40k Lasso gets a full social and charm NPC relationship system
- $60k Niko gets combat and rage mechanics — the city fights back
- $80k Second playable district added to the world
- $100k Full original soundtrack commissioned from an independent composer
- $125k Photo mode added — the world is worth looking at
- $150kTwo-player co-op mode prototyped — bring a friend into the city
- $175k Skip early access entirely — ship the full game at launch
- $200k Console port investigation begins — bring the dogs to your TV