# I'm releasing a World Model Ive trained that hallucinates GTA. It works on RTX 3060, 4060 and anything above, linking huggingface weights — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/983625f95016b979

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-11-14 17:30 ingested 2026-11-15 02:35

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1xw83dc/im_releasing_a_world_model_ive_trained_that/)

## retained text



```
Play it Now in browser (nonlocal, streaming) :[https://lucidml.ai/dashboard/minidemo?world=gta5](https://lucidml.ai/dashboard/minidemo?world=gta5)

Full Weights and Inference code : [https://huggingface.co/lucidml/gta](https://huggingface.co/lucidml/gta)

The architecture is a transformer which uses learned summaries. Ive written about it in detail in tech reports  and some more

I know there are a lot of limitations still like consistency and quality improvements, and am hoping to train this more to fix all of those.
```



document fetched 2026-11-15 00:28 · canonical https://reddit.com/r/StableDiffusion/comments/1xw83dc/im_releasing_a_world_model_ive_trained_that
