| # Attribution | |
| The Future Sight world-model weights are released under the MIT license in | |
| [LICENSE](LICENSE). They are associated with *Training Controllable World Models | |
| via Action Influence Maximization*, Samuel M. Bateman et al., CoRL 2026. | |
| Our base VAE is | |
| [**`stabilityai/sd-vae-ft-mse`**](https://huggingface.co/stabilityai/sd-vae-ft-mse), | |
| used through pcuenq's Flax conversion and included as a frozen StableVAE in each | |
| inference bundle. Credit belongs to the original autoencoder | |
| authors, Stability AI's improved-autoencoder contributors, and the Flax | |
| conversion contributors. The upstream Hugging Face cards identify these weights | |
| as MIT: | |
| - [Stability AI sd-vae-ft-mse](https://huggingface.co/stabilityai/sd-vae-ft-mse). | |
| - [pcuenq sd-vae-ft-mse-flax](https://huggingface.co/pcuenq/sd-vae-ft-mse-flax). | |
| - [CompVis latent-diffusion autoencoders](https://github.com/CompVis/latent-diffusion#pretrained-autoencoding-models). | |
| The original upstream HF revision used for conversion was not recorded. The | |
| bundle's `vae.source_sha256` records the converted source identity, and its state | |
| manifest records the tensor hashes used for restoration. Current upstream | |
| repository revisions are not represented as the original training provenance. | |