File size: 1,242 Bytes
b4e9a22
 
 
 
 
 
c23b9b6
 
 
 
b4e9a22
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
# 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.