| # Attribution |
|
|
| Licensing (code and weights) is stated once, in the License section of the model |
| card (README.md). |
|
|
| - **DINOv3 pretrained encoder:** Meta, |
| [official repository](https://github.com/facebookresearch/dinov3), |
| [paper](https://arxiv.org/abs/2508.10104), |
| [timm model](https://huggingface.co/timm/vit_base_patch16_dinov3.lvd_1689m). |
| The transformer blocks are used frozen; the input patch embedding is adapted |
| for this autoencoder. The weights are therefore distributed under the included |
| [DINOv3 License](LICENSE-DINOV3.md); see the model card's License section. |
| - **Encoder implementation:** provided by the separately installed |
| [timm](https://github.com/huggingface/pytorch-image-models) dependency, pinned |
| to `timm==1.0.26`. |
| - **semantic_vae** ([well9472/semantic_vae](https://huggingface.co/well9472/semantic_vae)): |
| the design follows its key ideas. These are a pretrained DINO encoder adapted |
| at its edges and a latent split into a reconstruction part and a DINO-derived |
| part. Its encoder has two DINOv2-B branches: one with a trainable patch |
| embedding, whose last six blocks are each LayerNormed and projected to 32 |
| channels, and one fully frozen, whose final block is projected to 32 channels. |
| We also follow it in training the decoder mainly with a DINO feature loss and |
| in using VISReg as the latent regularizer.¹ No code or weights from it are included. |
| - **Decoder up-path:** the VQGAN f16 decoder layout of Esser, Rombach and Ommer, |
| [Taming Transformers for High-Resolution Image Synthesis](https://arxiv.org/abs/2012.09841), |
| slightly modernized (per-position RMSNorm, folded upsampling) and behind a ViT |
| trunk. |
| - **Earlier decoder head (not in the release):** a pixel-level head after |
| [PixelDiT](https://arxiv.org/abs/2511.20645) |
| ([NVlabs/PixelDiT](https://github.com/NVlabs/PixelDiT)), with the FCDM block of |
| Kwon et al., |
| [Reviving ConvNeXt for Efficient Convolutional Diffusion Models](https://arxiv.org/abs/2603.09408), |
| as its token mixer. The released encoder and decoder trunk were trained with it |
| before the convolutional head replaced it. |
| - **Latent regularizer:** [VISReg](https://arxiv.org/abs/2606.02572) |
| ([project page](https://haiyuwu.github.io/visreg/)), used during training only. |
| - **Optimizer (training only, not a runtime dependency):** |
| [dionw](https://github.com/JTriggerFish/dionw), a single-GPU implementation of |
| [Dion](https://arxiv.org/abs/2504.05295) / |
| [Dion3](https://arxiv.org/abs/2608.11612) with |
| [NorMuon](https://arxiv.org/abs/2510.05491) normalization, using kernels from |
| [microsoft/dion](https://github.com/microsoft/dion) and |
| [Dao-AILab/gram-newton-schulz](https://github.com/Dao-AILab/gram-newton-schulz). |
| - **Training data:** about 14 million images from a mix of public and licensed |
| or curated datasets: mostly photographs, plus book covers, a few text-heavy |
| datasets and 1% synthetic rendered text. No training images are redistributed. |
| |
| ¹ semantic_vae's public model card names SIGReg (from LeJEPA) at 1e-3 as its |
| regularizer. We verified in its released training code that the regularizer is |
| VISReg, at 1e-3. |
| |
