Primal VAE 2x1
Primal VAE 2x1 is a downloadable image autoencoder checkpoint that encodes images into a one-channel latent at 2× spatial reduction. Its latent is trained to have unit-normal statistics and can be patched to match the Flux.1 VAE latent shape.
What it does
Developers can use it to encode and reconstruct images or supply Flux.1-shaped latents to compatible diffusion workflows.
- Availability
- Weight file: primal_vae_2x1.safetensors
- Pricing model
- Not established
- Deployment
- Load for inference with Diffusers; example uses CUDA
- Privacy
- Not established
- Licence
- Not established
Capability
Modality
Access
Deployment
Pricing
Target user
Evidence
5 claims- The 2x1 checkpoint is available as primal_vae_2x1.safetensors. [1]
vae = AutoModel.from_pretrained("ostris/primal_vae", variant="2x1", trust_remote_code=True, torch_dtype=torch.bfloat16).to("cuda")
Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights. - The 2x1 variant encodes to 1 channel at 2× spatial reduction. [1]
- With patch compression 8, its latent has 16 channels at 8×, matching the Flux.1 VAE latent shape. [1]
patch 4 the latent, you end up with 8x at 16 channels, the same as most modern VAEs, without the need for deep
Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights. - The 2x1 model has approximately 80 million parameters. [1]
- Image sides need to be multiples of 16. [1]
Image sides need to be multiples of 16.
Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights.
Sources
1
1
Hugging FaceOwner-controlled
https://huggingface.co/ostris/primal_vae
Retrieved 2026-10-31 00:47 UTC
Verified
Versions
1| Version | Effective | Change | Changed fields | State |
|---|---|---|---|---|
| 1 | 2026-10-31 00:47 UTC | publication | None | Current |
Original catalog content is licensed under CC BY 4.0, attributed to wowsuchai. Source quotations and linked content keep their original rights.