Hugging Face's logo

AiArtLab
/
sdxs-1b

Text-to-Image
Diffusers
Safetensors
Model card Files
xet
 Community
2
recoilme commited on
Commit
2b06e68
·
1 Parent(s): e2b305d

2604

Browse files
Files changed (6) hide show
  1. README.md +4 -1
  2. media/gandr-collage.jpg +3 -0
  3. media/result_grid.jpg +2 -2
  4. src/cuda.ipynb +2 -2
  5. test.ipynb +2 -2
  6. unet/diffusion_pytorch_model.safetensors +2 -2
README.md CHANGED
@@ -81,9 +81,12 @@ image.show()
81
  ```
82
 
83
  ### Diversiry
84
- Prompt: *cat* (Neg: low quality, bad quality, photo, low details)
85
  ![cat](media/cat.jpg)
86
 
 
 
 
87
  ### VAE
88
 
89
  The VAE in Simple Diffusion utilizes an asymmetric VAE architecture featuring an 8x encoder and a 16x decoder. While a compression factor of 8 is maintained during training, the resolution is effectively doubled during inference through an additional upscaling block. This strategy reduces training costs by an order of magnitude and boosts inference speed without perceptual quality loss. Effectively, this acts as an integrated latent upscaler. To ensure a fair comparison with other VAEs, we downsampled the generated images to match the input resolution for metric evaluation. The SDXS VAE was not trained from scratch but was initialized from weights of FLUX 2 VAE, then redisigned and retrained. We also trained [16 ch vae](https://huggingface.co/AiArtLab/simplevae) with flux.1 quality based on aura vae.
 
81
  ```
82
 
83
  ### Diversiry
84
+ Prompt: *cat*
85
  ![cat](media/cat.jpg)
86
 
87
+ Random prompts
88
+ ![collage](media/gandr-collage.jpg)
89
+
90
  ### VAE
91
 
92
  The VAE in Simple Diffusion utilizes an asymmetric VAE architecture featuring an 8x encoder and a 16x decoder. While a compression factor of 8 is maintained during training, the resolution is effectively doubled during inference through an additional upscaling block. This strategy reduces training costs by an order of magnitude and boosts inference speed without perceptual quality loss. Effectively, this acts as an integrated latent upscaler. To ensure a fair comparison with other VAEs, we downsampled the generated images to match the input resolution for metric evaluation. The SDXS VAE was not trained from scratch but was initialized from weights of FLUX 2 VAE, then redisigned and retrained. We also trained [16 ch vae](https://huggingface.co/AiArtLab/simplevae) with flux.1 quality based on aura vae.
media/gandr-collage.jpg ADDED

Git LFS Details

  • SHA256: 03ec331650d0841a0d5e6d81d5ea32f86031a542e040a4d7dad689137da5d862
  • Pointer size: 132 Bytes
  • Size of remote file: 1.15 MB
media/result_grid.jpg CHANGED

Git LFS Details

  • SHA256: 6a11c623d872189c8abcfaaa8969cd251e07f60edb96386c22cad971d2ad6edc
  • Pointer size: 132 Bytes
  • Size of remote file: 6.42 MB

Git LFS Details

  • SHA256: 28b01ae7faad9584a7f1efdb82d01e42c27e24b52c33489d65fe3f6156da0f99
  • Pointer size: 132 Bytes
  • Size of remote file: 6.46 MB
src/cuda.ipynb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:690a043172ca4a958ba7fa2547463f56f02869dbb816d673e53ad9399adf1c43
3
- size 1496
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1c49083d3defe77873103e3c528b087b93bcf202e742f699fbbceb81f4eefe51
3
+ size 2530
test.ipynb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ffb8b63e9a963aee212f993c38154f7fef82b7c94325cb2266c39093d9aad108
3
- size 13792585
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e2ad52116dadf67a00cc6303514409f17387875c4925d2a4b83625b23729b71d
3
+ size 13732127
unet/diffusion_pytorch_model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f08ca574d811e6b38bedb43da3c01931086c425e21992b33a5e03b9ba27130c4
3
- size 6420443856
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b4dc30ad1a2c8a6b060849ace4aa33d94889d7436d92892d2f21515ef00fe7d5
3
+ size 3210307232