how would that be implemented in batch? Will you have a button that runs stage1+stage2 as well as th
how would that be implemented in batch? Will you have a button that runs stage1+stage2 as well as the original independent ones?
RAlso, by the way, was wondering, are there hotkeys available in auto1111? Or some way to bind parameter buttons for a keyboard shortcut?
Fa checkbox
Fso if you enable checkbox
Fit will do first stage 1
Fand stage 2
Fi am adding metadata right now
FCo how can I fix this?
R@Dr. Furkan Gözükara is there a way to read the Supir metadata parameters in Supir, say, by dropping an upscale into input, like in auto1111 png tab?
Fi am working
Freally hard for me
Fbut i am adding meta data displayer tab
Fworking on it
Roh wow. ok
Rthat's perfect
Fye about to finish testing
FGreat, thank you
TIs it possible to run SUPIR on kaggle?
Fupdating
Fnot yet. they have to fix fp16
Rsuuuuuuper cool! @Dr. Furkan Gözükara
Rwhat did you name this metadata category/type? its specific to your Supir script no?
VSUPIR seems to be amazing and the 1-click installer too! Great job! -Is SDXL model with Juggernaut XL - V9 only used to generate the initial image that is then upscaled, so I do not need SDXL if I use other images to upscale ?
Fit should work with all images that uses same kind of metadata
Fno it is used during upscale
Fdo you need other images metadata too? i may try to make it more robust
Fi am testing output folders parameter right now
Rso, I use an called Allusion which sees the metadata from Auto1111 and Forge
Rhere's an example
R
R
Fok i will try to make it more robust
Ryou get all the data from the generation
Ri think this metadata that auto1111 saves is called Parameters
RSo if you use the same, i think it will be universal
Ffixing errors
Ptrying to use Supor and at 2nd stage getting:
[Tiled VAE]: split to 4x2 = 8 tiles. Optimal tile size 480x448, original tile size 512x512
[Tiled VAE]: Executing Encoder Task Queue: 100%|████████████████████████████████████| 728/728 [00:01<00:00, 405.80it/s]
[Tiled VAE]: Done in 2.045s, max VRAM alloc 9757.875 MB
P:\SUPIR_v7\SUPIR\venv\lib\site-packages\torch\nn\functional.py:5476: UserWarning: 1Torch was not compiled with flash attention. (Triggered internally at ..\aten\src\ATen\native\transformers\cuda\sdp_utils.cpp:263.)
attn_output = scaled_dot_product_attention(q, k, v, attn_mask, dropout_p, is_causal)
[Tiled VAE]: split to 4x2 = 8 tiles. Optimal tile size 480x448, original tile size 512x512
[Tiled VAE]: Executing Encoder Task Queue: 100%|████████████████████████████████████| 728/728 [00:01<00:00, 405.80it/s]
[Tiled VAE]: Done in 2.045s, max VRAM alloc 9757.875 MB
P:\SUPIR_v7\SUPIR\venv\lib\site-packages\torch\nn\functional.py:5476: UserWarning: 1Torch was not compiled with flash attention. (Triggered internally at ..\aten\src\ATen\native\transformers\cuda\sdp_utils.cpp:263.)
attn_output = scaled_dot_product_attention(q, k, v, attn_mask, dropout_p, is_causal)
F
Fyes not an error
Pit was stuck for a very long time. but in the end it did upscale. I thought it was dead xD
F
Fyou must be using shared vram
Fcheck and let me know 
Rsame metadata type as auto1111?
Khopefully fp8 with supir soon
Fi dont know
Fbut works on both
Fye would be great
Ffp8 added for auto1111
Pyou mean Nvidia offload to ram? its turned off. i have 4090 using 1-1 options to reduce vram usage
Ri can test it if you send me the image