iaaa

app.py

@@ -9,14 +9,19 @@ import os
 import time
 import uuid
 
-LOW_MEMORY = os.getenv("LOW_MEMORY", "0") == "1"
 
 device = "cuda" if torch.cuda.is_available() else "cpu"
 dtype = torch.float16
 
-multi_decoder = (
+MULTI_DECODER = (
     torch.cuda.get_device_properties(0).total_memory < 18 * 1024 * 1024 * 1024
 )
+LOW_MEMORY = os.getenv("LOW_MEMORY", not MULTI_DECODER) == "1"
+
+print(f"device: {device}")
+print(f"dtype: {dtype}")
+print(f"multi decoder: {MULTI_DECODER}")
+print(f"low memory: {LOW_MEMORY}")
 
 vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=dtype)
 pipe = DiffusionPipeline.from_pretrained(
@@ -91,7 +96,7 @@ def predict(
         cosine_scale_2=1,
         cosine_scale_3=1,
         sigma=0.8,
-        multi_decoder=multi_decoder,
+        multi_decoder=MULTI_DECODER,
         show_image=False,
         lowvram=LOW_MEMORY,
     )
@@ -122,7 +127,7 @@ with gr.Blocks(css=css) as demo:
 [DemoFusion](https://ruoyidu.github.io/demofusion/demofusion.html) enables higher-resolution image generation.  
 You can upload an initial image and prompt to generate an enhanced version. 
 [Duplicate Space](https://huggingface.co/spaces/radames/Enhance-This-DemoFusion-SDXL?duplicate=true) to avoid the queue.  
-GPU Time Comparison: T4: - A10G: ~175s A100: RTX 4090: ~88.8s
+GPU Time Comparison: T4: - A10G: ~175s A100: RTX 4090: ~48.1s
 
 <small>
 <b>Notes</b> The author advises against the term "super resolution" because it's more like image-to-image generation than enhancement, but it's still a lot of fun!

requirements.txt

@@ -10,4 +10,4 @@ accelerate
 invisible-watermark
 huggingface-hub
 hf-transfer
-gradio_imageslider==0.0.14
+https://huggingface.co/datasets/radames/gradio-components/resolve/main/gradio_imageslider-0.0.13-py3-none-any.whl