Upload 2 files

README.md

@@ -12,8 +12,8 @@ pinned: false
 
 This demo is an AI-assisted development using **Claude Sonnet 5**.
 
-Restores low-resolution, noisy, low-light CCTV/security camera frames using the pretrained **Stable Diffusion x4 Upscaler** (`stabilityai/stable-diffusion-x4-upscaler`). Runs on CPU.
+Restores low-resolution, noisy, low-light CCTV/security camera frames using the pretrained **Stable Diffusion x4 Upscaler** (`stabilityai/stable-diffusion-x4-upscaler`). 
 
 Upload a low-quality frame, adjust the prompt/steps if you like, and get a 4x upscaled, denoised, sharpened result.
 
-**Note:** CPU inference takes roughly 1-3 minutes per image depending on step count. Input images are resized so their longest side is ≤128px before upscaling, matching the resolution the model was trained on.
+**Note:** Input images are resized so their longest side is ≤128px before upscaling, matching the resolution the model was trained on. For CPU inference, it may take 1-3 minutes per image.

app.py

@@ -57,7 +57,7 @@ demo = gr.Interface(
     title="CCTV Frame Restoration with Stable Diffusion x4 Upscaler",
     description=(
         "Diffusion-based super-resolution for low-quality surveillance frames. "
-        "Runs on CPU; each image takes roughly 1-3 minutes depending on step count."
+        "The model was trained on 128x128 crops, so larger inputs are downscaled first. "
     ),
 )