Double generation time due to low demand

app.py

@@ -86,7 +86,7 @@ def detect(image):
     # ddim inversion
     img = transform_img(image).unsqueeze(0).to(pipe.unet.dtype).to(pipe.device)
     image_latents = pipe.vae.encode(img).latent_dist.mode() * 0.13025
-    inverted_latents = pipe(prompt="", latents=image_latents, guidance_scale=1, num_inference_steps=25, output_type="latent")
+    inverted_latents = pipe(prompt="", latents=image_latents, guidance_scale=1, num_inference_steps=50, output_type="latent")
     inverted_latents = inverted_latents.images
 
     # calculate p-value instead of detection threshold. more rigorous, plus we can do a non-boolean output
@@ -109,7 +109,7 @@ def detect(image):
         return max(0.0, 1-1/math.log(5/p_value,10))
 
 def generate(prompt):
-    return pipe(prompt=prompt, negative_prompt="monochrome", num_inference_steps=25, latents=get_noise()).images[0]
+    return pipe(prompt=prompt, negative_prompt="monochrome", num_inference_steps=50, latents=get_noise()).images[0]
 
 # optimize for speed
 pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)