Update app.py

app.py

@@ -88,7 +88,7 @@ upscaler = UpscaleWithModel.from_pretrained("Kim2091/ClearRealityV1").to(torch.d
 def scheduler_swap_callback(pipeline, step_index, timestep, callback_kwargs):
         # adjust the batch_size of prompt_embeds according to guidance_scale
     if step_index == int(pipeline.num_timesteps * 0.1):
-        print("-- swapping scheduler --")
+       # print("-- swapping scheduler --")
       #  pipeline.scheduler = euler_scheduler
         torch.set_float32_matmul_precision("high")
        # pipe.vae = vae_b
@@ -115,7 +115,7 @@ def scheduler_swap_callback(pipeline, step_index, timestep, callback_kwargs):
      #   pipe.unet = unet_a
         torch.backends.cudnn.deterministic = False
         #pipe.unet.set_default_attn_processor()
-        print("-- swapping scheduler --")
+        #print("-- swapping scheduler --")
       #  pipeline.scheduler = heun_scheduler
         #pipe.scheduler.set_timesteps(num_inference_steps*.70)
       #  print(f"-- setting step {pipeline.num_timesteps * 0.9} --")
@@ -199,7 +199,7 @@ FTP_PASS = "GoogleBez12!"
 def scheduler_swap_callback(pipeline, step_index, timestep, callback_kwargs):
         # adjust the batch_size of prompt_embeds according to guidance_scale
     if step_index == int(pipeline.num_timesteps * 0.1):
-        print("-- swapping scheduler --")
+     #   print("-- swapping scheduler --")
       #  pipeline.scheduler = euler_scheduler
         torch.set_float32_matmul_precision("high")
        # pipe.vae = vae_b
@@ -225,7 +225,7 @@ def scheduler_swap_callback(pipeline, step_index, timestep, callback_kwargs):
      #   pipe.vae = vae_a
      #   pipe.unet = unet_a
         torch.backends.cudnn.deterministic = False
-        print("-- swapping scheduler --")
+      #  print("-- swapping scheduler --")
       #  pipeline.scheduler = heun_scheduler
         #pipe.scheduler.set_timesteps(num_inference_steps*.70)
       #  print(f"-- setting step {pipeline.num_timesteps * 0.9} --")
@@ -275,6 +275,7 @@ def generate_30(
     num_inference_steps: int = 170,
     denoising_start: float = 0.0,
     denoising_end: float = 1.0,
+    strength: float = 0.3,
     use_resolution_binning: bool = True, 
     progress=gr.Progress(track_tqdm=True)  # Add progress as a keyword argument
 ):
@@ -285,6 +286,7 @@ def generate_30(
     #upload_to_ftp(filename) 
     #uploadNote(prompt,num_inference_steps,guidance_scale,timestamp)
     sd_image_path = f"rv_L_{timestamp}.png"
+         #  INITIAL GENERATE
     if denoising_start==0.0 and denoising_end!=1.0:
         generator = torch.Generator(device='cuda').manual_seed(seed)
         options = {
@@ -294,10 +296,9 @@ def generate_30(
             "width": width,
             "height": height,
             "guidance_scale": guidance_scale,
-            "num_inference_steps": int(num_inference_steps * (denoising_end-denoising_start)),
+            "num_inference_steps": num_inference_steps,
             "generator": generator,
             "output_type": "latent",
-          #  "denoising_start": denoising_start,                   #  for img2img
             "denoising_end": denoising_end,
             "callback_on_step_end": pyx.scheduler_swap_callback
         }
@@ -318,7 +319,7 @@ def generate_30(
             },
         sd_latent_path,
         )
-        
+         #  MIDDLE GENERATE
     if denoising_end!=1.0 and denoising_start!=0.0:
         latent_file = f'rv_L_{denoising_start}.pt'
         loaded_data = torch.load(latent_file)
@@ -333,30 +334,30 @@ def generate_30(
         generator = torch.Generator(device='cuda').manual_seed(seed)
         options = {
             "prompt": [prompt],
-            #"image": loaded_latents,                   #  for img2img
-           # "strength": 0.2,                           #  for img2img
+            "image": loaded_latents,                   #  for img2img
+            "strength": 0.3,                           #  for img2img
             "negative_prompt": [negative_prompt],
             "negative_prompt_2": [neg_prompt_2],
             "width": width,
             "height": height,
             "guidance_scale": guidance_scale,
-            "num_inference_steps": int(num_inference_steps * (denoising_end-denoising_start)),
+            "num_inference_steps": num_inference_steps,
             "generator": generator,
             "latents": loaded_latents,
             "output_type": "latent",
-           # "denoising_start": denoising_start,                   #  for img2img
+            "denoising_start": denoising_start,                   #  for img2img
             "denoising_end": denoising_end,
             "callback_on_step_end": pyx.scheduler_swap_callback
         }        
         if use_resolution_binning:
             options["use_resolution_binning"] = True
         batch_options = options.copy()
-       # pipe2 = StableDiffusionXLImg2ImgPipeline.from_pipe(pipe)
-       # rv_image = pipe2(**batch_options).images
+        pipe2 = StableDiffusionXLImg2ImgPipeline.from_pipe(pipe)
+        rv_image = pipe2(**batch_options).images
         #indx=int(num_inference_steps*denoising_start)
         #pipe.scheduler.set_begin_index(indx)
         #pipe.scheduler._step_index = indx
-        rv_image = pipe(**batch_options).images
+        #rv_image = pipe(**batch_options).images
         sd_latent_path = f"rv_L_{denoising_end}.pt"
          # Encode the latents before saving
         #pipe.vae.to(torch.float32)
@@ -369,7 +370,7 @@ def generate_30(
             },
         sd_latent_path,
         )        
-        
+         #  FINAL GENERATE
     if denoising_end==1.0 and denoising_start!=0.0:
         pyx.upload_to_ftp(filename) 
         latent_file = f'rv_L_{denoising_start}.pt'
@@ -385,30 +386,30 @@ def generate_30(
         print("-- using latent file --")
         options = {
             "prompt": [prompt],
-           # "image": loaded_latents,                   #  for img2img
-           # "strength": 0.2,                   #  for img2img
+            "image": loaded_latents,                   #  for img2img
+            "strength": 0.3,                   #  for img2img
             "negative_prompt": [negative_prompt],
             "negative_prompt_2": [neg_prompt_2],
             "width": width,
             "height": height,
             "guidance_scale": guidance_scale,
-            "num_inference_steps": int(num_inference_steps * (denoising_end-denoising_start)),
+            "num_inference_steps": num_inference_steps,
             "generator": generator,
             "latents": loaded_latents,
             "output_type": "pil",
-           # "denoising_start": denoising_start,                   #  for img2img
+            "denoising_start": denoising_start,                   #  for img2img
             "denoising_end": denoising_end,
             "callback_on_step_end": pyx.scheduler_swap_callback
         }       
         if use_resolution_binning:
             options["use_resolution_binning"] = True
         batch_options = options.copy()
-        #pipe2 = StableDiffusionXLImg2ImgPipeline.from_pipe(pipe)
-        #rv_image = pipe2(**batch_options).images[0]
+        pipe2 = StableDiffusionXLImg2ImgPipeline.from_pipe(pipe)
+        rv_image = pipe2(**batch_options).images[0]
         #indx=int(num_inference_steps*denoising_start)
         #pipe.scheduler.set_begin_index(indx)
         #pipe.scheduler._step_index = indx
-        rv_image = pipe(**batch_options).images[0]
+        #rv_image = pipe(**batch_options).images[0]
         print("-- got image file --")
         rv_image.save(sd_image_path,optimize=False,compress_level=0)
         pyx.upload_to_ftp(sd_image_path)    
@@ -419,7 +420,7 @@ def generate_30(
         downscale_path = f"rv_L_upscale_{timestamp}.png"
         downscale1.save(downscale_path,optimize=False,compress_level=0)
         pyx.upload_to_ftp(downscale_path) 
-
+         #  SINGLE RUN GENERATE
     if denoising_end==1.0 and denoising_start==0.0:
         generator = torch.Generator(device='cuda').manual_seed(seed)
         options = {
@@ -617,7 +618,7 @@ with gr.Blocks(theme=gr.themes.Origin(),css=css) as demo:
                 label="Guidance Scale",
                 minimum=0.1,
                 maximum=30,
-                step=0.1,
+                step=0.001,
                 value=3.8,
             )
             num_inference_steps = gr.Slider(
@@ -631,7 +632,7 @@ with gr.Blocks(theme=gr.themes.Origin(),css=css) as demo:
                 label="Denoising Start",
                 minimum=0.0,
                 maximum=0.9,
-                step=0.1,
+                step=0.05,
                 value=0.0,
             )
             
@@ -639,9 +640,16 @@ with gr.Blocks(theme=gr.themes.Origin(),css=css) as demo:
                 label="Denoising End",
                 minimum=0.1,
                 maximum=1.0,
-                step=0.1,
+                step=0.05,
                 value=1.0,
             )
+            strength = gr.Slider(
+                label="Denoising Strength",
+                minimum=0.1,
+                maximum=1.0,
+                step=0.01,
+                value=0.3,
+            )
             
 
     gr.Examples(
@@ -674,6 +682,7 @@ with gr.Blocks(theme=gr.themes.Origin(),css=css) as demo:
             num_inference_steps,
             denoising_start,
             denoising_end,
+            strength,
         ],
         outputs=[result],
     )