Update app.py

app.py

@@ -9,6 +9,7 @@ import base64
 import torch
 from diffusers import AutoPipelineForImage2Image
 from diffusers.utils import make_image_grid, load_image
+import uuid
 
 base_url=f'https://omnibus-top-20-img-img.hf.space/file='
 loaded_model=[]
@@ -23,31 +24,83 @@ print (loaded_model)
 pipeline = AutoPipelineForImage2Image.from_pretrained("runwayml/stable-diffusion-v1-5", safety_checker=None, variant="fp16", use_safetensors=True).to("cpu")
 pipeline.unet = torch.compile(pipeline.unet)
 
+grid_wide=10
+
+
+
+def get_concat_h_cut(in1, in2):
+    im1=Image.open(in1)
+    im2=Image.open(in2)
+    dst = Image.new('RGB', (im1.width + im2.width,
+                            min(im1.height, im2.height)))
+    dst.paste(im1, (0, 0))
+    dst.paste(im2, (im1.width, 0))
+    return dst
+
+
+def get_concat_v_cut(in1, in2):
+    im1=Image.open(in1)
+    im2=Image.open(in2)    
+    dst = Image.new(
+        'RGB', (min(im1.width, im2.width), im1.height + im2.height))
+    dst.paste(im1, (0, 0))
+    dst.paste(im2, (0, im1.height))
+    return dst
+
+
+
+
+
+
 def load_model(model_drop):
     pipeline = AutoPipelineForImage2Image.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float32, use_safetensors=True)
 
-def run_dif(prompt,im_path,model_drop,cnt,strength,guidance,infer):
+def run_dif(prompt,im_path,model_drop,cnt,strength,guidance,infer,im_height,im_width):
+    uid=uuid.uuid4()
     print(f'im_path:: {im_path}')
     print(f'im_path0:: {im_path.root[0]}')
     print(f'im_path0.image.path:: {im_path.root[0].image.path}')
     out_box=[]
-    
+    im_height=int(im_height)
+    im_width=int(im_width)
     for i,ea in enumerate(im_path.root):
-        print(f'root::{im_path.root[i]}')
-        #print(f'ea:: {ea}')        
-        #print(f'impath:: {im_path.path}')
-        url = base_url+im_path.root[i].image.path
-        print(url)
-        #init_image = load_image(url)
-        init_image=load_image(url)
-        #prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
-        
-        # pass prompt and image to pipeline
-        #image = pipeline(prompt, image=init_image, strength=0.8,guidance_scale=8.0,negative_prompt=negative_prompt,num_inference_steps=50).images[0]
-        image = pipeline(prompt, image=init_image, strength=float(strength),guidance_scale=float(guidance),num_inference_steps=int(infer)).images[0]
-        #make_image_grid([init_image, image], rows=1, cols=2)
-        out_box.append(image)
-        yield out_box,""
+        for hh in range(im_height/grid_wide):
+            for b in range(im_width/grid_wide):
+            
+                print(f'root::{im_path.root[i]}')
+                #print(f'ea:: {ea}')        
+                #print(f'impath:: {im_path.path}')
+                url = base_url+im_path.root[i].image.path
+                print(url)
+                #init_image = load_image(url)
+                init_image=load_image(url)
+                #prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
+                
+                # pass prompt and image to pipeline
+                #image = pipeline(prompt, image=init_image, strength=0.8,guidance_scale=8.0,negative_prompt=negative_prompt,num_inference_steps=50).images[0]
+                image = pipeline(prompt, image=init_image, strength=float(strength),guidance_scale=float(guidance),num_inference_steps=int(infer)).images[0]
+                #make_image_grid([init_image, image], rows=1, cols=2)
+                
+                
+                out_box.append(image)
+
+                if out_box:
+                    if len(out_box)>1:
+                        im_roll = get_concat_v_cut(f'{out_box[0]}',f'{out_box[1]}')
+                        im_roll.save(f'comb-{uid}-tmp.png')                
+                        for i in range(2,len(out_box)):
+                            im_roll = get_concat_v_cut(f'comb-{uid}-tmp.png',f'{out_box[i]}')
+                            im_roll.save(f'comb-{uid}-tmp.png')
+                        out = f'comb-{uid}-tmp.png'
+                    else:
+                        tmp_im = Image.open(out_box[0])
+                        tmp_im.save(f'comb-{uid}-tmp.png')
+                        out = f'comb-{uid}-tmp.png'
+
+
+
+                
+                yield out,""
 
 
 
@@ -155,14 +208,14 @@ def load_im(img):
     im_box=[]
     im = Image.open(img)
     width, height = im.size
-    new_w=int(width/10)
+    new_w=int(width/grid_wide)
     new_h=new_w
     w=0
     h=0
     newsize=(512,512)
     for i in range(int(height/new_h)):
         print(i)
-        for b in range(10):
+        for b in range(grid_wide):
             print(b)
             # Setting the points for cropped image
             left = w
@@ -180,7 +233,7 @@ def load_im(img):
         #yield im_box,[]
         h+=new_h
         w=0
-    yield im_box,im_box
+    yield im_box,im_box,height,width
 with gr.Blocks(css=css) as app:
     with gr.Row():
         with gr.Column():
@@ -198,11 +251,16 @@ with gr.Blocks(css=css) as app:
     with gr.Row():
         model_drop=gr.Dropdown(label="Models", choices=models, type='index', value=models[0])
         cnt = gr.Number(value=1)
+        
     out_html=gr.HTML()
-    outp=gr.Gallery(columns=10)
-    fingal=gr.Gallery(columns=10)
-    im_list=gr.Textbox()    
-    im_btn.click(load_im,inp_im,[outp,im_list])
-    go_btn = btn.click(run_dif,[inp,outp,model_drop,cnt,strength,guidance,infer],[fingal,out_html])
+    outp=gr.Gallery(columns=grid_wide)
+    #fingal=gr.Gallery(columns=grid_wide)
+    fin=gr.Image()
+    im_height=gr.Number()
+    im_width=gr.Number()
+    
+    im_list=gr.Textbox(visible=False)    
+    im_btn.click(load_im,inp_im,[outp,im_list,im_height,im_width])
+    go_btn = btn.click(run_dif,[inp,outp,model_drop,cnt,strength,guidance,infer,im_height,im_width],[fin,out_html])
     stop_btn.click(None,None,None,cancels=[go_btn])
 app.queue().launch()