Update app.py (#5)

- Update app.py (507cc0db0ff895562dd859ef9498947b711ec5fd)


Co-authored-by: Liao Shutao <[email protected]>

app.py

@@ -42,6 +42,21 @@ bise_net = BiSeNet(n_classes = 19)
 bise_net.load_state_dict(torch.load(bise_net_cp_path, map_location="cpu")) # device fail
 bise_net.cuda()
 
+sys.path.append("./models/LLaVA")
+from llava.model.builder import load_pretrained_model
+from llava.mm_utils import get_model_name_from_path
+from llava.eval.run_llava import eval_model
+
+# Load Llava for prompt enhancement
+llva_model_path = "liuhaotian/llava-v1.5-7b"  
+llva_tokenizer, llva_model, llva_image_processor, llva_context_len = load_pretrained_model(
+    model_path=llva_model_path,
+    model_base=None,
+    model_name=get_model_name_from_path(llva_model_path),)    
+llva_tokenizer.cuda()
+llva_model.cuda()
+llva_image_processor.cuda()
+
 ### Load consistentID_model checkpoint
 pipe.load_ConsistentID_model(
     os.path.dirname(consistentID_path),
@@ -58,52 +73,65 @@ pipe.image_encoder.to(device)
 pipe.image_proj_model.to(device)
 pipe.FacialEncoder.to(device)
 
-# @torch.inference_mode()
-# def Enhance_prompt(prompt,select_images):
-    
-#     llva_prompt = f'Please ignore the image. Enhance the following text prompt for me. You can associate more details with the character\'s gesture, environment, and decent clothing:"{prompt}".' 
-#     args = type('Args', (), {
-#         "model_path": llva_model_path,
-#         "model_base": None,
-#         "model_name": get_model_name_from_path(llva_model_path),
-#         "query": llva_prompt,
-#         "conv_mode": None,
-#         "image_file": select_images,
-#         "sep": ",",
-#         "temperature": 0,
-#         "top_p": None,
-#         "num_beams": 1,
-#         "max_new_tokens": 512
-#     })() 
-#     Enhanced_prompt = eval_model(args, llva_tokenizer, llva_model, llva_image_processor)
-
-#     return Enhanced_prompt
+
     
 @spaces.GPU
-def process(inputImage,prompt,negative_prompt):
+def process(selected_template_images,costum_image,prompt
+        ,negative_prompt,prompt_selected,retouching,model_selected_tab,prompt_selected_tab,width,height,merge_steps):
+    
+    if model_selected_tab==0:
+        select_images = load_image(Image.open(selected_template_images))
+    else:
+        select_images = load_image(Image.fromarray(costum_image))
+
+    if prompt_selected_tab==0:
+        prompt = prompt_selected
+        negative_prompt = ""
+        need_safetycheck = False
+    else:
+        need_safetycheck = True
 
     # hyper-parameter
-    select_images = load_image(Image.fromarray(inputImage))
     num_steps = 50
-    merge_steps = 30
+    # merge_steps = 30
+            
+    @torch.inference_mode()
+    def Enhance_prompt(prompt,select_images):
+        
+        llva_prompt = f'Please ignore the image. Enhance the following text prompt for me. You can associate more details with the character\'s gesture, environment, and decent clothing:"{prompt}".' 
+        args = type('Args', (), {
+            "model_path": llva_model_path,
+            "model_base": None,
+            "model_name": get_model_name_from_path(llva_model_path),
+            "query": llva_prompt,
+            "conv_mode": None,
+            "image_file": select_images,
+            "sep": ",",
+            "temperature": 0,
+            "top_p": None,
+            "num_beams": 1,
+            "max_new_tokens": 512
+        })() 
+        Enhanced_prompt = eval_model(args, llva_tokenizer, llva_model, llva_image_processor)
     
+        return Enhanced_prompt
+
     if prompt == "":
         prompt = "A man, in a forest"
         prompt = "A man, with backpack, in a raining tropical forest, adventuring, holding a flashlight, in mist, seeking animals"
         prompt = "A person, in a sowm, wearing santa hat and a scarf, with a cottage behind"
     else:
-        # prompt=Enhance_prompt(prompt,blank_image) # TODO
-        prompt=prompt
+        prompt=Enhance_prompt(prompt,Image.new('RGB', (200, 200), color = 'white'))
         print(prompt)
         pass
-        
+
     if negative_prompt == "":
-        negative_prompt = ",monochrome, lowres, bad anatomy, worst quality, low quality, blurry"
+        negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality, blurry"
 
-    # Extend Prompt
+    #Extend Prompt
     prompt = "cinematic photo," + prompt + ", 50mm photograph, half-length portrait, film, bokeh, professional, 4k, highly detailed"
 
-    negtive_prompt_group="((((ugly)))), (((duplicate))), ((morbid)), ((mutilated)), [out of frame], extra fingers, mutated hands, ((poorly drawn hands)), ((poorly drawn face)), (((mutation))), (((deformed))), ((ugly)), blurry, ((bad anatomy)), (((bad proportions))), ((extra limbs)), cloned face, (((disfigured))). out of frame, ugly, extra limbs, (bad anatomy), gross proportions, (malformed limbs), ((missing arms)), ((missing legs)), (((extra arms))), (((extra legs))), mutated hands, (fused fingers), (too many fingers), (((long neck)))"
+    negtive_prompt_group="((cross-eye)),((cross-eyed)),(((NFSW))),(nipple),((((ugly)))), (((duplicate))), ((morbid)), ((mutilated)), [out of frame], extra fingers, mutated hands, ((poorly drawn hands)), ((poorly drawn face)), (((mutation))), (((deformed))), ((ugly)), blurry, ((bad anatomy)), (((bad proportions))), ((extra limbs)), cloned face, (((disfigured))). out of frame, ugly, extra limbs, (bad anatomy), gross proportions, (malformed limbs), ((missing arms)), ((missing legs)), (((extra arms))), (((extra legs))), mutated hands, (fused fingers), (too many fingers), (((long neck)))"
     negative_prompt = negative_prompt + negtive_prompt_group
     
     seed = torch.randint(0, 1000, (1,)).item()
@@ -111,119 +139,90 @@ def process(inputImage,prompt,negative_prompt):
 
     images = pipe(
         prompt=prompt,
-        width=512,    
-        height=768,
+        width=width,    
+        height=height,
         input_id_images=select_images,
         negative_prompt=negative_prompt,
         num_images_per_prompt=1,
         num_inference_steps=num_steps,
         start_merge_step=merge_steps,
         generator=generator,
+        retouching=retouching,
+        need_safetycheck=need_safetycheck,
     ).images[0]
 
     current_date = datetime.today()
+    return np.array(images)
 
-    output_dir = script_directory + f"/images/gradio_outputs"
-    if not os.path.exists(output_dir):
-        os.makedirs(output_dir)
-
-    images.save(os.path.join(output_dir, f"{current_date}-{seed}.jpg"))
-
-    return os.path.join(output_dir, f"{current_date}-{seed}.jpg")
-
-# # Gets the templates
-# script_directory = os.path.dirname(os.path.realpath(__file__))
-# # preset_template = glob.glob(script_directory+"/images/templates/*.png")
-# preset_template = glob.glob("./images/templates/*.png")
-# preset_template = preset_template + glob.glob("./images/templates/*.jpg")
-
-# # Use Blocks Create Gradio 
-# with gr.Blocks(title="ConsistentID Demo") as demo:
-#     gr.Markdown("# ConsistentID Demo")
-#     gr.Markdown("\
-#         Put the reference figure to be redrawn into the box below (There is a small probability of referensing failure. You can submit it repeatedly)")
-#     gr.Markdown("\
-#         If you find our work interesting, please leave a star in GitHub for us!<br>\
-#         https://github.com/JackAILab/ConsistentID")
-#     with gr.Row():
-#         with gr.Column():
-#             model_selected_tab = gr.State(0)
-#             with gr.TabItem("template images") as template_images_tab:
-#                 template_gallery_list = [(i, i) for i in preset_template]
-#                 gallery = gr.Gallery(template_gallery_list,columns=[4], rows=[2], object_fit="contain", height="auto",show_label=False)
+# Gets the templates
+script_directory = os.path.dirname(os.path.realpath(__file__))
+preset_template = glob.glob("./images/templates/*.png")
+preset_template = preset_template + glob.glob("./images/templates/*.jpg")
+
+
+with gr.Blocks(title="ConsistentID Demo") as demo:
+    gr.Markdown("# ConsistentID Demo")
+    gr.Markdown("\
+        Put the reference figure to be redrawn into the box below (There is a small probability of referensing failure. You can submit it repeatedly)")
+    gr.Markdown("\
+        If you find our work interesting, please leave a star in GitHub for us!<br>\
+        https://github.com/JackAILab/ConsistentID")
+    with gr.Row():
+        with gr.Column():
+            model_selected_tab = gr.State(0)
+            with gr.TabItem("template images") as template_images_tab:
+                template_gallery_list = [(i, i) for i in preset_template]
+                gallery = gr.Gallery(template_gallery_list,columns=[4], rows=[2], object_fit="contain", height="auto",show_label=False)
                 
-#                 def select_function(evt: gr.SelectData):
-#                     return preset_template[evt.index]
-
-#                 selected_template_images = gr.Text(show_label=False, visible=False, placeholder="Selected")
-#                 gallery.select(select_function, None, selected_template_images)
-#             with gr.TabItem("Upload Image") as upload_image_tab:
-#                 costum_image = gr.Image(label="Upload Image")
-
-#             model_selected_tabs = [template_images_tab, upload_image_tab]
-#             for i, tab in enumerate(model_selected_tabs):
-#                 tab.select(fn=lambda tabnum=i: tabnum, inputs=[], outputs=[model_selected_tab])
-
-#             with gr.Column():
-#                 prompt_selected_tab = gr.State(0)
-#                 with gr.TabItem("template prompts") as template_prompts_tab:
-#                     prompt_selected = gr.Dropdown(value="A person, police officer, half body shot", elem_id='dropdown', choices=[
-#                         "A woman in a wedding dress",
-#                         "A woman, queen, in a gorgeous palace",
-#                         "A man sitting at the beach with sunset", 
-#                         "A person, police officer, half body shot", 
-#                         "A man, sailor, in a boat above ocean",
-#                         "A women wearing headphone, listening music", 
-#                         "A man, firefighter, half body shot"], label=f"prepared prompts")
-
-#                 with gr.TabItem("custom prompt") as custom_prompt_tab:
-#                     prompt = gr.Textbox(label="prompt",placeholder="A man/woman wearing a santa hat")
-#                     nagetive_prompt = gr.Textbox(label="negative prompt",placeholder="monochrome, lowres, bad anatomy, worst quality, low quality, blurry")
+                def select_function(evt: gr.SelectData):
+                    return preset_template[evt.index]
+
+                selected_template_images = gr.Text(show_label=False, visible=False, placeholder="Selected")
+                gallery.select(select_function, None, selected_template_images)
+            with gr.TabItem("Upload Image") as upload_image_tab:
+                costum_image = gr.Image(label="Upload Image")
+
+            model_selected_tabs = [template_images_tab, upload_image_tab]
+            for i, tab in enumerate(model_selected_tabs):
+                tab.select(fn=lambda tabnum=i: tabnum, inputs=[], outputs=[model_selected_tab])
+
+            with gr.Column():
+                prompt_selected_tab = gr.State(0)
+                with gr.TabItem("template prompts") as template_prompts_tab:
+                    prompt_selected = gr.Dropdown(value="A person, police officer, half body shot", elem_id='dropdown', choices=[
+                        "A woman in a wedding dress",
+                        "A woman, queen, in a gorgeous palace",
+                        "A man sitting at the beach with sunset", 
+                        "A person, police officer, half body shot", 
+                        "A man, sailor, in a boat above ocean",
+                        "A women wearing headphone, listening music", 
+                        "A man, firefighter, half body shot"], label=f"prepared prompts")
+
+                with gr.TabItem("custom prompt") as custom_prompt_tab:
+                    prompt = gr.Textbox(label="prompt",placeholder="A man/woman wearing a santa hat")
+                    nagetive_prompt = gr.Textbox(label="negative prompt",placeholder="monochrome, lowres, bad anatomy, worst quality, low quality, blurry")
             
-#                 prompt_selected_tabs = [template_prompts_tab, custom_prompt_tab]
-#                 for i, tab in enumerate(prompt_selected_tabs):
-#                     tab.select(fn=lambda tabnum=i: tabnum, inputs=[], outputs=[prompt_selected_tab])
+                prompt_selected_tabs = [template_prompts_tab, custom_prompt_tab]
+                for i, tab in enumerate(prompt_selected_tabs):
+                    tab.select(fn=lambda tabnum=i: tabnum, inputs=[], outputs=[prompt_selected_tab])
             
-#             retouching = gr.Checkbox(label="face retouching",value=False)
-#             width = gr.Slider(label="image width",minimum=256,maximum=768,value=512,step=8)
-#             height = gr.Slider(label="image height",minimum=256,maximum=768,value=768,step=8)
-#             width.release(lambda x,y: min(1280-x,y), inputs=[width,height], outputs=[height])
-#             height.release(lambda x,y: min(1280-y,x), inputs=[width,height], outputs=[width])
-#             merge_steps = gr.Slider(label="step starting to merge facial details(30 is recommended)",minimum=10,maximum=50,value=30,step=1)
+            retouching = gr.Checkbox(label="face retouching",value=False)
+            width = gr.Slider(label="image width",minimum=256,maximum=768,value=512,step=8)
+            height = gr.Slider(label="image height",minimum=256,maximum=768,value=768,step=8)
+            width.release(lambda x,y: min(1280-x,y), inputs=[width,height], outputs=[height])
+            height.release(lambda x,y: min(1280-y,x), inputs=[width,height], outputs=[width])
+            merge_steps = gr.Slider(label="step starting to merge facial details(30 is recommended)",minimum=10,maximum=50,value=30,step=1)
             
-#             btn = gr.Button("Run")
-#         with gr.Column():
-#             out = gr.Image(label="Output")
-#             gr.Markdown('''
-#                 N.B.:<br/>
-#                 - If the proportion of face in the image is too small, the probability of an error will be slightly higher, and the similarity will also significantly decrease.)
-#                 - At the same time, use prompt with \"man\" or \"woman\" instead of \"person\" as much as possible, as that may cause the model to be confused whether the protagonist is male or female.
-#                 - Due to insufficient graphics memory on the demo server, there is an upper limit on the resolution for generating samples. We will support the generation of SDXL as soon as possible<br/><br/>
-#                 ''')
-#         btn.click(fn=process, inputs=[selected_template_images,costum_image,prompt,nagetive_prompt,prompt_selected,retouching
-#             ,model_selected_tab,prompt_selected_tab,width,height,merge_steps], outputs=out)
-
-iface = gr.Interface(
-    fn=process,
-    inputs=[
-        gr.Image(label="Upload Image"), 
-        gr.Textbox(label="prompt",placeholder="A man, in a forest, adventuring"),
-        gr.Textbox(label="negative prompt",placeholder="monochrome, lowres, bad anatomy, worst quality, low quality, blurry"),
-    ],
-    outputs=[
-        gr.Image(label="Output"), 
-    ],
-    title="ConsistentID Demo",
-    description="Put reference portrait below" ,
-    allow_flagging="never"
-)
-
-iface.launch() # zero.device
-
-# @spaces.GPU
-# def greet(n):
-#     print(zero.device) # <-- 'cuda:0' 🤗
-#     return f"Hello {zero + n} Tensor"
-
-# demo = gr.Interface(fn=greet, inputs=gr.Number(), outputs=gr.Text())
-# demo.launch()
+            btn = gr.Button("Run")
+        with gr.Column():
+            out = gr.Image(label="Output")
+            gr.Markdown('''
+                N.B.:<br/>
+                - If the proportion of face in the image is too small, the probability of an error will be slightly higher, and the similarity will also significantly decrease.)
+                - At the same time, use prompt with \"man\" or \"woman\" instead of \"person\" as much as possible, as that may cause the model to be confused whether the protagonist is male or female.
+                - Due to insufficient graphics memory on the demo server, there is an upper limit on the resolution for generating samples. We will support the generation of SDXL as soon as possible<br/><br/>
+                ''')
+        btn.click(fn=process, inputs=[selected_template_images,costum_image,prompt,nagetive_prompt,prompt_selected,retouching
+            ,model_selected_tab,prompt_selected_tab,width,height,merge_steps], outputs=out)
+
+demo.launch()