Update app.py

app.py

@@ -111,7 +111,7 @@ hf_hub_download(
 antelope_download = snapshot_download(repo_id="DIAMONIK7777/antelopev2", local_dir="/data/models/antelopev2")
 print(antelope_download)
 app = FaceAnalysis(name='antelopev2', root='/data', providers=['CPUExecutionProvider'])
-app.prepare(ctx_id=0, det_size=(640, 640))
+app.prepare(ctx_id=0, det_size=(768, 768))
 
 # prepare models under ./checkpoints
 face_adapter = f'/data/checkpoints/ip-adapter.bin'
@@ -144,7 +144,7 @@ pipe = StableDiffusionXLInstantIDImg2ImgPipeline.from_pretrained("SG161222/RealV
 
 pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config, use_karras_sigmas=True)
 pipe.load_ip_adapter_instantid(face_adapter)
-pipe.set_ip_adapter_scale(0.8)
+pipe.set_ip_adapter_scale(0.9)
 et = time.time()
 elapsed_time = et - st
 print('Loading pipeline took: ', elapsed_time, 'seconds')
@@ -181,7 +181,7 @@ def update_selection(selected_state: gr.SelectData, sdxl_loras, face_strength, i
 
     for lora_list in lora_defaults:
         if lora_list["model"] == sdxl_loras[selected_state.index]["repo"]:
-            face_strength = lora_list.get("face_strength", 0.95)
+            face_strength = lora_list.get("face_strength", 0.9)
             image_strength = lora_list.get("image_strength", 0.2)
             weight = lora_list.get("weight", 0.95)
             depth_control_scale = lora_list.get("depth_control_scale", 0.8)
@@ -229,8 +229,7 @@ def merge_incompatible_lora(full_path_lora, lora_scale):
                 )
                 del weights_sd
                 del lora_model
-
-def resize_image_aspect_ratio(img, max_dim=512):
+def resize_image_aspect_ratio(img, max_dim=1280):
     width, height = img.size
     aspect_ratio = width / height
 
@@ -247,7 +246,6 @@ def resize_image_aspect_ratio(img, max_dim=512):
     return img.resize((new_width, new_height), Image.LANCZOS)
 
 
-
 def run_lora(face_image, prompt, negative, lora_scale, selected_state, face_strength, image_strength, guidance_scale, depth_control_scale, sdxl_loras, custom_lora, progress=gr.Progress(track_tqdm=True)):
     print("Custom LoRA:", custom_lora)
     custom_lora_path = custom_lora[0] if custom_lora else None
@@ -265,7 +263,7 @@ def run_lora(face_image, prompt, negative, lora_scale, selected_state, face_stre
     except:
         face_detected = False
         face_emb = None
-        face_kps = face_image  # Optional face handling, verify intentional
+        face_kps = face_image
 
     et = time.time()
     print('Cropping and calculating face embeds took:', et - st, 'seconds')
@@ -299,6 +297,8 @@ def run_lora(face_image, prompt, negative, lora_scale, selected_state, face_stre
         repo_name = sdxl_loras[selected_state_index]["repo"]
         full_path_lora = state_dicts[repo_name]["saved_name"]
 
+    repo_name = repo_name.rstrip("/").lower()
+
     print("Full path LoRA", full_path_lora)
 
     et = time.time()
@@ -315,6 +315,7 @@ def run_lora(face_image, prompt, negative, lora_scale, selected_state, face_stre
 
 run_lora.zerogpu = True
 
+
 @spaces.GPU(duration=100)
 def generate_image(prompt, negative, face_emb, face_image, face_kps, image_strength, guidance_scale, face_strength, depth_control_scale, repo_name, loaded_state_dict, lora_scale, sdxl_loras, selected_state_index, face_detected, st):
     global last_fused, last_lora
@@ -325,7 +326,6 @@ def generate_image(prompt, negative, face_emb, face_image, face_kps, image_stren
     control_images = [face_kps, zoe(face_image)] if face_detected else [zoe(face_image)]
     control_scales = [face_strength, depth_control_scale] if face_detected else [depth_control_scale]
 
-    # Handle Hugging Face URL-based LoRA
     if repo_name.startswith("https://huggingface.co"):
         repo_id = repo_name.split("huggingface.co/")[-1]
         fs = HfFileSystem()
@@ -335,24 +335,21 @@ def generate_image(prompt, negative, face_emb, face_image, face_kps, image_stren
         if not safetensors_files:
             raise gr.Error("No .safetensors file found in this Hugging Face repository.")
 
-        weight_file = safetensors_files[0]  # Dynamically select the first available .safetensors file
+        weight_file = safetensors_files[0]
         full_path_lora = hf_hub_download(repo_id=repo_id, filename=weight_file, repo_type="model")
-        loaded_state_dict = load_file(full_path_lora)
     else:
-        # Use the previously loaded state_dict if not using a Hugging Face URL
-        loaded_state_dict = state_dicts[repo_name]["state_dict"]
+        full_path_lora = loaded_state_dict
 
-    # Manage LoRA weights and textual inversion embeddings
     if last_lora != repo_name:
         if last_fused:
             pipe.unfuse_lora()
             pipe.unload_lora_weights()
             pipe.unload_textual_inversion()
-        pipe.load_lora_weights(loaded_state_dict)
+
+        pipe.load_lora_weights(full_path_lora)
         pipe.fuse_lora(lora_scale)
         last_fused = True
 
-        # Handle pivotal tuning (textual inversion embeddings)
         is_pivotal = sdxl_loras[selected_state_index]["is_pivotal"]
         if is_pivotal:
             text_embedding_name = sdxl_loras[selected_state_index]["text_embedding_weights"]
@@ -361,12 +358,10 @@ def generate_image(prompt, negative, face_emb, face_image, face_kps, image_stren
             pipe.load_textual_inversion(state_dict_embedding["clip_l" if "clip_l" in state_dict_embedding else "text_encoders_0"], token=["<s0>", "<s1>"], text_encoder=pipe.text_encoder, tokenizer=pipe.tokenizer)
             pipe.load_textual_inversion(state_dict_embedding["clip_g" if "clip_g" in state_dict_embedding else "text_encoders_1"], token=["<s0>", "<s1>"], text_encoder=pipe.text_encoder_2, tokenizer=pipe.tokenizer_2)
 
-    # Prompt embeddings
     print("Processing prompt...")
     conditioning, pooled = compel(prompt)
     negative_conditioning, negative_pooled = compel(negative) if negative else (None, None)
 
-    # Image generation
     print("Generating image...")
     image = pipe(
         prompt_embeds=conditioning,
@@ -379,7 +374,7 @@ def generate_image(prompt, negative, face_emb, face_image, face_kps, image_stren
         image=face_image,
         strength=1-image_strength,
         control_image=control_images,
-        num_inference_steps=36,
+        num_inference_steps=40,
         guidance_scale=guidance_scale,
         controlnet_conditioning_scale=control_scales,
     ).images[0]
@@ -560,9 +555,9 @@ with gr.Blocks(css="custom.css") as demo:
             with gr.Accordion("Advanced options", open=False):
                 negative = gr.Textbox(label="Negative Prompt")
                 weight = gr.Slider(0, 10, value=0.9, step=0.1, label="LoRA weight")
-                face_strength = gr.Slider(0, 2, value=0.85, step=0.01, label="Face strength", info="Higher values increase the face likeness but reduce the creative liberty of the models")
-                image_strength = gr.Slider(0, 1, value=0.15, step=0.01, label="Image strength", info="Higher values increase the similarity with the structure/colors of the original photo")
-                guidance_scale = gr.Slider(0, 50, value=7, step=0.1, label="Guidance Scale")
+                face_strength = gr.Slider(0, 2, value=0.9, step=0.01, label="Face strength", info="Higher values increase the face likeness but reduce the creative liberty of the models")
+                image_strength = gr.Slider(0, 1, value=0.20, step=0.01, label="Image strength", info="Higher values increase the similarity with the structure/colors of the original photo")
+                guidance_scale = gr.Slider(0, 50, value=8, step=0.1, label="Guidance Scale")
                 depth_control_scale = gr.Slider(0, 1, value=0.8, step=0.01, label="Zoe Depth ControlNet strenght")
             prompt_title = gr.Markdown(
                 value="### Click on a LoRA in the gallery to select it",