minor process and ui improvements

app.py

@@ -217,13 +217,17 @@ def run_inference(config, config_history, pipeline, progress=gr.Progress(track_t
     # str_config = str_config.replace("'", '"').replace('None', 'null').replace('False', 'false')
     # config = json.loads(str_config)
 
-    if str(config["model"]) != 'None' and str(config["model"]) != 'null' and str(config["scheduler"]) != 'None':
+    if str(config["model"]) != 'None' and str(config["model"]) != 'null' and str(config["model"]) != '' and str(config["scheduler"]) != 'None':
         
-        progress((1,3), desc="Preparing pipeline initialization...")
+        progress(1, desc="Initializing pipeline...")
         
         torch.backends.cuda.matmul.allow_tf32 = get_bool(config["allow_tensorfloat32"]) # Use TensorFloat-32 as of https://huggingface.co/docs/diffusers/main/en/optimization/fp16 faster, but slightly less accurate computations
-        
-        progress((2,3), desc="Initializing pipeline...")
+                
+        # MANUAL SEED/GENERATOR
+        if config["manual_seed"] is None or config["manual_seed"] == '' or int(config["manual_seed"]) < 0:
+            generator = None
+        else:
+            generator = torch.manual_seed(int(config["manual_seed"])) 
         
         # INIT PIPELINE
         pipeline = DiffusionPipeline.from_pretrained(
@@ -232,6 +236,8 @@ def run_inference(config, config_history, pipeline, progress=gr.Progress(track_t
             torch_dtype = get_data_type(config["data_type"]), 
             variant = get_variant(config["variant"])).to(config["device"])
         
+        progress(2, desc="Setting pipeline params...")
+
         if str(config["cpu_offload"]).lower() != 'false':
             pipeline.enable_model_cpu_offload()
 
@@ -242,8 +248,17 @@ def run_inference(config, config_history, pipeline, progress=gr.Progress(track_t
         if str(config["enable_vae_slicing"]).lower() != 'false': pipeline.enable_vae_slicing()
         if str(config["enable_vae_tiling"]).lower() != 'false': pipeline.enable_vae_tiling()
                 
+        # SAFETY CHECKER
+        if str(config["safety_checker"]).lower() == 'false': pipeline.safety_checker = None 
+        pipeline.requires_safety_checker = get_bool(config["requires_safety_checker"])
+            
+        # SCHEDULER/SOLVER
+        pipeline.scheduler = get_scheduler(config["scheduler"], pipeline.scheduler.config)
+        
         # INIT REFINER
         if str(config['refiner']).lower() != 'none' and str(config['refiner']).lower() != 'null':
+
+            progress(3, desc="Initializing refiner...")
             refiner = DiffusionPipeline.from_pretrained(
                     config['refiner'],
                     text_encoder_2=pipeline.text_encoder_2,
@@ -258,38 +273,28 @@ def run_inference(config, config_history, pipeline, progress=gr.Progress(track_t
             if str(config["enable_vae_slicing"]).lower() != 'false': refiner.enable_vae_slicing()
             if str(config["enable_vae_tiling"]).lower() != 'false': refiner.enable_vae_tiling()
             
-        # SAFETY CHECKER
-        if str(config["safety_checker"]).lower() == 'false': pipeline.safety_checker = None 
-        pipeline.requires_safety_checker = get_bool(config["requires_safety_checker"])
             
-        # SCHEDULER/SOLVER
-        pipeline.scheduler = get_scheduler(config["scheduler"], pipeline.scheduler.config)
-        
-        # MANUAL SEED/GENERATOR
-        if config["manual_seed"] is None or config["manual_seed"] == '' or int(config["manual_seed"]) < 0:
-            generator = None
-        else:
-            generator = torch.manual_seed(int(config["manual_seed"])) 
-        
         # ADAPTERS
         # TEXTUAL INVERSION
-        if str(config["adapter_textual_inversion"]).lower() != 'none' and str(config["adapter_textual_inversion"]).lower() != 'null':
+        if str(config["adapter_textual_inversion"]).lower() != 'none' and str(config["adapter_textual_inversion"]).lower() != 'null' and str(config["adapter_textual_inversion"]).lower() != '':
+            progress(4, desc=f"Loading textual inversion adapter {config['adapter_textual_inversion']}...")
             pipeline.load_textual_inversion(config["adapter_textual_inversion"], token=config["adapter_textual_inversion_token"])
         
         # LoRA
         if len(config["adapter_lora"]) > 0 and len(config["adapter_lora"]) == len(config["adapter_lora_weight"]):
             adapter_lora_balancing = []
             for adapter_lora_index, adapter_lora in enumerate(config["adapter_lora"]):
+                progress(5, desc=f"Loading LoRA adapters {config['adapter_lora']}...")
                 if str(config["adapter_lora_weight"][adapter_lora_index]).lower() != 'none':
                     pipeline.load_lora_weights(adapter_lora, weight_name=config["adapter_lora_weight"][adapter_lora_index], adapter_name=config["adapter_lora_token"][adapter_lora_index])
                 else:
                     pipeline.load_lora_weights(adapter_lora, adapter_name=config["adapter_lora_token"][adapter_lora_index])
                 adapter_lora_balancing.append(config["adapter_lora_balancing"][adapter_lora])
         
-            adapter_weights = {adapter_lora_balancing}
-            pipeline.set_adapters({config["adapter_lora_token"]}, adapter_weights=adapter_weights)
+            adapter_weights = adapter_lora_balancing
+            pipeline.set_adapters(config["adapter_lora_token"], adapter_weights=adapter_weights)
             
-        progress((3,3), desc="Creating the result...")
+        progress(6, desc="Inferencing...")
 
         prompt = config["prompt"] + config["trigger_token"] + config["adapter_textual_inversion_token"] + ' '.join(config["adapter_lora_token"])
         
@@ -309,11 +314,12 @@ def run_inference(config, config_history, pipeline, progress=gr.Progress(track_t
             
         config_history.append(config.copy())
 
+    # expected output: out_image, out_config_history, config_history, pipeline
         return image[0], dict_list_to_markdown_table(config_history), config_history, pipeline
     
     else:
     
-        return "Please select a model AND a scheduler.", None, config_history, pipeline
+        return "Please select a model AND a scheduler.", "Please select a model AND a scheduler.", None, pipeline
 
 appConfig = load_app_config()
 models = appConfig.get("models", {})

config.py

@@ -212,7 +212,7 @@ def assemble_code(str_config):
         code.append(f'generator = torch.manual_seed(manual_seed)')
 
     # ADAPTER
-    if str(config["adapter_textual_inversion"]).lower() != 'none':
+    if str(config["adapter_textual_inversion"]).lower() != 'none' and str(config["adapter_textual_inversion"]).lower() != 'null' and str(config["adapter_textual_inversion"]).lower() != '':
         code.append(f'pipeline.load_textual_inversion("{config["adapter_textual_inversion"]}", token="{config["adapter_textual_inversion_token"]}")')
 
     if len(config["adapter_lora"]) > 0 and len(config["adapter_lora"]) == len(config["adapter_lora_weight"]):
@@ -227,7 +227,7 @@ def assemble_code(str_config):
         code.append(f'adapter_weights = {adapter_lora_balancing}')
         code.append(f'pipeline.set_adapters({config["adapter_lora_token"]}, adapter_weights=adapter_weights)')
 
-    code.append(f'prompt = "{config["prompt"]} {config["trigger_token"]} {config["adapter_textual_inversion_token"]} {" ".join(config["adapter_lora_token"])}"')
+    code.append(f'prompt = "{config["prompt"]} {config["trigger_token"]} {config["adapter_textual_inversion_token"]} {", ".join(config["adapter_lora_token"])}"')
     code.append(f'negative_prompt = "{config["negative_prompt"]}"')
     code.append(f'inference_steps = {config["inference_steps"]}')
     code.append(f'guidance_scale = {config["guidance_scale"]}')