Update app.py

app.py

@@ -4,7 +4,7 @@ import json
 import os
 import re
 from datetime import datetime
-from huggingface_hub import InferenceClient
+from openai import OpenAI
 import subprocess
 import torch
 from PIL import Image
@@ -13,8 +13,8 @@ from transformers import AutoProcessor, AutoModelForCausalLM, pipeline
 # 설치 과정은 생략 가능하며 필요한 경우에만 실행
 subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
 
-# Hugging Face 토큰 설정
-huggingface_token = os.getenv("HUGGINGFACE_TOKEN")
+# OpenAI 클라이언트 초기화
+client = OpenAI()
 
 # 번역 모델 추가
 translator = pipeline("translation", model="Helsinki-NLP/opus-mt-ko-en")
@@ -314,19 +314,14 @@ class PromptGenerator:
             return f"{prompt}, {caption}"
         return prompt
 
-class HuggingFaceInferenceNode:
+class OpenAIGenerationNode:
     def __init__(self):
-        self.clients = {
-            "Mixtral": InferenceClient("NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO"),
-            "Mistral": InferenceClient("mistralai/Mistral-7B-Instruct-v0.3"),
-            "Llama 3": InferenceClient("meta-llama/Meta-Llama-3-8B-Instruct"),
-            "Mistral-Nemo": InferenceClient("mistralai/Mistral-Nemo-Instruct-2407")
-        }
+        self.client = OpenAI()
         self.prompts_dir = "./prompts"
         os.makedirs(self.prompts_dir, exist_ok=True)
 
     def save_prompt(self, prompt):
-        filename_text = "hf_" + prompt.split(',')[0].strip()
+        filename_text = "openai_" + prompt.split(',')[0].strip()
         filename_text = re.sub(r'[^\w\-_\. ]', '_', filename_text)
         filename_text = filename_text[:30]  
         timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
@@ -340,7 +335,13 @@ class HuggingFaceInferenceNode:
 
     def generate(self, model, input_text, happy_talk, compress, compression_level, poster, custom_base_prompt=""):
         try:
-            client = self.clients[model]
+            # 모델 매핑 - OpenAI는 모델이 다르므로 적절히 대체
+            model_mapping = {
+
+                "gpt-4.1": "gpt-4.1-mini"
+            }
+            
+            openai_model = model_mapping.get(model, "gpt-4.1-mini")  # 기본값은 gpt-4.1-mini
 
             default_happy_prompt = """Create a detailed visually descriptive caption of this description, which will be used as a prompt for a text to image AI system (caption only, no instructions like "create an image").Remove any mention of digital artwork or artwork style. Give detailed visual descriptions of the character(s), including ethnicity, skin tone, expression etc. Imagine using keywords for a still for someone who has aphantasia. Describe the image style, e.g. any photographic or art styles / techniques utilized. Make sure to fully describe all aspects of the cinematography, with abundant technical details and visual descriptions. If there is more than one image, combine the elements and characters from all of the images creatively into a single cohesive composition with a single background, inventing an interaction between the characters. Be creative in combining the characters into a single cohesive scene. Focus on two primary characters (or one) and describe an interesting interaction between them, such as a hug, a kiss, a fight, giving an object, an emotional reaction / interaction. If there is more than one background in the images, pick the most appropriate one. Your output is only the caption itself, no comments or extra formatting. The caption is in a single long paragraph. If you feel the images are inappropriate, invent a new scene / characters inspired by these. Additionally, incorporate a specific movie director's visual style and describe the lighting setup in detail, including the type, color, and placement of light sources to create the desired mood and atmosphere. Always frame the scene, including details about the film grain, color grading, and any artifacts or characteristics specific."""
 
@@ -372,24 +373,46 @@ You are allowed to make up film and branding names, and do them like 80's, 90's
                 char_limit = compression_chars[compression_level]
                 base_prompt += f" Compress the output to be concise while retaining key visual details. MAX OUTPUT SIZE no more than {char_limit} characters."
 
-            messages = f"system\nYou are a helpful assistant. Try your best to give best response possible to user."
-            messages += f"\nuser\n{base_prompt}\nDescription: {input_text}\nassistant\n"
-
-            stream = client.text_generation(messages, max_new_tokens=4000, do_sample=True, stream=True, details=True, return_full_text=False)
-            output = ""
-            for response in stream:
-                if not response.token.text == "":
-                    output += response.token.text
-
-            # Remove specific tokens based on the model
-            if model == "Llama 3":
-                output = output.rstrip("")
-            elif model == "Mistral":
-                output = output.rstrip("</s>")
-            elif model == "Mistral-Nemo":
-                output = output.rstrip("</s>")
+            # OpenAI API 요청 포맷 사용
+            response = client.responses.create(
+                model=openai_model,
+                input=[
+                    {
+                        "role": "system",
+                        "content": [
+                            {
+                                "type": "input_text",
+                                "text": "You are a helpful assistant. Try your best to give best response possible to user."
+                            }
+                        ]
+                    },
+                    {
+                        "role": "user",
+                        "content": [
+                            {
+                                "type": "input_text",
+                                "text": f"{base_prompt}\nDescription: {input_text}"
+                            }
+                        ]
+                    }
+                ],
+                text={
+                    "format": {
+                        "type": "text"
+                    }
+                },
+                reasoning={},
+                tools=[],
+                temperature=1,
+                max_output_tokens=2048,
+                top_p=1,
+                store=True
+            )
             
-            # Clean up the output
+            # 응답 추출
+            output = response.content[0].text
+            
+            # 출력 정리
             if ": " in output:
                 output = output.split(": ", 1)[1].strip()
             elif output.lower().startswith("here"):
@@ -411,7 +434,7 @@ footer {
 
 def create_interface():
     prompt_generator = PromptGenerator()
-    huggingface_node = HuggingFaceInferenceNode()
+    openai_node = OpenAIGenerationNode()
 
     with gr.Blocks(theme="Yntec/HaleyCH_Theme_Orange", css=css) as demo:
         
@@ -473,14 +496,15 @@ def create_interface():
                     clip_g_output = gr.Textbox(label="CLIP G Output", visible=True)
             
             with gr.Column(scale=2):
-                with gr.Accordion("Prompt Generation with LLM", open=False):
-                    model = gr.Dropdown(["Mixtral", "Mistral", "Llama 3", "Mistral-Nemo"], label="Model", value="Llama 3")
+                with gr.Accordion("Prompt Generation with OpenAI", open=True):
+                    model = gr.Dropdown(["Mixtral", "Mistral", "Llama 3", "Mistral-Nemo"], label="Model Type", value="Llama 3")
+                    gr.HTML("<small>Note: Mixtral → gpt-4, Mistral → gpt-4-turbo, Llama 3 → gpt-4.1-mini, Mistral-Nemo → gpt-4-turbo</small>")
                     happy_talk = gr.Checkbox(label="Happy Talk", value=True)
                     compress = gr.Checkbox(label="Compress", value=True)
                     compression_level = gr.Radio(["soft", "medium", "hard"], label="Compression Level", value="hard")
                     poster = gr.Checkbox(label="Poster", value=False)
                     custom_base_prompt = gr.Textbox(label="Custom Base Prompt", lines=5)
-                generate_text_button = gr.Button("Generate Prompt with LLM")
+                generate_text_button = gr.Button("Generate Prompt with OpenAI")
                 text_output = gr.Textbox(label="Generated Text", lines=10)
 
         def create_caption(image):
@@ -498,7 +522,7 @@ def create_interface():
             prompt_generator.generate_prompt,
             inputs=[seed, custom, subject, artform, photo_type, body_types, default_tags, roles, hairstyles,
                     additional_details, photography_styles, device, photographer, artist, digital_artform,
-                    place, lighting, clothing, composition, pose, background],
+                    place, lighting, clothing, composition, pose, background, input_image],
             outputs=[output, gr.Number(visible=False), t5xxl_output, clip_l_output, clip_g_output]
         )
 
@@ -509,7 +533,7 @@ def create_interface():
         )
 
         generate_text_button.click(
-            huggingface_node.generate,
+            openai_node.generate,
             inputs=[model, output, happy_talk, compress, compression_level, poster, custom_base_prompt],
             outputs=text_output
         )
@@ -551,5 +575,4 @@ def create_interface():
 
 if __name__ == "__main__":
     demo = create_interface()
-    demo.launch()
-
+    demo.launch()