update

app.py

@@ -7,15 +7,15 @@ import torch
 
 from transformers import AutoProcessor, AutoModelForCausalLM
 
-checkpoint = "microsoft/git-base"
-processor = AutoProcessor.from_pretrained(checkpoint)
-model = AutoModelForCausalLM.from_pretrained(checkpoint)
-
 openai.organization = os.getenv("API_ORG")
 openai.api_key = os.getenv("API_KEY")
 app_password = os.getenv("APP_PASSWORD")
 app_username = os.getenv("APP_USERNAME")
 
+checkpoint = "openai/clip-vit-base-patch32"
+processor = AutoProcessor.from_pretrained(checkpoint)
+model = AutoModelForCausalLM.from_pretrained(checkpoint)
+
 def generate(input_image):
     device = "cuda" if torch.cuda.is_available() else "cpu"
     inputs = processor(images=input_image, return_tensors="pt").to(device)

app2.py

@@ -0,0 +1,69 @@
+import gradio as gr
+import openai
+import os
+import requests
+
+openai.organization = os.getenv("API_ORG")
+openai.api_key = os.getenv("API_KEY")
+app_password = os.getenv("APP_PASSWORD")
+app_username = os.getenv("APP_USERNAME")
+
+def generate_prompt(input):
+    prompt = """You are a prompt writing support system for image generation AI.
+From the user's input, You output prompt in English that should be input to the image generation AI, imagining its intent as much as possible.
+You are not allowed to ask questions of the user.
+You will always output only brief prompt in English to be input to the image generation AI.
+Your output will always English.
+Input from user:
+"""
+    response = openai.ChatCompletion.create(
+        model = "gpt-3.5-turbo",
+        messages = [{"role": "system", "content": prompt+input}],
+        max_tokens=256
+    )
+    generated_text = response['choices'][0]['message']['content'].strip()
+    return "Make the illustration a photo: "+generated_text
+
+def get_related_caption(prompt):
+    url = "https://api.irasutoya.nibo.sh/semantic-search"
+    params = {'q': prompt}
+    headers = {"content-type": "application/json"}
+    r = requests.get(url, params=params, headers=headers)
+    data = r.json()
+    return data['illustrations'][0]['description']
+
+def generate(prompt):
+    caption = get_related_caption(prompt)
+    generated_prompt = generate_prompt(caption)
+    response = openai.Image.create(
+      prompt=generated_prompt,
+      n=1,
+      size="256x256"
+    )
+    return caption, generated_prompt, response['data'][0]['url']
+
+with gr.Blocks() as demo:
+    with gr.Column():
+        with gr.Row():
+            with gr.Column():
+                prompt_text = gr.Textbox(lines=5, label="Prompt")
+                prompt_examples = gr.Examples(
+                            examples=[
+                                "きのこの山",
+                                "たけのこの里",
+                                "お菓子の家",
+                            ],
+                            inputs=[prompt_text],
+                            outputs=None,
+                        )
+                btn = gr.Button(value="Generate Image")
+
+            with gr.Column():
+                caption = gr.Textbox(lines=5, label="Related Caption")
+                generated_prompt = gr.Textbox(lines=5, label="Generated Prompt")
+                out_image = gr.components.Image(type="filepath", label="Generated Image")
+
+    btn.click(generate, inputs=[prompt_text], outputs=[caption, generated_prompt, out_image])
+    demo.load()
+
+demo.launch(share=True, auth=(app_username, app_password))