Spaces:
Paused
Paused
Update utils/planner.py
Browse filesvariations to a single prompt
- utils/planner.py +84 -0
utils/planner.py
CHANGED
@@ -46,3 +46,87 @@ def extract_scene_plan(prompt: str) -> dict:
|
|
46 |
}
|
47 |
|
48 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
46 |
}
|
47 |
|
48 |
|
49 |
+
# utils/planner.py
|
50 |
+
|
51 |
+
import openai
|
52 |
+
import os
|
53 |
+
from openai import OpenAI
|
54 |
+
from dotenv import load_dotenv
|
55 |
+
import json
|
56 |
+
load_dotenv()
|
57 |
+
|
58 |
+
client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
|
59 |
+
|
60 |
+
# 🧠 Scene Plan Extractor
|
61 |
+
SYSTEM_INSTRUCTIONS = """
|
62 |
+
You are a scene planning assistant for an AI image generation system.
|
63 |
+
Your job is to take the user's prompt and return a structured JSON with:
|
64 |
+
- scene (environment, setting)
|
65 |
+
- subject (main actor)
|
66 |
+
- objects (main product or items)
|
67 |
+
- layout (foreground/background elements and their placement)
|
68 |
+
- rules (validation rules to ensure visual correctness)
|
69 |
+
Respond ONLY in raw JSON format. Do NOT include explanations.
|
70 |
+
"""
|
71 |
+
|
72 |
+
def extract_scene_plan(prompt: str) -> dict:
|
73 |
+
try:
|
74 |
+
response = client.chat.completions.create(
|
75 |
+
model="gpt-4o-mini-2024-07-18",
|
76 |
+
messages=[
|
77 |
+
{"role": "system", "content": SYSTEM_INSTRUCTIONS},
|
78 |
+
{"role": "user", "content": prompt}
|
79 |
+
],
|
80 |
+
temperature=0.3,
|
81 |
+
max_tokens=500
|
82 |
+
)
|
83 |
+
json_output = response.choices[0].message.content
|
84 |
+
print("🧠 Scene Plan (Raw):", json_output)
|
85 |
+
return json.loads(json_output)
|
86 |
+
|
87 |
+
except Exception as e:
|
88 |
+
print("❌ extract_scene_plan() Error:", e)
|
89 |
+
return {
|
90 |
+
"scene": None,
|
91 |
+
"subject": None,
|
92 |
+
"objects": [],
|
93 |
+
"layout": {},
|
94 |
+
"rules": {}
|
95 |
+
}
|
96 |
+
|
97 |
+
# 🧠 GPT-Based Prompt Variation Generator
|
98 |
+
def generate_prompt_variations_from_scene(scene_plan: dict, base_prompt: str, n: int = 3) -> list:
|
99 |
+
try:
|
100 |
+
system_msg = f"""
|
101 |
+
You are a creative prompt variation generator for an AI image generation system.
|
102 |
+
|
103 |
+
Given a base user prompt and its structured scene plan, generate {n} diverse image generation prompts.
|
104 |
+
Each prompt should:
|
105 |
+
- Be visually rich and descriptive
|
106 |
+
- Include stylistic or contextual variation
|
107 |
+
- Reference the same product and environment
|
108 |
+
- Stay faithful to the base prompt and extracted plan
|
109 |
+
|
110 |
+
Respond ONLY with a JSON array of strings. No explanations.
|
111 |
+
"""
|
112 |
+
|
113 |
+
response = client.chat.completions.create(
|
114 |
+
model="gpt-4o-mini-2024-07-18",
|
115 |
+
messages=[
|
116 |
+
{"role": "system", "content": system_msg},
|
117 |
+
{"role": "user", "content": json.dumps({
|
118 |
+
"base_prompt": base_prompt,
|
119 |
+
"scene_plan": scene_plan
|
120 |
+
})}
|
121 |
+
],
|
122 |
+
temperature=0.7,
|
123 |
+
max_tokens=600
|
124 |
+
)
|
125 |
+
|
126 |
+
content = response.choices[0].message.content
|
127 |
+
print("🧠 Prompt Variations (Raw):", content)
|
128 |
+
return json.loads(content)
|
129 |
+
|
130 |
+
except Exception as e:
|
131 |
+
print("❌ generate_prompt_variations_from_scene() Error:", e)
|
132 |
+
return [base_prompt] # fallback to original if anything fails
|