Create prompt_expander.py
Browse files- prompt_expander.py +67 -0
prompt_expander.py
ADDED
@@ -0,0 +1,67 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
from typing import List
|
3 |
+
from diffusers.modular_pipelines import (
|
4 |
+
PipelineState,
|
5 |
+
ModularPipelineBlocks,
|
6 |
+
InputParam,
|
7 |
+
OutputParam,
|
8 |
+
)
|
9 |
+
import google.generativeai as genai
|
10 |
+
import os
|
11 |
+
|
12 |
+
SYSTEM_PROMPT = (
|
13 |
+
"You are an expert image generation assistant. "
|
14 |
+
"Take the user's short description and expand it into a vivid, detailed, and clear image generation prompt. "
|
15 |
+
"Ensure rich colors, depth, realistic lighting, and an imaginative composition. "
|
16 |
+
"Avoid vague terms — be specific about style, perspective, and mood. "
|
17 |
+
"Try to keep the output under 512 tokens."
|
18 |
+
)
|
19 |
+
|
20 |
+
class GeminiPromptExpander(ModularPipelineBlocks):
|
21 |
+
def __init__(self, model_id="gemini-2.5-flash-lite", system_prompt=SYSTEM_PROMPT):
|
22 |
+
super().__init__()
|
23 |
+
api_key = os.getenv("GOOGLE_API_KEY")
|
24 |
+
if api_key is None:
|
25 |
+
raise ValueError("Must provide an API key for Gemini through the `GOOGLE_API_KEY` env variable.")
|
26 |
+
genai.configure(api_key=api_key)
|
27 |
+
self.model = genai.GenerativeModel(model_name=model_id, system_instruction=system_prompt)
|
28 |
+
|
29 |
+
@property
|
30 |
+
def expected_components(self):
|
31 |
+
return []
|
32 |
+
|
33 |
+
@property
|
34 |
+
def inputs(self) -> List[InputParam]:
|
35 |
+
return [
|
36 |
+
InputParam(
|
37 |
+
"prompt",
|
38 |
+
type_hint=str,
|
39 |
+
required=True,
|
40 |
+
description="Prompt to use",
|
41 |
+
)
|
42 |
+
]
|
43 |
+
|
44 |
+
@property
|
45 |
+
def intermediate_outputs(self) -> List[OutputParam]:
|
46 |
+
return [
|
47 |
+
OutputParam(
|
48 |
+
"prompt",
|
49 |
+
type_hint=str,
|
50 |
+
description="Expanded prompt by the LLM",
|
51 |
+
),
|
52 |
+
OutputParam(
|
53 |
+
"old_prompt",
|
54 |
+
type_hint=str,
|
55 |
+
description="Old prompt provided by the user",
|
56 |
+
)
|
57 |
+
]
|
58 |
+
|
59 |
+
|
60 |
+
def __call__(self, components, state: PipelineState) -> PipelineState:
|
61 |
+
block_state = self.get_block_state(state)
|
62 |
+
|
63 |
+
block_state.old_prompt = block_state.prompt
|
64 |
+
block_state.prompt = self.model.generate_content(block_state.old_prompt).text
|
65 |
+
self.set_block_state(state, block_state)
|
66 |
+
|
67 |
+
return components, state
|