Create nano_banana.py
Browse files- nano_banana.py +83 -0
nano_banana.py
ADDED
|
@@ -0,0 +1,83 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
|
| 2 |
+
from typing import List
|
| 3 |
+
from diffusers.modular_pipelines import (
|
| 4 |
+
PipelineState,
|
| 5 |
+
ModularPipelineBlocks,
|
| 6 |
+
InputParam,
|
| 7 |
+
OutputParam,
|
| 8 |
+
)
|
| 9 |
+
from PIL import Image
|
| 10 |
+
import google.generativeai as genai
|
| 11 |
+
import os
|
| 12 |
+
|
| 13 |
+
|
| 14 |
+
class NanoBanana(ModularPipelineBlocks):
|
| 15 |
+
def __init__(self, model_id="gemini-2.5-flash-image-preview"):
|
| 16 |
+
super().__init__()
|
| 17 |
+
api_key = os.getenv("GEMINI_API_KEY")
|
| 18 |
+
if api_key is None:
|
| 19 |
+
raise ValueError("Must provide an API key for Gemini through the `GEMINI_API_KEY` env variable.")
|
| 20 |
+
genai.configure(api_key=api_key)
|
| 21 |
+
self.model = genai.GenerativeModel(model_name=model_id)
|
| 22 |
+
|
| 23 |
+
@property
|
| 24 |
+
def expected_components(self):
|
| 25 |
+
return []
|
| 26 |
+
|
| 27 |
+
@property
|
| 28 |
+
def inputs(self) -> List[InputParam]:
|
| 29 |
+
return [
|
| 30 |
+
InputParam(
|
| 31 |
+
"image",
|
| 32 |
+
type_hint=Image.Image,
|
| 33 |
+
required=False,
|
| 34 |
+
description="Image to use"
|
| 35 |
+
),
|
| 36 |
+
InputParam(
|
| 37 |
+
"prompt",
|
| 38 |
+
type_hint=str,
|
| 39 |
+
required=True,
|
| 40 |
+
description="Prompt to use",
|
| 41 |
+
)
|
| 42 |
+
]
|
| 43 |
+
|
| 44 |
+
@property
|
| 45 |
+
def intermediate_inputs(self) -> List[InputParam]:
|
| 46 |
+
return []
|
| 47 |
+
|
| 48 |
+
@property
|
| 49 |
+
def intermediate_outputs(self) -> List[OutputParam]:
|
| 50 |
+
return [
|
| 51 |
+
OutputParam(
|
| 52 |
+
"output_image",
|
| 53 |
+
type_hint=str,
|
| 54 |
+
description="Output image",
|
| 55 |
+
),
|
| 56 |
+
OutputParam(
|
| 57 |
+
"old_image",
|
| 58 |
+
type_hint=str,
|
| 59 |
+
description="Old image (if) provided by the user",
|
| 60 |
+
)
|
| 61 |
+
]
|
| 62 |
+
|
| 63 |
+
|
| 64 |
+
def __call__(self, components, state: PipelineState) -> PipelineState:
|
| 65 |
+
block_state = self.get_block_state(state)
|
| 66 |
+
|
| 67 |
+
old_image = block_state.image
|
| 68 |
+
prompt = block_state.state.prompt
|
| 69 |
+
contents = [prompt]
|
| 70 |
+
if old_image is not None:
|
| 71 |
+
contents.expand(old_image)
|
| 72 |
+
|
| 73 |
+
output = self.model.generate_content(contents=contents)
|
| 74 |
+
block_state.output_image = output
|
| 75 |
+
|
| 76 |
+
if old_image is not None:
|
| 77 |
+
block_state.old_image = old_image
|
| 78 |
+
else:
|
| 79 |
+
block_state.old_image = None
|
| 80 |
+
|
| 81 |
+
self.set_block_state(state, block_state)
|
| 82 |
+
|
| 83 |
+
return components, state
|