Create ideogram_block.py
Browse files- ideogram_block.py +150 -0
ideogram_block.py
ADDED
|
@@ -0,0 +1,150 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import requests
|
| 3 |
+
from typing import Optional
|
| 4 |
+
from PIL import Image
|
| 5 |
+
from io import BytesIO
|
| 6 |
+
from diffusers.utils import load_image
|
| 7 |
+
from diffusers.modular_pipelines import (
|
| 8 |
+
PipelineState,
|
| 9 |
+
ModularPipelineBlocks,
|
| 10 |
+
InputParam,
|
| 11 |
+
OutputParam,
|
| 12 |
+
)
|
| 13 |
+
|
| 14 |
+
IDEOGRAM_GENERATE_URL = "https://api.ideogram.ai/v1/ideogram-v3/generate"
|
| 15 |
+
|
| 16 |
+
class CreateCharacterImageBlock(ModularPipelineBlocks):
|
| 17 |
+
|
| 18 |
+
@property
|
| 19 |
+
def inputs(self) -> list[InputParam]:
|
| 20 |
+
return [
|
| 21 |
+
InputParam(
|
| 22 |
+
name="prompt",
|
| 23 |
+
type_hint=str,
|
| 24 |
+
description="Text prompt describing the desired image."
|
| 25 |
+
),
|
| 26 |
+
InputParam(
|
| 27 |
+
name="character_image",
|
| 28 |
+
type_hint=Image.Image,
|
| 29 |
+
description="A single PIL Image to use as the character reference."
|
| 30 |
+
),
|
| 31 |
+
]
|
| 32 |
+
|
| 33 |
+
@property
|
| 34 |
+
def intermediate_outputs(self) -> list[OutputParam]:
|
| 35 |
+
return [
|
| 36 |
+
OutputParam(
|
| 37 |
+
name="image",
|
| 38 |
+
type_hint=Image.Image,
|
| 39 |
+
description="Generated image."
|
| 40 |
+
),
|
| 41 |
+
OutputParam(
|
| 42 |
+
name="image_url",
|
| 43 |
+
type_hint=str,
|
| 44 |
+
description="URL to the generated image."
|
| 45 |
+
)
|
| 46 |
+
]
|
| 47 |
+
|
| 48 |
+
@staticmethod
|
| 49 |
+
def generate_image_with_character_reference(
|
| 50 |
+
prompt: str,
|
| 51 |
+
character_image: Image.Image, *,
|
| 52 |
+
style_type: str = "AUTO", # "GENERAL", "REALISTIC", "DESIGN", "FICTION"
|
| 53 |
+
seed: Optional[int] = None,
|
| 54 |
+
resolution: Optional[str] = "1024x1024",
|
| 55 |
+
request_timeout: int = 60,
|
| 56 |
+
) -> str:
|
| 57 |
+
"""
|
| 58 |
+
Generate an image using Ideogram's character reference feature.
|
| 59 |
+
|
| 60 |
+
Args:
|
| 61 |
+
api_key: Your Ideogram API key.
|
| 62 |
+
prompt: Text prompt describing the desired image.
|
| 63 |
+
character_image: A single PIL Image to use as the character reference.
|
| 64 |
+
style_type: Ideogram style selection. "AUTO" lets the API decide.
|
| 65 |
+
seed: Random seed for reproducibility.
|
| 66 |
+
resolution: Image resolution.
|
| 67 |
+
request_timeout: Timeout for HTTP requests (seconds).
|
| 68 |
+
|
| 69 |
+
Returns:
|
| 70 |
+
str: URL to the generated image (default)
|
| 71 |
+
|
| 72 |
+
Raises:
|
| 73 |
+
RuntimeError on API errors or missing results.
|
| 74 |
+
"""
|
| 75 |
+
# Resolve API key from environment
|
| 76 |
+
api_key = os.getenv("IDEOGRAM_API_KEY")
|
| 77 |
+
|
| 78 |
+
if not api_key:
|
| 79 |
+
raise RuntimeError(
|
| 80 |
+
"IDEOGRAM_API_KEY is not set. Provide api_key param or set env var IDEOGRAM_API_KEY."
|
| 81 |
+
)
|
| 82 |
+
|
| 83 |
+
headers = {"Api-Key": api_key}
|
| 84 |
+
|
| 85 |
+
# Validate image type
|
| 86 |
+
if not isinstance(character_image, Image.Image):
|
| 87 |
+
raise TypeError("character_image must be a PIL.Image.Image")
|
| 88 |
+
|
| 89 |
+
# Validate seed and mutually exclusive resolution/aspect_ratio
|
| 90 |
+
if seed is not None:
|
| 91 |
+
if not (0 <= int(seed) <= 2147483647):
|
| 92 |
+
raise ValueError("seed must be between 0 and 2147483647 inclusive")
|
| 93 |
+
|
| 94 |
+
# Build multipart with a single character reference image
|
| 95 |
+
with BytesIO() as buf:
|
| 96 |
+
character_image.save(buf, format="PNG")
|
| 97 |
+
buf.seek(0)
|
| 98 |
+
files = [
|
| 99 |
+
(
|
| 100 |
+
"character_reference_images",
|
| 101 |
+
("character.png", buf, "image/png"),
|
| 102 |
+
)
|
| 103 |
+
]
|
| 104 |
+
|
| 105 |
+
data: dict = {
|
| 106 |
+
"prompt": prompt,
|
| 107 |
+
"style_type": style_type,
|
| 108 |
+
}
|
| 109 |
+
if seed is not None:
|
| 110 |
+
data["seed"] = str(int(seed))
|
| 111 |
+
if resolution is not None:
|
| 112 |
+
data["resolution"] = resolution
|
| 113 |
+
|
| 114 |
+
resp = requests.post(
|
| 115 |
+
IDEOGRAM_GENERATE_URL,
|
| 116 |
+
headers=headers,
|
| 117 |
+
data=data,
|
| 118 |
+
files=files,
|
| 119 |
+
timeout=request_timeout,
|
| 120 |
+
)
|
| 121 |
+
if not resp.ok:
|
| 122 |
+
raise RuntimeError(f"Ideogram API error {resp.status_code}: {resp.text}")
|
| 123 |
+
|
| 124 |
+
payload = resp.json()
|
| 125 |
+
# Expected: payload['data'][0]['url']
|
| 126 |
+
try:
|
| 127 |
+
image_url = payload["data"][0]["url"]
|
| 128 |
+
except Exception as e:
|
| 129 |
+
raise RuntimeError(f"Unexpected Ideogram response format: {payload}") from e
|
| 130 |
+
|
| 131 |
+
return image_url
|
| 132 |
+
|
| 133 |
+
def __call__(self, components: dict, state: PipelineState):
|
| 134 |
+
|
| 135 |
+
block_state = self.get_block_state(state)
|
| 136 |
+
|
| 137 |
+
if isinstance(block_state.character_image, str):
|
| 138 |
+
character_image = load_image(block_state.character_image)
|
| 139 |
+
elif isinstance(block_state.character_image, Image.Image):
|
| 140 |
+
character_image = block_state.character_image
|
| 141 |
+
else:
|
| 142 |
+
raise ValueError(f"Invalid character image type: {type(block_state.character_image)}")
|
| 143 |
+
|
| 144 |
+
block_state.image_url = self.generate_image_with_character_reference(
|
| 145 |
+
prompt=block_state.prompt,
|
| 146 |
+
character_image=character_image,
|
| 147 |
+
)
|
| 148 |
+
block_state.image = load_image(block_state.image_url)
|
| 149 |
+
self.set_block_state(state, block_state)
|
| 150 |
+
return components, state
|