Aatricks's picture
Deploy ZeroGPU Gradio Space snapshot
b701455
Raw
History Blame Contribute Delete
15.4 kB
import io
import os
import threading
import queue
import numpy as np
import logging
from PIL import Image
from PIL.PngImagePlugin import PngInfo
logger = logging.getLogger(__name__)
output_directory = "./output"
# Maximum number of images that will be saved in a single `save_images` call.
# Higher counts are likely to indicate tiled intermediate outputs which should
# not be saved as individual image files to avoid filling disk with tiles.
# Can be configured at runtime via the `LD_MAX_IMAGES_PER_SAVE` environment
# variable (default: 16).
MAX_IMAGES_PER_SAVE = int(os.getenv("LD_MAX_IMAGES_PER_SAVE", "16"))
# In-memory image buffer for API responses (avoids disk round-trip)
# Maps request_filename_prefix -> list of (filename, subfolder, png_bytes)
_image_bytes_buffer: dict[str, list[tuple[str, str, bytes]]] = {}
_image_bytes_lock = threading.Lock()
def store_image_bytes(prefix: str, filename: str, subfolder: str, data: bytes) -> None:
"""Store image bytes in memory for later retrieval by the API server."""
with _image_bytes_lock:
_image_bytes_buffer.setdefault(prefix, []).append((filename, subfolder, data))
def pop_image_bytes(prefix: str) -> list[tuple[str, str, bytes]]:
"""Pop and return all stored image byte entries for a given prefix.
Returns a list of (filename, subfolder, png_bytes) tuples.
"""
with _image_bytes_lock:
return _image_bytes_buffer.pop(prefix, [])
def get_output_directory() -> str:
"""#### Get the output directory.
#### Returns:
- `str`: The output directory.
"""
global output_directory
return output_directory
def get_save_image_path(
filename_prefix: str, output_dir: str, image_width: int = 0, image_height: int = 0
) -> tuple:
"""#### Get the save image path.
#### Args:
- `filename_prefix` (str): The filename prefix.
- `output_dir` (str): The output directory.
- `image_width` (int, optional): The image width. Defaults to 0.
- `image_height` (int, optional): The image height. Defaults to 0.
#### Returns:
- `tuple`: The full output folder, filename, counter, subfolder, and filename prefix.
"""
def map_filename(filename: str) -> tuple:
prefix_len = len(os.path.basename(filename_prefix))
prefix = filename[: prefix_len + 1]
try:
digits = int(filename[prefix_len + 1 :].split("_")[0])
except (ValueError, IndexError):
digits = 0
return (digits, prefix)
def compute_vars(input: str, image_width: int, image_height: int) -> str:
input = input.replace("%width%", str(image_width))
input = input.replace("%height%", str(image_height))
return input
filename_prefix = compute_vars(filename_prefix, image_width, image_height)
subfolder = os.path.dirname(os.path.normpath(filename_prefix))
filename = os.path.basename(os.path.normpath(filename_prefix))
full_output_folder = os.path.join(output_dir, subfolder)
subfolder_paths = [
os.path.join(full_output_folder, x)
for x in ["Classic", "HiresFix", "Img2Img", "Adetailer", "ControlNet"]
]
for path in subfolder_paths:
os.makedirs(path, exist_ok=True)
# Find highest counter across all subfolders
counter = 1
for path in subfolder_paths:
if os.path.exists(path):
files = os.listdir(path)
if files:
numbers = [
map_filename(f)[0]
for f in files
if f.startswith(filename) and f.endswith(".png")
]
if numbers:
counter = max(max(numbers) + 1, counter)
return full_output_folder, filename, counter, subfolder, filename_prefix
MAX_RESOLUTION = 16384
class SaveImage:
"""#### Class for saving images."""
def __init__(self):
"""#### Initialize the SaveImage class."""
self.output_dir = get_output_directory()
self.type = "output"
self.prefix_append = ""
self.compress_level = 4
def save_images(
self,
images: list,
filename_prefix: str = "LD",
prompt: str = None,
extra_pnginfo: dict = None,
store_bytes_prefix: str | None = None,
) -> dict:
"""#### Save images to the output directory.
#### Args:
- `images` (list): The list of images.
- `filename_prefix` (str, optional): The filename prefix. Defaults to "LD".
- `prompt` (str, optional): The prompt. Defaults to None.
- `extra_pnginfo` (dict, optional): Additional PNG info. Defaults to None.
- `store_bytes_prefix` (str, optional): If set, also buffer PNG bytes in memory
under this key for zero-disk-IO API retrieval.
#### Returns:
- `dict`: The saved images information.
"""
filename_prefix += self.prefix_append
# Safety: compute total number of images to be saved in this call, counting
# batched tensors as multiple images. Abort early if count exceeds threshold.
total_images = 0
for image in images:
shape = getattr(image, 'shape', None)
if shape is None:
total_images += 1
continue
try:
if len(shape) >= 4:
total_images += int(shape[0])
else:
total_images += 1
except Exception:
total_images += 1
if total_images > MAX_IMAGES_PER_SAVE:
# Diagnostic: record basic info about incoming images to help trace
# the source of excessive image counts (tiling issues, batched tensors)
details = []
try:
for idx, image in enumerate(images[:10]):
try:
shape = getattr(image, 'shape', None)
dtype = getattr(image, 'dtype', None)
tname = type(image).__name__
details.append(f"idx={idx} type={tname} shape={shape} dtype={dtype}")
except Exception as e:
details.append(f"idx={idx} inspect_failed: {e}")
more = f" (+{max(0, len(images)-10)} more)" if len(images) > 10 else ""
except Exception:
details = ["failed to enumerate images"]
more = ""
logger.warning(
"Attempting to save %d images in a single call (exceeds MAX_IMAGES_PER_SAVE=%d). "
"This may indicate tiled intermediate outputs; aborting save to avoid creating many tile files. "
"filename_prefix=%s store_bytes_prefix=%s Details: %s%s",
total_images,
MAX_IMAGES_PER_SAVE,
filename_prefix,
store_bytes_prefix,
"; ".join(details),
more,
)
return {"ui": {"images": []}}
full_output_folder, filename, counter, subfolder, filename_prefix = (
get_save_image_path(
filename_prefix,
self.output_dir,
images[0].shape[-2],
images[0].shape[-1],
)
)
results = list()
for batch_number, image in enumerate(images):
# Convert tensor to numpy and handle different dimensions
i = image.cpu().numpy()
# Handle batched tensors (4D: [batch, channels, height, width] or [batch, height, width, channels])
if i.ndim == 4:
# Process each image in the batch separately
for sub_batch_idx in range(i.shape[0]):
sub_image = i[sub_batch_idx] # Extract single image from batch
# Convert to HWC format if in CHW format
if sub_image.shape[0] in [1, 3, 4] and sub_image.shape[0] < min(
sub_image.shape[1], sub_image.shape[2]
):
sub_image = np.transpose(sub_image, (1, 2, 0)) # CHW -> HWC
# Squeeze single channel dimension if present
if sub_image.shape[-1] == 1:
sub_image = sub_image.squeeze(-1)
# Scale to 0-255 range
sub_image_scaled = np.clip(sub_image * 255.0, 0, 255).astype(
np.uint8
)
img = Image.fromarray(sub_image_scaled)
# Attach PNG text metadata if provided
if extra_pnginfo:
metadata = PngInfo()
for k, v in extra_pnginfo.items():
try:
metadata.add_text(str(k), str(v))
except Exception:
# Ensure metadata writing never blocks saving
pass
else:
metadata = None
filename_with_batch_num = filename.replace(
"%batch_num%", str(batch_number)
)
file = f"{filename_with_batch_num}_{counter:05}_.png"
# Save the image to appropriate subfolder
save_path = full_output_folder
if filename_prefix == "LD-HF":
save_path = os.path.join(full_output_folder, "HiresFix")
elif filename_prefix == "LD-I2I":
save_path = os.path.join(full_output_folder, "Img2Img")
elif filename_prefix == "LD-CN":
save_path = os.path.join(full_output_folder, "ControlNet")
elif filename_prefix == "LD-head" or filename_prefix == "LD-body":
save_path = os.path.join(full_output_folder, "Adetailer")
else:
save_path = os.path.join(full_output_folder, "Classic")
img.save(
os.path.join(save_path, file),
pnginfo=metadata,
compress_level=self.compress_level,
)
# Buffer PNG bytes in memory for API responses (avoids re-read)
if store_bytes_prefix:
buf = io.BytesIO()
img.save(buf, format="PNG", pnginfo=metadata, compress_level=self.compress_level)
save_rel_bytes = os.path.relpath(save_path, "./output")
store_image_bytes(store_bytes_prefix, file, save_rel_bytes, buf.getvalue())
# Return the actual subfolder relative to ./output so callers can locate files
save_rel = os.path.relpath(save_path, "./output")
results.append(
{
"filename": file,
"subfolder": save_rel,
"requested_subfolder": subfolder,
"type": self.type,
}
)
counter += 1
continue # Skip the rest of the loop for this batch
# Handle 3D tensors (single image: [channels, height, width] or [height, width, channels])
elif i.ndim == 3:
# Convert to HWC format if in CHW format
if i.shape[0] in [1, 3, 4] and i.shape[0] < min(i.shape[1], i.shape[2]):
i = np.transpose(i, (1, 2, 0)) # CHW -> HWC
# Squeeze single channel dimension if present
if i.shape[-1] == 1:
i = i.squeeze(-1)
# Handle 2D tensors (grayscale: [height, width])
elif i.ndim == 2:
pass # Already in correct format
else:
raise ValueError(f"Unexpected tensor dimensions: {i.shape}")
# Scale to 0-255 range and convert to PIL Image
i_scaled = np.clip(i * 255.0, 0, 255).astype(np.uint8)
img = Image.fromarray(i_scaled)
# Attach PNG text metadata if provided
if extra_pnginfo:
metadata = PngInfo()
for k, v in extra_pnginfo.items():
try:
metadata.add_text(str(k), str(v))
except Exception:
pass
else:
metadata = None
filename_with_batch_num = filename.replace("%batch_num%", str(batch_number))
file = f"{filename_with_batch_num}_{counter:05}_.png"
# Save the image to appropriate subfolder
save_path = full_output_folder
if filename_prefix == "LD-HF":
save_path = os.path.join(full_output_folder, "HiresFix")
elif filename_prefix == "LD-I2I":
save_path = os.path.join(full_output_folder, "Img2Img")
elif filename_prefix == "LD-CN":
save_path = os.path.join(full_output_folder, "ControlNet")
elif filename_prefix == "LD-Flux":
save_path = os.path.join(full_output_folder, "Flux")
elif filename_prefix == "LD-head" or filename_prefix == "LD-body":
save_path = os.path.join(full_output_folder, "Adetailer")
else:
save_path = os.path.join(full_output_folder, "Classic")
img.save(
os.path.join(save_path, file),
pnginfo=metadata,
compress_level=self.compress_level,
)
# Buffer PNG bytes in memory for API responses (avoids re-read)
if store_bytes_prefix:
buf = io.BytesIO()
img.save(buf, format="PNG", pnginfo=metadata, compress_level=self.compress_level)
save_rel_bytes = os.path.relpath(save_path, "./output")
store_image_bytes(store_bytes_prefix, file, save_rel_bytes, buf.getvalue())
# Return the actual subfolder relative to ./output so callers can locate files
save_rel = os.path.relpath(save_path, "./output")
results.append(
{
"filename": file,
"subfolder": save_rel,
"requested_subfolder": subfolder,
"type": self.type,
}
)
counter += 1
return {"ui": {"images": results}}
def save_images_async(
self,
images: list,
filename_prefix: str = "LD",
prompt: str = None,
extra_pnginfo: dict = None,
) -> threading.Thread:
"""#### Save images asynchronously in a background thread.
#### Returns:
- `threading.Thread`: The background thread handling the save.
"""
# Create copies of tensors on CPU to free GPU memory immediately
cpu_images = [img.detach().cpu().clone() for img in images]
thread = threading.Thread(
target=self.save_images,
args=(cpu_images, filename_prefix, prompt, extra_pnginfo)
)
thread.start()
return thread