fix resizing and versioning

invokeai/app/invocations/flux_denoise.py

@@ -63,7 +63,7 @@ from invokeai.backend.util.devices import TorchDevice
     title="FLUX Denoise",
     tags=["image", "flux"],
     category="image",
-    version="4.0.0",
+    version="4.1.0",
 )
 class FluxDenoiseInvocation(BaseInvocation):
     """Run denoising process with a FLUX transformer model."""

invokeai/backend/flux/extensions/kontext_extension.py

@@ -3,6 +3,9 @@ import numpy as np
 import torch
 from einops import repeat
 from PIL import Image
+import torchvision.transforms as T
+from dataclasses import dataclass
+from typing
 
 from invokeai.app.invocations.fields import FluxKontextConditioningField
 from invokeai.app.invocations.flux_vae_encode import FluxVaeEncodeInvocation
@@ -12,6 +15,80 @@ from invokeai.backend.flux.sampling_utils import pack
 from invokeai.backend.flux.util import PREFERED_KONTEXT_RESOLUTIONS
 
 
+@dataclass
+class KontextResizeConstraints:
+    """Configurable constraints for FLUX Kontext image resizing."""
+
+    min_dimension: int = 672
+    max_dimension: int = 1568
+    multiple: int = 16  # Ensure dimensions are multiples of this value
+
+
+def calculate_optimal_kontext_dimensions(
+    width: int, height: int, constraints: KontextResizeConstraints = KontextResizeConstraints()
+) -> tuple[int, int]:
+    """Calculate optimal dimensions for FLUX Kontext conditioning.
+
+    This function ensures:
+    1. Dimensions are multiples of the specified multiple
+    2. Dimensions are within the valid range
+    3. Aspect ratio is preserved as closely as possible
+    4. Implicit optimization for performance by maximizing within constraints
+
+    Args:
+        width: Original image width
+        height: Original image height
+        constraints: Resize constraints configuration
+
+    Returns:
+        Tuple of (optimal_width, optimal_height)
+    """
+    aspect_ratio = width / height
+
+    # Calculate the maximum possible dimensions while staying within bounds
+    if aspect_ratio >= 1.0:  # Landscape or square
+        # Width is the larger dimension
+        max_width = min(constraints.max_dimension, width)
+        max_width = (max_width // constraints.multiple) * constraints.multiple
+        optimal_width = max(constraints.min_dimension, max_width)
+        optimal_height = int(optimal_width / aspect_ratio)
+        optimal_height = (optimal_height // constraints.multiple) * constraints.multiple
+        optimal_height = max(constraints.min_dimension, optimal_height)
+    else:  # Portrait
+        # Height is the larger dimension
+        max_height = min(constraints.max_dimension, height)
+        max_height = (max_height // constraints.multiple) * constraints.multiple
+        optimal_height = max(constraints.min_dimension, max_height)
+        optimal_width = int(optimal_height * aspect_ratio)
+        optimal_width = (optimal_width // constraints.multiple) * constraints.multiple
+        optimal_width = max(constraints.min_dimension, optimal_width)
+
+    return optimal_width, optimal_height
+
+
+def resize_to_optimal_kontext_resolution(
+    image: Image.Image, constraints: KontextResizeConstraints = KontextResizeConstraints()
+) -> Image.Image:
+    """Resize image to optimal FLUX Kontext resolution.
+
+    This ensures the image dimensions are compatible with the FLUX packing operation
+    and provides optimal performance for Kontext conditioning.
+
+    Args:
+        image: PIL Image to resize
+        constraints: Resize constraints configuration
+
+    Returns:
+        Resized PIL Image with optimal dimensions
+    """
+    width, height = image.size
+    optimal_width, optimal_height = calculate_optimal_kontext_dimensions(width, height, constraints)
+
+    # Resize the image to the optimal resolution
+    resized_image = image.resize((optimal_width, optimal_height), Image.Resampling.LANCZOS)
+    return resized_image
+
+
 def generate_img_ids_with_offset(
     latent_height: int,
     latent_width: int,