Add files via upload

nodes/FV_ClampImagesToMegapixels.py

@@ -0,0 +1,67 @@
+import torch
+import torch.nn.functional as F
+import math
+
+class ClampImagesMegapixels:
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "images": ("IMAGE",),  # batch [B, H, W, C]
+                "min_mp": ("FLOAT", {"min": 0.1, "max": 100.0, "step": 0.1, "default": 0.6}),
+                "max_mp": ("FLOAT", {"min": 0.1, "max": 100.0, "step": 0.1, "default": 1.0}),
+                "multiple_of": ("INT", {"min": 1, "max": 512, "step": 1, "default": 64}),
+            }
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    RETURN_NAMES = ("images",)
+    FUNCTION = "run"
+    CATEGORY = "Fictiverse/Resize"
+
+    def run(self, images, min_mp, max_mp, multiple_of):
+        B, H, W, C = images.shape
+        current_pixels = H * W
+        
+        # Conversion MP -> Pixels absolus
+        limit_upper_pixels = int(max_mp * 1_000_000)
+        limit_lower_pixels = int(min_mp * 1_000_000)
+        
+        # 1. Définir la cible (Clamp)
+        # Si < min, on vise min.
+        # Si > max, on vise max.
+        # Sinon, on vise la taille actuelle.
+        target_pixels = current_pixels
+        if target_pixels < limit_lower_pixels:
+            target_pixels = limit_lower_pixels
+        elif target_pixels > limit_upper_pixels:
+            target_pixels = limit_upper_pixels
+        
+        # Préparation calculs dimensions
+        aspect_ratio = W / H
+        new_height = math.sqrt(target_pixels / aspect_ratio)
+        new_width = new_height * aspect_ratio
+
+        # 2. Quantification Spatiale (Arrondi au multiple de 'multiple_of')
+        new_width = int(round(new_width / multiple_of) * multiple_of)
+        new_height = int(round(new_height / multiple_of) * multiple_of)
+        
+        # Optimisation : Si les dimensions calculées sont identiques à l'original, on renvoie l'original
+        if new_height == H and new_width == W:
+             return (images,)
+
+        # 3. Exécution du Resize
+        # Conversion [B, H, W, C] -> [B, C, H, W] pour pytorch
+        img_batch = images.permute(0, 3, 1, 2).float()
+        
+        resized = F.interpolate(img_batch, size=(new_height, new_width), mode='bilinear', align_corners=False)
+
+        # Retour au format ComfyUI [B, H, W, C]
+        resized = resized.permute(0, 2, 3, 1).to(images.dtype)
+
+        return (resized,)
+
+# Enregistrement de la node
+NODE_CLASS_MAPPINGS = {
+    "Clamp Images To Megapixels": ClampImagesMegapixels
+}