diff --git a/README.md b/README.md
index 65959bc..9a5462f 100644
--- a/README.md
+++ b/README.md
@@ -1,7 +1,19 @@
 # ComfyUI-MaskBatchPermutations
- Permutes a mask batch to present possible additive combinations. Passing a mask batch (e.g. out of [SEGS to Mask Batch](https://github.com/ltdrdata/ComfyUI-Impact-Pack)) will return a new mask batch representing all the possible combinations of the included masks. So, a mask batch with two mask sections, "A" and "B, will return a batch containing an empty mask, an empty mask & A, an empty mask & B, and an empty mask & A & B.
- ## example workflow
- ![An image embedding a workflow showing this node being used with a mask batch with three items.](workflow_example.png)
- This image contains an embedded workflow.
- ## why?
- "Automatic" face detailing without direct operator intervention usually works well, but occasionally it wrecks an otherwise good face. Instead of having to hand compose them back together in something like GIMP and fix the metadata, I decided to create this instead.
+Provides two nodes, Permute Mask Batch and Combinatorial Detailer.
+
+## Permute Mask Batch
+Passing a mask batch (e.g. out of [SEGS to Mask Batch](https://github.com/ltdrdata/ComfyUI-Impact-Pack)) will return a new mask batch representing all the possible combinations of the included masks. So, a mask batch with two masks, "A" and "B, will return a new batch containing an empty mask, an empty mask & A, an empty mask & B, and an empty mask & A & B.
+
+### example workflow
+![An image embedding a workflow showing this node being used with a mask batch with three items.](workflow_example.png)
+This image contains an embedded workflow.
+
+## Combinatorial Detailer
+Similar to Permute Mask Batch but accepts a mask batch, a base image, and then a batch of candidate images (for example, the batched outputs of several separate detailer passes using different prompts or seeds). Provides a batch of images representing the possible combinations of the base image, masks and candidates. Be advised that this can create very large batches - a set of masks representing three regions and with two candidate images will generate (2 + 1)<sup>3</sup> combinations (27) as each mask area will present with either the base image, or one of the two candidates.
+
+### example workflow
+In this example, I have given both detailers deliberately divergent prompts to make it clearer in the example output what is going on.
+![An image embedding a workflow showing this node being used with a mask batch with three items and two candidates.](workflow_combi.png)
+
+# why?
+"Automatic" face detailing without direct operator intervention usually works well, but occasionally it wrecks an otherwise good face. Instead of having to hand compose them back together in something like GIMP and fix the metadata, I decided to create this instead.
diff --git a/__init__.py b/__init__.py
index e82ed67..fddc816 100644
--- a/__init__.py
+++ b/__init__.py
@@ -35,14 +35,67 @@ class PermuteMaskBatch:
                 output[i] = combined
         return (output,)
 
+class CombinatorialDetailer:
+    
+    # no internal state
+    def __init__(self):
+        pass
+
+    @classmethod
+    def INPUT_TYPES(s):
+        """
+        Input: mask
+        """
+        return {
+            "required": {
+                "masks": ("MASK",),
+                "base_image": ("IMAGE",),
+                "candidates": ("IMAGE",),
+            },
+        }
+    
+    RETURN_TYPES = ("IMAGE",)
+    RETURN_NAMES = ("IMAGE",)
+    FUNCTION = "combinatorialDetailer"
+    OUTPUT_NODE = False
+    CATEGORY = "image"
+
+    def combinatorialDetailer(self, masks, base_image, candidates):
+        candidate_count, height, width, _ = candidates.shape
+        mask_count = masks.shape[0]
+        expanded_masks = [x.unsqueeze(-1) for x in masks]
+
+        # Each mask area can be in one of `n + 1` states (all candidates + base)
+        num_combinations = (candidate_count + 1) ** mask_count
+
+        output_images = torch.zeros((num_combinations, height, width, 3), dtype=base_image.dtype)
+        output_images[0] = base_image[0]
+
+        # Iterate over all other possible combinations
+        for i in range(1, num_combinations):
+            combined_image = base_image[0].clone()
+            current_combination = i
+            for mask_index in range(mask_count):
+                selected_candidate = current_combination % (candidate_count + 1)
+                # print("out image", i, "mask index", mask_index, "selected candidate", selected_candidate)
+                if selected_candidate != 0:
+                    combined_image = torch.where(expanded_masks[mask_index] == 1, candidates[selected_candidate - 1], combined_image)
+                current_combination //= (candidate_count + 1)
+            output_images[i] = combined_image
+
+        return (output_images,)
+
+
 NODE_CLASS_MAPPINGS= {
     "PermuteMaskBatch": PermuteMaskBatch,
+    "CombinatorialDetailer": CombinatorialDetailer,
 }
 NODE_DISPLAY_NAME_MAPPINGS = {
-    "PermuteMaskBatch": 'Permute Mask Batch'
+    "PermuteMaskBatch": "Permute Mask Batch",
+    "CombinatorialDetailer": "Combinatorial Detailer",
 }
 
-__version__ = '1.0.0'
-__all__ = ['NODE_CLASS_MAPPINGS', 'NODE_DISPLAY_NAME_MAPPINGS']
+__version__ = "1.1.0"
+__all__ = ["NODE_CLASS_MAPPINGS", "NODE_DISPLAY_NAME_MAPPINGS"]
 
 
diff --git a/workflow_combi.png b/workflow_combi.png
new file mode 100644
index 0000000..4f69e47
Binary files /dev/null and b/workflow_combi.png differ