Merge pull request #4 from deforum-art/akatz-dev

Updated with new latent node, bumped pyproject version

Author: akatz-ai

__init__.py

@@ -2,15 +2,18 @@
 from .utility_nodes import UTILITY_NODE_CLASS_MAPPINGS, UTILITY_NODE_DISPLAY_NAME_MAPPINGS
 from .conditions import CONDITION_NODE_CLASS_MAPPINGS, CONDITION_NODE_DISPLAY_NAME_MAPPINGS
 from .lazy_nodes import GENERAL_NODE_CLASS_MAPPINGS, GENERAL_NODE_DISPLAY_NAME_MAPPINGS
+from .latent_nodes import LATENT_NODE_CLASS_MAPPINGS, LATENT_NODE_DISPLAY_NAME_MAPPINGS
 
 NODE_CLASS_MAPPINGS = {}
 NODE_CLASS_MAPPINGS.update(FLOW_CONTROL_NODE_CLASS_MAPPINGS)
 NODE_CLASS_MAPPINGS.update(UTILITY_NODE_CLASS_MAPPINGS)
 NODE_CLASS_MAPPINGS.update(CONDITION_NODE_CLASS_MAPPINGS)
 NODE_CLASS_MAPPINGS.update(GENERAL_NODE_CLASS_MAPPINGS)
+NODE_CLASS_MAPPINGS.update(LATENT_NODE_CLASS_MAPPINGS)
 
 NODE_DISPLAY_NAME_MAPPINGS = {}
 NODE_DISPLAY_NAME_MAPPINGS.update(FLOW_CONTROL_NODE_DISPLAY_NAME_MAPPINGS)
 NODE_DISPLAY_NAME_MAPPINGS.update(UTILITY_NODE_DISPLAY_NAME_MAPPINGS)
 NODE_DISPLAY_NAME_MAPPINGS.update(CONDITION_NODE_DISPLAY_NAME_MAPPINGS)
 NODE_DISPLAY_NAME_MAPPINGS.update(GENERAL_NODE_DISPLAY_NAME_MAPPINGS)
+NODE_DISPLAY_NAME_MAPPINGS.update(LATENT_NODE_DISPLAY_NAME_MAPPINGS)

base_node.py

@@ -23,4 +23,6 @@ class DebugNode(BaseNode):
 
 class ListNode(BaseNode):
     CATEGORY = f"{NODE_NAME}/Lists"
-    
+    
+class LatentNode(BaseNode):
+    CATEGORY = f"{NODE_NAME}/Latent"

latent_nodes.py

@@ -0,0 +1,111 @@
+import re
+import torch
+from .tools import VariantSupport
+from .base_node import NODE_NAME, LatentNode
+import comfy.model_management
+
+MAX_RESOLUTION=32768
+
+# from  https://discuss.pytorch.org/t/help-regarding-slerp-function-for-generative-model-sampling/32475
+def slerp(val, low, high):
+    dims = low.shape
+
+    #flatten to batches
+    low = low.reshape(dims[0], -1)
+    high = high.reshape(dims[0], -1)
+
+    low_norm = low/torch.norm(low, dim=1, keepdim=True)
+    high_norm = high/torch.norm(high, dim=1, keepdim=True)
+
+    # in case we divide by zero
+    low_norm[low_norm != low_norm] = 0.0
+    high_norm[high_norm != high_norm] = 0.0
+
+    omega = torch.acos((low_norm*high_norm).sum(1))
+    so = torch.sin(omega)
+    res = (torch.sin((1.0-val)*omega)/so).unsqueeze(1)*low + (torch.sin(val*omega)/so).unsqueeze(1) * high
+    return res.reshape(dims)
+
+@VariantSupport()
+class SeedInterpNoise(LatentNode):
+    """
+    Produce a batch of noise tensors whose seeds advance by +1 after
+    every (interp_steps + 1) frames, with SLERP interpolation between
+    the two anchor seeds.
+
+    Example (start_seed=1, interp_steps=2, frames=7):
+        seeds: 1   1--2   1--2   2   2--3   2--3   3
+                ε0  ε1    ε2    ε3  ε4     ε5     ε6
+                ▲   ▲          ▲           ▲
+                |   |__________|           |
+                |______ segment 1 _____|   |___ seg 2 …
+    """
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {"required": {
+                    "source": (["CPU", "GPU"],),
+                    "start_seed": ("INT",  {"default": 0, "min": 0,
+                                            "max": 0xffffffffffffffff}),
+                    "frames":     ("INT",  {"default": 8, "min": 1,
+                                            "max": 9999999}),
+                    "interp_steps": ("INT", {"default": 1, "min": 0,
+                                             "max": 1024}),
+                    "width":  ("INT", {"default": 512, "min": 64,
+                                       "max": MAX_RESOLUTION, "step": 8}),
+                    "height": ("INT", {"default": 512, "min": 64,
+                                       "max": MAX_RESOLUTION, "step": 8}),
+               }}
+
+    RETURN_TYPES = ("LATENT",)          # matches NoisyLatentImage
+    FUNCTION = "build"
+
+    def build(self, source, start_seed, frames, interp_steps,
+              width, height):
+        device = "cpu" if source == "CPU" \
+                 else comfy.model_management.get_torch_device()
+        c, h, w = 4, height // 8, width // 8
+
+        # pre-allocate output tensor
+        batch = torch.empty((frames, c, h, w),
+                            dtype=torch.float32, device=device)
+
+        # helper: epsilon(seed)
+        def eps(seed):
+            g = torch.Generator(device).manual_seed(seed)
+            return torch.randn((c, h, w), generator=g, device=device)
+
+        seg_len = interp_steps + 1      # ε_a + k in-betweens + ε_b
+        cur_seed = start_seed
+        frame_idx = 0
+
+        while frame_idx < frames:
+            eps_a = eps(cur_seed)
+            eps_b = eps(cur_seed + 1)
+
+            # anchor A
+            batch[frame_idx] = eps_a
+            frame_idx += 1
+            if frame_idx >= frames:
+                break
+
+            # interpolated frames
+            for k in range(1, seg_len):
+                if frame_idx >= frames:
+                    break
+                t = k / seg_len
+                batch[frame_idx] = slerp(t, eps_a, eps_b)
+                frame_idx += 1
+
+            # anchor B (start of next segment) only if we still need frames
+            cur_seed += 1
+
+        return ({"samples": batch.cpu()}, )
+
+LATENT_NODE_CLASS_MAPPINGS = {
+    "SeedInterpNoise": SeedInterpNoise,
+}
+
+LATENT_NODE_DISPLAY_NAME_MAPPINGS = {
+    "SeedInterpNoise": f"Seed Interp Noise | {NODE_NAME}",
+}

pyproject.toml

@@ -1,7 +1,7 @@
 [project]
 name = "comfyui-deforum"
 description = "Custom nodes for deforum workflows"
-version = "0.2.0"
+version = "0.2.1"
 license = { file = "LICENSE.txt" }
 
 [project.urls]