feat: add CurveInput ABC with MonotoneCubicCurve implementation

CurveInput is an abstract base class so future curve representations
(bezier, LUT-based, analytical functions) can be added without breaking
downstream nodes that type-check against CurveInput.

MonotoneCubicCurve is the concrete implementation that:
- Mirrors frontend createMonotoneInterpolator (curveUtils.ts) exactly
- Pre-computes slopes as numpy arrays at construction time
- Provides vectorised interp_array() using numpy for batch evaluation
- interp() for single-value evaluation
- to_lut() for generating lookup tables

CurveEditor node wraps raw widget points in MonotoneCubicCurve.

comfy/comfy_types/node_typing.py

@@ -176,8 +176,8 @@ class InputTypeOptions(TypedDict):
     """COMBO type only. Specifies the configuration for a multi-select widget.
     Available after ComfyUI frontend v1.13.4
     https://github.com/Comfy-Org/ComfyUI_frontend/pull/2987"""
-    gradient_stops: NotRequired[list[list[float]]]
-    """Gradient color stops for gradientslider display mode. Each stop is [offset, r, g, b] (``FLOAT``)."""
+    gradient_stops: NotRequired[list[dict]]
+    """Gradient color stops for gradientslider display mode. Each stop is {"offset": float, "color": [r, g, b]}."""
 
 
 class HiddenInputTypeDict(TypedDict):

comfy/ldm/flux/model.py

@@ -44,6 +44,22 @@ class FluxParams:
     txt_norm: bool = False
 
 
+def invert_slices(slices, length):
+    sorted_slices = sorted(slices)
+    result = []
+    current = 0
+
+    for start, end in sorted_slices:
+        if current < start:
+            result.append((current, start))
+        current = max(current, end)
+
+    if current < length:
+        result.append((current, length))
+
+    return result
+
+
 class Flux(nn.Module):
     """
     Transformer model for flow matching on sequences.
@@ -138,6 +154,7 @@ class Flux(nn.Module):
         y: Tensor,
         guidance: Tensor = None,
         control = None,
+        timestep_zero_index=None,
         transformer_options={},
         attn_mask: Tensor = None,
     ) -> Tensor:
@@ -164,10 +181,6 @@ class Flux(nn.Module):
             txt = self.txt_norm(txt)
         txt = self.txt_in(txt)
 
-        vec_orig = vec
-        if self.params.global_modulation:
-            vec = (self.double_stream_modulation_img(vec_orig), self.double_stream_modulation_txt(vec_orig))
-
         if "post_input" in patches:
             for p in patches["post_input"]:
                 out = p({"img": img, "txt": txt, "img_ids": img_ids, "txt_ids": txt_ids, "transformer_options": transformer_options})
@@ -182,6 +195,24 @@ class Flux(nn.Module):
         else:
             pe = None
 
+        vec_orig = vec
+        txt_vec = vec
+        extra_kwargs = {}
+        if timestep_zero_index is not None:
+            modulation_dims = []
+            batch = vec.shape[0] // 2
+            vec_orig = vec_orig.reshape(2, batch, vec.shape[1]).movedim(0, 1)
+            invert = invert_slices(timestep_zero_index, img.shape[1])
+            for s in invert:
+                modulation_dims.append((s[0], s[1], 0))
+            for s in timestep_zero_index:
+                modulation_dims.append((s[0], s[1], 1))
+            extra_kwargs["modulation_dims_img"] = modulation_dims
+            txt_vec = vec[:batch]
+
+        if self.params.global_modulation:
+            vec = (self.double_stream_modulation_img(vec_orig), self.double_stream_modulation_txt(txt_vec))
+
         blocks_replace = patches_replace.get("dit", {})
         transformer_options["total_blocks"] = len(self.double_blocks)
         transformer_options["block_type"] = "double"
@@ -195,7 +226,8 @@ class Flux(nn.Module):
                                                    vec=args["vec"],
                                                    pe=args["pe"],
                                                    attn_mask=args.get("attn_mask"),
-                                                   transformer_options=args.get("transformer_options"))
+                                                   transformer_options=args.get("transformer_options"),
+                                                   **extra_kwargs)
                     return out
 
                 out = blocks_replace[("double_block", i)]({"img": img,
@@ -213,7 +245,8 @@ class Flux(nn.Module):
                                  vec=vec,
                                  pe=pe,
                                  attn_mask=attn_mask,
-                                 transformer_options=transformer_options)
+                                 transformer_options=transformer_options,
+                                 **extra_kwargs)
 
             if control is not None: # Controlnet
                 control_i = control.get("input")
@@ -230,6 +263,12 @@ class Flux(nn.Module):
         if self.params.global_modulation:
             vec, _ = self.single_stream_modulation(vec_orig)
 
+        extra_kwargs = {}
+        if timestep_zero_index is not None:
+            lambda a: 0 if a == 0 else a + txt.shape[1]
+            modulation_dims_combined = list(map(lambda x: (0 if x[0] == 0 else x[0] + txt.shape[1], x[1] + txt.shape[1], x[2]), modulation_dims))
+            extra_kwargs["modulation_dims"] = modulation_dims_combined
+
         transformer_options["total_blocks"] = len(self.single_blocks)
         transformer_options["block_type"] = "single"
         transformer_options["img_slice"] = [txt.shape[1], img.shape[1]]
@@ -242,7 +281,8 @@ class Flux(nn.Module):
                                        vec=args["vec"],
                                        pe=args["pe"],
                                        attn_mask=args.get("attn_mask"),
-                                       transformer_options=args.get("transformer_options"))
+                                       transformer_options=args.get("transformer_options"),
+                                       **extra_kwargs)
                     return out
 
                 out = blocks_replace[("single_block", i)]({"img": img,
@@ -253,7 +293,7 @@ class Flux(nn.Module):
                                                           {"original_block": block_wrap})
                 img = out["img"]
             else:
-                img = block(img, vec=vec, pe=pe, attn_mask=attn_mask, transformer_options=transformer_options)
+                img = block(img, vec=vec, pe=pe, attn_mask=attn_mask, transformer_options=transformer_options, **extra_kwargs)
 
             if control is not None: # Controlnet
                 control_o = control.get("output")
@@ -264,7 +304,11 @@ class Flux(nn.Module):
 
         img = img[:, txt.shape[1] :, ...]
 
-        img = self.final_layer(img, vec_orig)  # (N, T, patch_size ** 2 * out_channels)
+        extra_kwargs = {}
+        if timestep_zero_index is not None:
+            extra_kwargs["modulation_dims"] = modulation_dims
+
+        img = self.final_layer(img, vec_orig, **extra_kwargs)  # (N, T, patch_size ** 2 * out_channels)
         return img
 
     def process_img(self, x, index=0, h_offset=0, w_offset=0, transformer_options={}):
@@ -312,13 +356,16 @@ class Flux(nn.Module):
         w_len = ((w_orig + (patch_size // 2)) // patch_size)
         img, img_ids = self.process_img(x, transformer_options=transformer_options)
         img_tokens = img.shape[1]
+        timestep_zero_index = None
         if ref_latents is not None:
+            ref_num_tokens = []
             h = 0
             w = 0
             index = 0
             ref_latents_method = kwargs.get("ref_latents_method", self.params.default_ref_method)
+            timestep_zero = ref_latents_method == "index_timestep_zero"
             for ref in ref_latents:
-                if ref_latents_method == "index":
+                if ref_latents_method in ("index", "index_timestep_zero"):
                     index += self.params.ref_index_scale
                     h_offset = 0
                     w_offset = 0
@@ -342,6 +389,13 @@ class Flux(nn.Module):
                 kontext, kontext_ids = self.process_img(ref, index=index, h_offset=h_offset, w_offset=w_offset)
                 img = torch.cat([img, kontext], dim=1)
                 img_ids = torch.cat([img_ids, kontext_ids], dim=1)
+                ref_num_tokens.append(kontext.shape[1])
+            if timestep_zero:
+                if index > 0:
+                    timestep = torch.cat([timestep, timestep * 0], dim=0)
+                    timestep_zero_index = [[img_tokens, img_ids.shape[1]]]
+            transformer_options = transformer_options.copy()
+            transformer_options["reference_image_num_tokens"] = ref_num_tokens
 
         txt_ids = torch.zeros((bs, context.shape[1], len(self.params.axes_dim)), device=x.device, dtype=torch.float32)
 
@@ -349,6 +403,6 @@ class Flux(nn.Module):
             for i in self.params.txt_ids_dims:
                 txt_ids[:, :, i] = torch.linspace(0, context.shape[1] - 1, steps=context.shape[1], device=x.device, dtype=torch.float32)
 
-        out = self.forward_orig(img, img_ids, context, txt_ids, timestep, y, guidance, control, transformer_options, attn_mask=kwargs.get("attention_mask", None))
+        out = self.forward_orig(img, img_ids, context, txt_ids, timestep, y, guidance, control, timestep_zero_index=timestep_zero_index, transformer_options=transformer_options, attn_mask=kwargs.get("attention_mask", None))
         out = out[:, :img_tokens]
         return rearrange(out, "b (h w) (c ph pw) -> b c (h ph) (w pw)", h=h_len, w=w_len, ph=self.patch_size, pw=self.patch_size)[:,:,:h_orig,:w_orig]

comfy_api/input/__init__.py

@@ -5,6 +5,8 @@ from comfy_api.latest._input import (
     MaskInput,
     LatentInput,
     VideoInput,
+    CurveInput,
+    MonotoneCubicCurve,
 )
 
 __all__ = [
@@ -13,4 +15,6 @@ __all__ = [
     "MaskInput",
     "LatentInput",
     "VideoInput",
+    "CurveInput",
+    "MonotoneCubicCurve",
 ]

comfy_api/latest/_input/__init__.py

@@ -1,4 +1,4 @@
-from .basic_types import ImageInput, AudioInput, MaskInput, LatentInput
+from .basic_types import ImageInput, AudioInput, MaskInput, LatentInput, CurveInput, MonotoneCubicCurve
 from .video_types import VideoInput
 
 __all__ = [
@@ -7,4 +7,6 @@ __all__ = [
     "VideoInput",
     "MaskInput",
     "LatentInput",
+    "CurveInput",
+    "MonotoneCubicCurve",
 ]

comfy_api/latest/_input/basic_types.py

@@ -1,3 +1,8 @@
+from __future__ import annotations
+
+import math
+from abc import ABC, abstractmethod
+import numpy as np
 import torch
 from typing import TypedDict, Optional
 
@@ -40,3 +45,153 @@ class LatentInput(TypedDict):
     """
 
     batch_index: Optional[list[int]]
+
+
+CurvePoint = tuple[float, float]
+
+
+class CurveInput(ABC):
+    """Abstract base class for curve inputs.
+
+    Subclasses represent different curve representations (control-point
+    interpolation, analytical functions, LUT-based, etc.) while exposing a
+    uniform evaluation interface to downstream nodes.
+    """
+
+    @property
+    @abstractmethod
+    def points(self) -> list[CurvePoint]:
+        """The control points that define this curve."""
+
+    @abstractmethod
+    def interp(self, x: float) -> float:
+        """Evaluate the curve at a single *x* value in [0, 1]."""
+
+    def interp_array(self, xs: np.ndarray) -> np.ndarray:
+        """Vectorised evaluation over a numpy array of x values.
+
+        Subclasses should override this for better performance. The default
+        falls back to scalar ``interp`` calls.
+        """
+        return np.fromiter((self.interp(float(x)) for x in xs), dtype=np.float64, count=len(xs))
+
+    def to_lut(self, size: int = 256) -> np.ndarray:
+        """Generate a float64 lookup table of *size* evenly-spaced samples in [0, 1]."""
+        return self.interp_array(np.linspace(0.0, 1.0, size))
+
+
+class MonotoneCubicCurve(CurveInput):
+    """Monotone cubic Hermite interpolation over control points.
+
+    Mirrors the frontend ``createMonotoneInterpolator`` in
+    ``ComfyUI_frontend/src/components/curve/curveUtils.ts`` so that
+    backend evaluation matches the editor preview exactly.
+
+    All heavy work (sorting, slope computation) happens once at construction.
+    ``interp_array`` is fully vectorised with numpy.
+    """
+
+    def __init__(self, control_points: list[CurvePoint]):
+        sorted_pts = sorted(control_points, key=lambda p: p[0])
+        self._points = [(float(x), float(y)) for x, y in sorted_pts]
+        self._xs = np.array([p[0] for p in self._points], dtype=np.float64)
+        self._ys = np.array([p[1] for p in self._points], dtype=np.float64)
+        self._slopes = self._compute_slopes()
+
+    @property
+    def points(self) -> list[CurvePoint]:
+        return list(self._points)
+
+    def _compute_slopes(self) -> np.ndarray:
+        xs, ys = self._xs, self._ys
+        n = len(xs)
+        if n < 2:
+            return np.zeros(n, dtype=np.float64)
+
+        dx = np.diff(xs)
+        dy = np.diff(ys)
+        dx_safe = np.where(dx == 0, 1.0, dx)
+        deltas = np.where(dx == 0, 0.0, dy / dx_safe)
+
+        slopes = np.empty(n, dtype=np.float64)
+        slopes[0] = deltas[0]
+        slopes[-1] = deltas[-1]
+        for i in range(1, n - 1):
+            if deltas[i - 1] * deltas[i] <= 0:
+                slopes[i] = 0.0
+            else:
+                slopes[i] = (deltas[i - 1] + deltas[i]) / 2
+
+        for i in range(n - 1):
+            if deltas[i] == 0:
+                slopes[i] = 0.0
+                slopes[i + 1] = 0.0
+            else:
+                alpha = slopes[i] / deltas[i]
+                beta = slopes[i + 1] / deltas[i]
+                s = alpha * alpha + beta * beta
+                if s > 9:
+                    t = 3 / math.sqrt(s)
+                    slopes[i] = t * alpha * deltas[i]
+                    slopes[i + 1] = t * beta * deltas[i]
+        return slopes
+
+    def interp(self, x: float) -> float:
+        xs, ys, slopes = self._xs, self._ys, self._slopes
+        n = len(xs)
+        if n == 0:
+            return 0.0
+        if n == 1:
+            return float(ys[0])
+        if x <= xs[0]:
+            return float(ys[0])
+        if x >= xs[-1]:
+            return float(ys[-1])
+
+        hi = int(np.searchsorted(xs, x, side='right'))
+        hi = min(hi, n - 1)
+        lo = hi - 1
+
+        dx = xs[hi] - xs[lo]
+        if dx == 0:
+            return float(ys[lo])
+
+        t = (x - xs[lo]) / dx
+        t2 = t * t
+        t3 = t2 * t
+        h00 = 2 * t3 - 3 * t2 + 1
+        h10 = t3 - 2 * t2 + t
+        h01 = -2 * t3 + 3 * t2
+        h11 = t3 - t2
+        return float(h00 * ys[lo] + h10 * dx * slopes[lo] + h01 * ys[hi] + h11 * dx * slopes[hi])
+
+    def interp_array(self, xs_in: np.ndarray) -> np.ndarray:
+        """Fully vectorised evaluation using numpy."""
+        xs, ys, slopes = self._xs, self._ys, self._slopes
+        n = len(xs)
+        if n == 0:
+            return np.zeros_like(xs_in)
+        if n == 1:
+            return np.full_like(xs_in, ys[0])
+
+        hi = np.searchsorted(xs, xs_in, side='right').clip(1, n - 1)
+        lo = hi - 1
+
+        dx = xs[hi] - xs[lo]
+        dx_safe = np.where(dx == 0, 1.0, dx)
+        t = np.where(dx == 0, 0.0, (xs_in - xs[lo]) / dx_safe)
+        t2 = t * t
+        t3 = t2 * t
+
+        h00 = 2 * t3 - 3 * t2 + 1
+        h10 = t3 - 2 * t2 + t
+        h01 = -2 * t3 + 3 * t2
+        h11 = t3 - t2
+
+        result = h00 * ys[lo] + h10 * dx * slopes[lo] + h01 * ys[hi] + h11 * dx * slopes[hi]
+        result = np.where(xs_in <= xs[0], ys[0], result)
+        result = np.where(xs_in >= xs[-1], ys[-1], result)
+        return result
+
+    def __repr__(self) -> str:
+        return f"MonotoneCubicCurve(points={self._points})"

comfy_api/latest/_io.py

@@ -23,7 +23,7 @@ if TYPE_CHECKING:
     from comfy.samplers import CFGGuider, Sampler
     from comfy.sd import CLIP, VAE
     from comfy.sd import StyleModel as StyleModel_
-    from comfy_api.input import VideoInput
+    from comfy_api.input import VideoInput, CurveInput as CurveInput_
 from comfy_api.internal import (_ComfyNodeInternal, _NodeOutputInternal, classproperty, copy_class, first_real_override, is_class,
     prune_dict, shallow_clone_class)
 from comfy_execution.graph_utils import ExecutionBlocker
@@ -297,7 +297,7 @@ class Float(ComfyTypeIO):
         '''Float input.'''
         def __init__(self, id: str, display_name: str=None, optional=False, tooltip: str=None, lazy: bool=None,
                     default: float=None, min: float=None, max: float=None, step: float=None, round: float=None,
-                    display_mode: NumberDisplay=None, gradient_stops: list[list[float]]=None,
+                    display_mode: NumberDisplay=None, gradient_stops: list[dict]=None,
                     socketless: bool=None, force_input: bool=None, extra_dict=None, raw_link: bool=None, advanced: bool=None):
             super().__init__(id, display_name, optional, tooltip, lazy, default, socketless, None, force_input, extra_dict, raw_link, advanced)
             self.min = min
@@ -1243,7 +1243,8 @@ class BoundingBox(ComfyTypeIO):
 @comfytype(io_type="CURVE")
 class Curve(ComfyTypeIO):
     CurvePoint = tuple[float, float]
-    Type = list[CurvePoint]
+    if TYPE_CHECKING:
+        Type = CurveInput_
 
     class Input(WidgetInput):
         def __init__(self, id: str, display_name: str=None, optional=False, tooltip: str=None,

comfy_extras/nodes_flux.py

@@ -6,6 +6,7 @@ import comfy.model_management
 import torch
 import math
 import nodes
+import comfy.ldm.flux.math
 
 class CLIPTextEncodeFlux(io.ComfyNode):
     @classmethod
@@ -231,6 +232,68 @@ class Flux2Scheduler(io.ComfyNode):
         sigmas = get_schedule(steps, round(seq_len))
         return io.NodeOutput(sigmas)
 
+class KV_Attn_Input:
+    def __init__(self):
+        self.cache = {}
+
+    def __call__(self, q, k, v, extra_options, **kwargs):
+        reference_image_num_tokens = extra_options.get("reference_image_num_tokens", [])
+        if len(reference_image_num_tokens) == 0:
+            return {}
+
+        ref_toks = sum(reference_image_num_tokens)
+        cache_key = "{}_{}".format(extra_options["block_type"], extra_options["block_index"])
+        if cache_key in self.cache:
+            kk, vv = self.cache[cache_key]
+            self.set_cache = False
+            return {"q": q, "k": torch.cat((k, kk), dim=2), "v": torch.cat((v, vv), dim=2)}
+
+        self.cache[cache_key] = (k[:, :, -ref_toks:], v[:, :, -ref_toks:])
+        self.set_cache = True
+        return {"q": q, "k": k, "v": v}
+
+    def cleanup(self):
+        self.cache = {}
+
+
+class FluxKVCache(io.ComfyNode):
+    @classmethod
+    def define_schema(cls) -> io.Schema:
+        return io.Schema(
+            node_id="FluxKVCache",
+            display_name="Flux KV Cache",
+            description="Enables KV Cache optimization for reference images on Flux family models.",
+            category="",
+            is_experimental=True,
+            inputs=[
+                io.Model.Input("model", tooltip="The model to use KV Cache on."),
+            ],
+            outputs=[
+                io.Model.Output(tooltip="The patched model with KV Cache enabled."),
+            ],
+        )
+
+    @classmethod
+    def execute(cls, model: io.Model.Type) -> io.NodeOutput:
+        m = model.clone()
+        input_patch_obj = KV_Attn_Input()
+
+        def model_input_patch(inputs):
+            if len(input_patch_obj.cache) > 0:
+                ref_image_tokens = sum(inputs["transformer_options"].get("reference_image_num_tokens", []))
+                if ref_image_tokens > 0:
+                    img = inputs["img"]
+                    inputs["img"] = img[:, :-ref_image_tokens]
+            return inputs
+
+        m.set_model_attn1_patch(input_patch_obj)
+        m.set_model_post_input_patch(model_input_patch)
+        if hasattr(model.model.diffusion_model, "params"):
+            m.add_object_patch("diffusion_model.params.default_ref_method", "index_timestep_zero")
+        else:
+            m.add_object_patch("diffusion_model.default_ref_method", "index_timestep_zero")
+
+        return io.NodeOutput(m)
 
 class FluxExtension(ComfyExtension):
     @override
@@ -243,6 +306,7 @@ class FluxExtension(ComfyExtension):
             FluxKontextMultiReferenceLatentMethod,
             EmptyFlux2LatentImage,
             Flux2Scheduler,
+            FluxKVCache,
         ]
 
 

comfy_extras/nodes_painter.py

@@ -0,0 +1,127 @@
+from __future__ import annotations
+
+import hashlib
+import os
+
+import numpy as np
+import torch
+from PIL import Image
+
+import folder_paths
+import node_helpers
+from comfy_api.latest import ComfyExtension, io, UI
+from typing_extensions import override
+
+
+def hex_to_rgb(hex_color: str) -> tuple[float, float, float]:
+    hex_color = hex_color.lstrip("#")
+    if len(hex_color) != 6:
+        return (0.0, 0.0, 0.0)
+    r = int(hex_color[0:2], 16) / 255.0
+    g = int(hex_color[2:4], 16) / 255.0
+    b = int(hex_color[4:6], 16) / 255.0
+    return (r, g, b)
+
+
+class PainterNode(io.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        return io.Schema(
+            node_id="Painter",
+            display_name="Painter",
+            category="image",
+            inputs=[
+                io.Image.Input(
+                    "image",
+                    optional=True,
+                    tooltip="Optional base image to paint over",
+                ),
+                io.String.Input(
+                    "mask",
+                    default="",
+                    socketless=True,
+                    extra_dict={"widgetType": "PAINTER", "image_upload": True},
+                ),
+                io.Int.Input(
+                    "width",
+                    default=512,
+                    min=64,
+                    max=4096,
+                    step=64,
+                    socketless=True,
+                    extra_dict={"hidden": True},
+                ),
+                io.Int.Input(
+                    "height",
+                    default=512,
+                    min=64,
+                    max=4096,
+                    step=64,
+                    socketless=True,
+                    extra_dict={"hidden": True},
+                ),
+                io.Color.Input("bg_color", default="#000000"),
+            ],
+            outputs=[
+                io.Image.Output("IMAGE"),
+                io.Mask.Output("MASK"),
+            ],
+        )
+
+    @classmethod
+    def execute(cls, mask, width, height, bg_color="#000000", image=None) -> io.NodeOutput:
+        if image is not None:
+            base_image = image[:1]
+            h, w = base_image.shape[1], base_image.shape[2]
+        else:
+            h, w = height, width
+            r, g, b = hex_to_rgb(bg_color)
+            base_image = torch.zeros((1, h, w, 3), dtype=torch.float32)
+            base_image[0, :, :, 0] = r
+            base_image[0, :, :, 1] = g
+            base_image[0, :, :, 2] = b
+
+        if mask and mask.strip():
+            mask_path = folder_paths.get_annotated_filepath(mask)
+            painter_img = node_helpers.pillow(Image.open, mask_path)
+            painter_img = painter_img.convert("RGBA")
+
+            if painter_img.size != (w, h):
+                painter_img = painter_img.resize((w, h), Image.LANCZOS)
+
+            painter_np = np.array(painter_img).astype(np.float32) / 255.0
+            painter_rgb = painter_np[:, :, :3]
+            painter_alpha = painter_np[:, :, 3:4]
+
+            mask_tensor = torch.from_numpy(painter_np[:, :, 3]).unsqueeze(0)
+
+            base_np = base_image[0].cpu().numpy()
+            composited = painter_rgb * painter_alpha + base_np * (1.0 - painter_alpha)
+            out_image = torch.from_numpy(composited).unsqueeze(0)
+        else:
+            mask_tensor = torch.zeros((1, h, w), dtype=torch.float32)
+            out_image = base_image
+
+        return io.NodeOutput(out_image, mask_tensor, ui=UI.PreviewImage(out_image))
+
+    @classmethod
+    def fingerprint_inputs(cls, mask, width, height, bg_color="#000000", image=None):
+        if mask and mask.strip():
+            mask_path = folder_paths.get_annotated_filepath(mask)
+            if os.path.exists(mask_path):
+                m = hashlib.sha256()
+                with open(mask_path, "rb") as f:
+                    m.update(f.read())
+                return m.digest().hex()
+        return ""
+
+
+
+class PainterExtension(ComfyExtension):
+    @override
+    async def get_node_list(self):
+        return [PainterNode]
+
+
+async def comfy_entrypoint():
+    return PainterExtension()

nodes.py

@@ -2034,6 +2034,27 @@ class ImagePadForOutpaint:
         return (new_image, mask.unsqueeze(0))
 
 
+class CurveEditor:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+                "curve": ("CURVE", {"default": [[0, 0], [1, 1]]}),
+            }
+        }
+
+    RETURN_TYPES = ("CURVE",)
+    RETURN_NAMES = ("curve",)
+    FUNCTION = "execute"
+    CATEGORY = "utils"
+
+    def execute(self, curve):
+        from comfy_api.input import CurveInput, MonotoneCubicCurve
+        if isinstance(curve, CurveInput):
+            return (curve,)
+        return (MonotoneCubicCurve([(float(x), float(y)) for x, y in curve]),)
+
+
 NODE_CLASS_MAPPINGS = {
     "KSampler": KSampler,
     "CheckpointLoaderSimple": CheckpointLoaderSimple,
@@ -2102,6 +2123,7 @@ NODE_CLASS_MAPPINGS = {
     "ConditioningZeroOut": ConditioningZeroOut,
     "ConditioningSetTimestepRange": ConditioningSetTimestepRange,
     "LoraLoaderModelOnly": LoraLoaderModelOnly,
+    "CurveEditor": CurveEditor,
 }
 
 NODE_DISPLAY_NAME_MAPPINGS = {
@@ -2170,6 +2192,7 @@ NODE_DISPLAY_NAME_MAPPINGS = {
     # _for_testing
     "VAEDecodeTiled": "VAE Decode (Tiled)",
     "VAEEncodeTiled": "VAE Encode (Tiled)",
+    "CurveEditor": "Curve Editor",
 }
 
 EXTENSION_WEB_DIRS = {}
@@ -2450,6 +2473,7 @@ async def init_builtin_extra_nodes():
         "nodes_nag.py",
         "nodes_sdpose.py",
         "nodes_math.py",
+        "nodes_painter.py",
     ]
 
     import_failed = []

requirements.txt

@@ -22,7 +22,7 @@ alembic
 SQLAlchemy
 filelock
 av>=14.2.0
-comfy-kitchen>=0.2.7
+comfy-kitchen>=0.2.8
 comfy-aimdo>=0.2.10
 requests
 simpleeval>=1.0.0