224 lines
9.0 KiB
Python
224 lines
9.0 KiB
Python
from inspect import cleandoc
|
|
|
|
from comfy_api.latest import io
|
|
|
|
import torch
|
|
|
|
from .helper_functions import getIndexTensorAlongDim, comonLazy, parse_expr, eval_tensor_expr_with_tree, make_zero_like
|
|
|
|
from .MathNodeBase import MathNodeBase
|
|
|
|
# try to import NestedTensor type if available
|
|
try:
|
|
import comfy.nested_tensor as _nested_tensor_module
|
|
_NESTED_TENSOR_AVAILABLE = True
|
|
except Exception:
|
|
_nested_tensor_module = None
|
|
_NESTED_TENSOR_AVAILABLE = False
|
|
|
|
|
|
class LatentMathNode(MathNodeBase):
|
|
"""
|
|
This node enables the use of math expressions on Latents.
|
|
inputs:
|
|
a, b, c, d:
|
|
Latent, bound to variables with the same name. Defaults to zero latent if not provided.
|
|
w, x, y, z:
|
|
Floats, bound to variables of the expression. Defaults to 0.0 if not provided.
|
|
Latent expression:
|
|
String, describing expression to aply to latents.
|
|
|
|
outputs:
|
|
LATENT:
|
|
Returns a LATENT object that contains the result of the math expression applied to the input conditionings.
|
|
"""
|
|
def __init__(self):
|
|
pass
|
|
|
|
@classmethod
|
|
def define_schema(cls) -> io.Schema:
|
|
"""
|
|
"""
|
|
return io.Schema(
|
|
node_id="mrmth_LatentMathNode",
|
|
display_name="Latent math",
|
|
category="More math",
|
|
inputs=[
|
|
io.Latent.Input(id="a"),
|
|
io.Latent.Input(id="b", optional=True, lazy=True),
|
|
io.Latent.Input(id="c", optional=True, lazy=True),
|
|
io.Latent.Input(id="d", optional=True, lazy=True),
|
|
io.Float.Input(id="w", default=0.0,optional=True,lazy=True, force_input=True),
|
|
io.Float.Input(id="x", default=0.0,optional=True,lazy=True, force_input=True),
|
|
io.Float.Input(id="y", default=0.0,optional=True,lazy=True, force_input=True),
|
|
io.Float.Input(id="z", default=0.0,optional=True,lazy=True, force_input=True),
|
|
io.String.Input(id="Latent", default="a*(1-w)+b*w", tooltip="Expression to apply on input latents"),
|
|
],
|
|
outputs=[
|
|
io.Latent.Output(),
|
|
],
|
|
)
|
|
|
|
#RETURN_NAMES = ("image_output_name",)
|
|
tooltip = cleandoc(__doc__)
|
|
|
|
#OUTPUT_NODE = False
|
|
#OUTPUT_TOOLTIPS = ("",) # Tooltips for the output node
|
|
@classmethod
|
|
def execute(cls, Latent, a, b=None, c=None, d=None, w=0.0, x=0.0, y=0.0, z=0.0) -> io.NodeOutput:
|
|
# Extract raw sample tensors (may be Tensor or NestedTensor)
|
|
a_in = a["samples"]
|
|
b_in = None if b is None else b["samples"]
|
|
c_in = None if c is None else c["samples"]
|
|
d_in = None if d is None else d["samples"]
|
|
|
|
# parse expression once
|
|
tree = parse_expr(Latent)
|
|
|
|
# Helper to evaluate for a single tensor
|
|
def eval_single_tensor(a_t, b_t, c_t, d_t):
|
|
# support tensors with >=4 dims (e.g. 4D latents [B,C,H,W] or 5D [B,T,C,H,W])
|
|
ndim = a_t.ndim
|
|
# use negative indexing so that channel/height/width mapping works for 4D and 5D
|
|
batch_dim = 0
|
|
channel_dim = -3
|
|
height_dim = -2
|
|
width_dim = -1
|
|
time_dim = None
|
|
if ndim >= 5:
|
|
# time/frame dim is the one before channels when present
|
|
time_dim = -4
|
|
|
|
B = getIndexTensorAlongDim(a_t, batch_dim)
|
|
C = getIndexTensorAlongDim(a_t, channel_dim)
|
|
H = getIndexTensorAlongDim(a_t, height_dim)
|
|
W = getIndexTensorAlongDim(a_t, width_dim)
|
|
|
|
# fill scalar/value tensors
|
|
width_val = a_t.shape[width_dim]
|
|
height_val = a_t.shape[height_dim]
|
|
channel_count = a_t.shape[channel_dim]
|
|
batch_count = a_t.shape[batch_dim]
|
|
frame_count = a_t.shape[time_dim] if time_dim is not None else a_t.shape[batch_dim]
|
|
|
|
variables = {
|
|
# core tensors and floats
|
|
'a': a_t, 'b': b_t, 'c': c_t, 'd': d_t,
|
|
'w': w, 'x': x, 'y': y, 'z': z,
|
|
|
|
'X': W, 'Y': H,
|
|
'B': B, 'batch': B,
|
|
'C': C, 'channel': C,
|
|
# scalar dims and counts
|
|
'W': width_val, 'width': width_val,
|
|
'H': height_val, 'height': height_val,
|
|
'T': frame_count, 'batch_count': batch_count,
|
|
'N': channel_count, 'channel_count': channel_count,
|
|
}
|
|
|
|
# expose time/frame if present
|
|
if time_dim is not None:
|
|
F = getIndexTensorAlongDim(a_t, time_dim)
|
|
variables.update({'frame_idx': F, 'frame': F, 'frame_count': frame_count})
|
|
|
|
return eval_tensor_expr_with_tree(tree, variables, a_t.shape)
|
|
|
|
# If input is a NestedTensor (from comfy), evaluate per-subtensor and return NestedTensor result
|
|
if hasattr(a_in, 'is_nested') and getattr(a_in, 'is_nested'):
|
|
# get underlying lists
|
|
a_list = a_in.unbind()
|
|
sizes = [t.shape[0] for t in a_list]
|
|
# merge all a subtensors along batch (dim=0)
|
|
merged_a = torch.cat(a_list, dim=0)
|
|
|
|
def merge_to_tensor(val, ref):
|
|
# ref is merged_a
|
|
if val is None:
|
|
return make_zero_like(ref)
|
|
if hasattr(val, 'is_nested') and getattr(val, 'is_nested'):
|
|
lst = val.unbind()
|
|
return torch.cat(lst, dim=0)
|
|
if isinstance(val, (list, tuple)):
|
|
return torch.cat(list(val), dim=0)
|
|
if torch.is_tensor(val):
|
|
# if val already matches merged shape
|
|
if val.shape == ref.shape:
|
|
return val
|
|
# if val is per-subtensor with same per-subtensor batch, replicate
|
|
try:
|
|
if val.shape[0] in sizes and val.shape[1:] == a_list[0].shape[1:]:
|
|
# broadcast by concatenating copies
|
|
return torch.cat([val for _ in a_list], dim=0)
|
|
except Exception:
|
|
pass
|
|
# if val has batch equal to combined, return as is
|
|
if val.shape[0] == sum(sizes):
|
|
return val
|
|
# fallback
|
|
return make_zero_like(ref)
|
|
|
|
merged_b = merge_to_tensor(b_in, merged_a)
|
|
merged_c = merge_to_tensor(c_in, merged_a)
|
|
merged_d = merge_to_tensor(d_in, merged_a)
|
|
|
|
# evaluate once on merged tensors
|
|
merged_result = eval_single_tensor(merged_a, merged_b, merged_c, merged_d)
|
|
|
|
# split back into list
|
|
split_results = list(merged_result.split(sizes, dim=0))
|
|
if _NESTED_TENSOR_AVAILABLE and _nested_tensor_module is not None:
|
|
out_samples = _nested_tensor_module.NestedTensor(split_results)
|
|
else:
|
|
out_samples = split_results
|
|
return ({"samples": out_samples},)
|
|
|
|
# Non-nested (single tensor) path
|
|
# ensure b/c/d are set appropriately (zeros_like if None)
|
|
def to_tensor(val, ref):
|
|
if val is None:
|
|
return make_zero_like(ref)
|
|
if hasattr(val, 'is_nested') and getattr(val, 'is_nested'):
|
|
lst = val.unbind()
|
|
elif isinstance(val, (list, tuple)):
|
|
lst = list(val)
|
|
else:
|
|
return val
|
|
|
|
if len(lst) == 0:
|
|
return make_zero_like(ref)
|
|
if len(lst) == 1:
|
|
return lst[0]
|
|
try:
|
|
cat = torch.cat(lst, dim=0)
|
|
if cat.shape == ref.shape or (cat.shape[0] == ref.shape[0] and cat.shape[1:] == ref.shape[1:]):
|
|
return cat
|
|
except Exception:
|
|
pass
|
|
try:
|
|
stk = torch.stack(lst, dim=0)
|
|
if stk.shape == ref.shape or (stk.shape[0] == ref.shape[0] and stk.shape[1:] == ref.shape[1:]):
|
|
return stk
|
|
except Exception:
|
|
pass
|
|
return lst[0]
|
|
|
|
b_single = to_tensor(b_in, a_in)
|
|
c_single = to_tensor(c_in, a_in)
|
|
d_single = to_tensor(d_in, a_in)
|
|
|
|
result1 = eval_single_tensor(a_in, b_single, c_single, d_single)
|
|
result = {"samples": result1}
|
|
return (result,)
|
|
|
|
"""
|
|
The node will always be re executed if any of the inputs change but
|
|
this method can be used to force the node to execute again even when the inputs don't change.
|
|
You can make this node return a number or a string. This value will be compared to the one returned the last time the node was
|
|
executed, if it is different the node will be executed again.
|
|
This method is used in the core repo for the LoadImage node where they return the image hash as a string, if the image hash
|
|
changes between executions the LoadImage node is executed again.
|
|
"""
|
|
#@classmethod
|
|
#def IS_CHANGED(s, image, string_field, int_field, float_field, print_to_screen):
|
|
# return ""
|