|
"""A set of constants and utilities for handling contexts.
|
|
|
|
Sets up the inputs and outputs for the Context going forward, with additional functions for
|
|
creating and exporting context objects.
|
|
"""
|
|
import comfy.samplers
|
|
import folder_paths
|
|
|
|
_all_context_input_output_data = {
|
|
"base_ctx": ("base_ctx", "RGTHREE_CONTEXT", "CONTEXT"),
|
|
"model": ("model", "MODEL", "MODEL"),
|
|
"clip": ("clip", "CLIP", "CLIP"),
|
|
"vae": ("vae", "VAE", "VAE"),
|
|
"positive": ("positive", "CONDITIONING", "POSITIVE"),
|
|
"negative": ("negative", "CONDITIONING", "NEGATIVE"),
|
|
"latent": ("latent", "LATENT", "LATENT"),
|
|
"images": ("images", "IMAGE", "IMAGE"),
|
|
"seed": ("seed", "INT", "SEED"),
|
|
"steps": ("steps", "INT", "STEPS"),
|
|
"step_refiner": ("step_refiner", "INT", "STEP_REFINER"),
|
|
"cfg": ("cfg", "FLOAT", "CFG"),
|
|
"ckpt_name": ("ckpt_name", folder_paths.get_filename_list("checkpoints"), "CKPT_NAME"),
|
|
"sampler": ("sampler", comfy.samplers.KSampler.SAMPLERS, "SAMPLER"),
|
|
"scheduler": ("scheduler", comfy.samplers.KSampler.SCHEDULERS, "SCHEDULER"),
|
|
"clip_width": ("clip_width", "INT", "CLIP_WIDTH"),
|
|
"clip_height": ("clip_height", "INT", "CLIP_HEIGHT"),
|
|
"text_pos_g": ("text_pos_g", "STRING", "TEXT_POS_G"),
|
|
"text_pos_l": ("text_pos_l", "STRING", "TEXT_POS_L"),
|
|
"text_neg_g": ("text_neg_g", "STRING", "TEXT_NEG_G"),
|
|
"text_neg_l": ("text_neg_l", "STRING", "TEXT_NEG_L"),
|
|
"mask": ("mask", "MASK", "MASK"),
|
|
"control_net": ("control_net", "CONTROL_NET", "CONTROL_NET"),
|
|
}
|
|
|
|
force_input_types = ["INT", "STRING", "FLOAT"]
|
|
force_input_names = ["sampler", "scheduler", "ckpt_name"]
|
|
|
|
|
|
def _create_context_data(input_list=None):
|
|
"""Returns a tuple of context inputs, return types, and return names to use in a node"s def"""
|
|
if input_list is None:
|
|
input_list = _all_context_input_output_data.keys()
|
|
list_ctx_return_types = []
|
|
list_ctx_return_names = []
|
|
ctx_optional_inputs = {}
|
|
for inp in input_list:
|
|
data = _all_context_input_output_data[inp]
|
|
list_ctx_return_types.append(data[1])
|
|
list_ctx_return_names.append(data[2])
|
|
ctx_optional_inputs[data[0]] = tuple([data[1]] + ([{
|
|
"forceInput": True
|
|
}] if data[1] in force_input_types or data[0] in force_input_names else []))
|
|
|
|
ctx_return_types = tuple(list_ctx_return_types)
|
|
ctx_return_names = tuple(list_ctx_return_names)
|
|
return (ctx_optional_inputs, ctx_return_types, ctx_return_names)
|
|
|
|
|
|
ALL_CTX_OPTIONAL_INPUTS, ALL_CTX_RETURN_TYPES, ALL_CTX_RETURN_NAMES = _create_context_data()
|
|
|
|
_original_ctx_inputs_list = [
|
|
"base_ctx", "model", "clip", "vae", "positive", "negative", "latent", "images", "seed"
|
|
]
|
|
ORIG_CTX_OPTIONAL_INPUTS, ORIG_CTX_RETURN_TYPES, ORIG_CTX_RETURN_NAMES = _create_context_data(
|
|
_original_ctx_inputs_list)
|
|
|
|
|
|
def new_context(base_ctx, **kwargs):
|
|
"""Creates a new context from the provided data, with an optional base ctx to start."""
|
|
context = base_ctx if base_ctx is not None else None
|
|
new_ctx = {}
|
|
for key in _all_context_input_output_data:
|
|
if key == "base_ctx":
|
|
continue
|
|
v = kwargs[key] if key in kwargs else None
|
|
new_ctx[key] = v if v is not None else context[
|
|
key] if context is not None and key in context else None
|
|
return new_ctx
|
|
|
|
|
|
def merge_new_context(*args):
|
|
"""Creates a new context by merging provided contexts with the latter overriding same fields."""
|
|
new_ctx = {}
|
|
for key in _all_context_input_output_data:
|
|
if key == "base_ctx":
|
|
continue
|
|
v = None
|
|
|
|
for ctx in reversed(args):
|
|
v = ctx[key] if not is_context_empty(ctx) and key in ctx else None
|
|
if v is not None:
|
|
break
|
|
new_ctx[key] = v
|
|
return new_ctx
|
|
|
|
|
|
def get_context_return_tuple(ctx, inputs_list=None):
|
|
"""Returns a tuple for returning in the order of the inputs list."""
|
|
if inputs_list is None:
|
|
inputs_list = _all_context_input_output_data.keys()
|
|
tup_list = [
|
|
ctx,
|
|
]
|
|
for key in inputs_list:
|
|
if key == "base_ctx":
|
|
continue
|
|
tup_list.append(ctx[key] if ctx is not None and key in ctx else None)
|
|
return tuple(tup_list)
|
|
|
|
|
|
def get_orig_context_return_tuple(ctx):
|
|
"""Returns a tuple for returning from a node with only the original context keys."""
|
|
return get_context_return_tuple(ctx, _original_ctx_inputs_list)
|
|
|
|
|
|
def is_context_empty(ctx):
|
|
"""Checks if the provided ctx is None or contains just None values."""
|
|
return not ctx or all(v is None for v in ctx.values())
|
|
|