digital-garbageandClaude Opus 5
090669e950
v4's Chain Sampler became a hub: 8183 lines, roughly seventy optional inputs, a _sample_chunk that grew six ALG arguments, and a hardcoded list of knobs silently inert on one model family. Every technique that needed the loop was added to it, so it collected all of them. This one picks an algorithm and a schedule from ComfyUI's own registries -- so a sampler it gains appears here without an edit -- and runs them. Three things stop it becoming a hub again. It never learns what a model-side modifier is. Sampler-side ones reach it through ONE call site, chain.process(ctx, latent), which cannot grow because the step travels in a context object rather than in the signature. And it declares the hook points it offers, so core hands it only what fits and it never names a modifier. Some modifiers genuinely cannot ride on a model: they need the step index, the schedule, or a second pass. Those attach here instead, and one asking for a hook this sampler does not offer is absent and said so, not half-running. Verified against ComfyUI's own guider path with real weights: bit-identical output. That took correcting the method, not the code -- see below. MEASUREMENT, corrected: sampling on this Mac's MPS backend is NONDETERMINISTIC. The same node with the same seed and the same inputs differs run to run: 2.42 by max latent difference, 0.0008 by mean pixel gap. Comparing by MAX said this sampler diverged from core by 14.2 and it was about to be "fixed"; by MEAN against a measured noise floor it is 0.00000 while core against itself is 0.00183. Every comparison here now runs a control twice first. Re-checked the Detail modifier the same way: 145x the floor, so that claim stands. Also matched what core's sampler actually does and this did not: fix_empty_latent_channels using the ratios our own empty-latent node reports, the preview and progress callback, the move back to the intermediate device, and dropping the ratio keys so nothing rescales twice. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
99 lines
4.6 KiB
Python
99 lines
4.6 KiB
Python
"""What a module declares, and the vocabulary it may use.
|
|
|
|
Core holds no list of features. A module announces itself and this file says
|
|
what a valid announcement looks like -- nothing here names an implementation.
|
|
"""
|
|
|
|
from dataclasses import dataclass, field
|
|
from typing import Any, Callable, Dict, List, Optional
|
|
|
|
# Bumped when the shape of an announcement changes. A module declaring an older
|
|
# version is not quietly adapted: it is refused, because guessing at what an old
|
|
# declaration meant is how silent behaviour changes happen.
|
|
CONTRACT_VERSION = 1
|
|
|
|
# Closed set. A type outside it has no renderer and no validation, so accepting
|
|
# one would mean rendering something approximate.
|
|
# A setting is a PREFERENCE. Anything that is data -- an image, a drawn mask,
|
|
# a video -- travels as a ComfyUI socket instead, because it is a tensor the app
|
|
# uploads and the graph wires, not a value that belongs in a JSON payload. That
|
|
# is why there is no "image" or "mask" here despite both being planned features.
|
|
TYPES = frozenset({"bool", "int", "float", "enum", "text", "multiline", "path", "color"})
|
|
|
|
NUMERIC = frozenset({"int", "float"})
|
|
|
|
# Which renderers each type allows, first being the default. A hint outside its
|
|
# type's list is refused at import: the alternative is a panel that silently
|
|
# renders something other than what was asked for.
|
|
UI_HINTS: Dict[str, List[str]] = {
|
|
"bool": ["checkboxRow", "toggle"],
|
|
"int": ["number", "slider", "stepper", "macroSlider"],
|
|
"float": ["number", "slider", "stepper", "macroSlider"],
|
|
"enum": ["select", "segmented", "radioGroup", "filterList", "wheel"],
|
|
"text": ["input", "search"],
|
|
"multiline": ["textarea", "autoTextarea"],
|
|
"path": ["filterList"],
|
|
"color": ["swatch"],
|
|
}
|
|
|
|
# Ordering is coarse-grained by stage, then by declared relations within it.
|
|
# Numbers are deliberately absent: a priority number is a global namespace, so
|
|
# inserting one module means renumbering the others.
|
|
STAGES: List[str] = ["load", "conditioning", "latent", "guidance", "sampling", "post"]
|
|
|
|
|
|
@dataclass(frozen=True)
|
|
class ModuleSpec:
|
|
"""One module's announcement, after validation."""
|
|
|
|
id: str
|
|
title: str
|
|
mount: str
|
|
settings: Dict[str, dict] = field(default_factory=dict)
|
|
requires: List[str] = field(default_factory=list) # model traits
|
|
after: List[str] = field(default_factory=list) # module ids
|
|
before: List[str] = field(default_factory=list)
|
|
stage: str = "sampling"
|
|
# ComfyUI node classes this module contributes. Kept out of `settings` on
|
|
# purpose: a setting is a preference, a node is graph structure, and letting
|
|
# a new setting change a node's socket list would rot every saved workflow.
|
|
nodes: List[type] = field(default_factory=list)
|
|
# A callable(model) -> iterable[str], contributing traits core cannot read on
|
|
# its own. This is how a model's own module teaches the system to recognise
|
|
# it, so supporting a new model is a new folder and never an edit to core.
|
|
traits: Optional[Callable] = None
|
|
# Named capabilities this module offers to OTHER modules. Core never reads a
|
|
# name out of here and never defines one: it is a lookup, so a node can ask
|
|
# "who can build one of these" without core learning what the thing is.
|
|
provides: Dict[str, Callable] = field(default_factory=dict)
|
|
# Hook points a sampler must offer before this module can run. Empty means
|
|
# "any sampler will do", the same narrowing rule traits follow.
|
|
hooks: List[str] = field(default_factory=list)
|
|
ui: Optional[str] = None # served path to its ui.js
|
|
status: str = "experimental" # or "proven"
|
|
source: str = "" # dotted import path
|
|
|
|
def defaults(self) -> Dict[str, Any]:
|
|
"""The values a headless run gets when no panel has been rendered.
|
|
|
|
Derived from the same declaration the panel renders, so the two cannot
|
|
disagree about what a setting means when nobody has touched it.
|
|
"""
|
|
return {key: spec["default"] for key, spec in self.settings.items()}
|
|
|
|
def to_manifest(self) -> dict:
|
|
# `nodes` is deliberately absent: it is not JSON, and the browser has no
|
|
# use for it. The graph is ComfyUI's surface, the manifest is the app's.
|
|
return {
|
|
"id": self.id,
|
|
"title": self.title,
|
|
"mount": self.mount,
|
|
"settings": self.settings,
|
|
"requires": list(self.requires),
|
|
"after": list(self.after),
|
|
"before": list(self.before),
|
|
"stage": self.stage,
|
|
"ui": self.ui,
|
|
"status": self.status,
|
|
}
|