Claude
7b02fcfd58
Wildcard options, fixed values and combines can carry a `negative`, and a
derivation action gains the `negative` ("Add to negative") mode. The
engine files negatives per variable in `__wp_negatives__`, carries them
across nodes, and the Assembler gets a `negative` output built from the
variables its template rendered, filled into a `negative_template` at
`$negatives`. Negatives resolve on their own seed stream and refs inside
them resolve quietly, so picks never shift. Injector rows can set a
negative, the Prompt Cleaner cleans an optional negative with its own
rule column, and the preview endpoint returns the negatives table.
Schema v8 stamps payloads that use a negative.
133 lines
6.8 KiB
Python
133 lines
6.8 KiB
Python
"""WP_PromptAssembler — fills $var placeholders from a PipelineContext."""
|
|
|
|
from comfy_api.latest import io # pyright: ignore[reportMissingImports]
|
|
|
|
from engine import negatives
|
|
from engine.context import strip_internals, with_resolver_tables
|
|
from engine.template import resolve_variables, resolve_variables_raw
|
|
from wp_nodes.types import PipelineContext
|
|
|
|
|
|
class WPPromptAssembler(io.ComfyNode):
|
|
"""Template-fills $var placeholders using the incoming PipelineContext."""
|
|
|
|
@classmethod
|
|
def define_schema(cls):
|
|
return io.Schema(
|
|
node_id="WP_PromptAssembler",
|
|
display_name="WP Prompt Assembler",
|
|
category="wildcard-pipeline",
|
|
inputs=[
|
|
PipelineContext.Input(
|
|
"context",
|
|
tooltip=(
|
|
"The resolved $variable context from any upstream "
|
|
"WP Context / Loop / Injector chain. Required — "
|
|
"without it the template can't resolve any $vars."
|
|
),
|
|
),
|
|
io.String.Input(
|
|
"template",
|
|
multiline=True,
|
|
default="",
|
|
# Placeholder example shown only while the textarea
|
|
# is empty — the workflow JSON stores no template
|
|
# by default, so re-creating the node leaves it
|
|
# blank instead of pre-populating prose users
|
|
# immediately have to delete.
|
|
placeholder="A $style portrait of $subject",
|
|
tooltip=(
|
|
"Your prompt template. Type free text and insert "
|
|
"$variable names where the upstream Context's "
|
|
"values should appear. Inline {a|b|c} picks render "
|
|
"verbatim here — produce randomness in a seeded "
|
|
"module instead."
|
|
),
|
|
# Render this input with our own editor instead of the
|
|
# stock textarea, WITHOUT changing what the socket
|
|
# accepts. `widget_type=` would do the first and break
|
|
# the second: `WidgetInput.get_io_type` returns
|
|
# widget_type when set, so the socket would stop being
|
|
# STRING and no upstream STRING output could feed it.
|
|
# `extra_dict` is merged last in `Input.as_dict` and
|
|
# never consulted by `get_io_type`, so the spec carries
|
|
# `widgetType` for the frontend's widget lookup while
|
|
# the socket stays STRING. The value is still a plain
|
|
# string in `widgets_values`, so workflows saved before
|
|
# this change load their template unchanged.
|
|
extra_dict={"widgetType": "WP_TEMPLATE_EDITOR"},
|
|
),
|
|
# Send-to-negative. Declared AFTER `template` so a workflow
|
|
# saved before it existed still maps its one widget value to
|
|
# the template; the missing value takes this default.
|
|
io.String.Input(
|
|
"negative_template",
|
|
multiline=True,
|
|
default="",
|
|
optional=True,
|
|
placeholder="$negatives",
|
|
tooltip=(
|
|
"Your negative prompt. $negatives is where the "
|
|
"negatives of the variables the prompt used go. "
|
|
"Empty = just those words; without $negatives they "
|
|
"are added at the end."
|
|
),
|
|
# Same editor and same STRING socket as `template`.
|
|
extra_dict={"widgetType": "WP_TEMPLATE_EDITOR"},
|
|
),
|
|
],
|
|
# `prompt` stays at index 0 so existing links keep their slot.
|
|
outputs=[
|
|
io.String.Output("prompt"),
|
|
# The negative template filled with the negatives of every
|
|
# variable the prompt rendered.
|
|
io.String.Output("negative"),
|
|
],
|
|
)
|
|
|
|
@classmethod
|
|
def execute(cls, context, template, negative_template=""):
|
|
# Build the render context from the socket payload. `context.context`
|
|
# holds user-named vars including those flagged internal (the
|
|
# PIPELINE_CONTEXT socket now propagates internal vars across
|
|
# nodes so Combine / Derivation downstream of an internal var
|
|
# can still read it). `__wp_internal_flags__` rides in
|
|
# `context.internals` as a cross-node-survivor so this assembler
|
|
# can re-apply the user's "hide from prompt" intent. Merge the
|
|
# flag map back into the render dict, then strip_internals
|
|
# drops both engine `__` keys AND user-flagged internal vars
|
|
# before resolution — net effect: `$var` for an internal var
|
|
# never substitutes in the rendered prompt.
|
|
render_ctx = dict(context.context)
|
|
# The accessor tables (`__wp_axes__`, `__wp_picks__`) AND the
|
|
# internal-flag map ride on `context.internals`, NOT `context.context`:
|
|
# every one is `__`-prefixed, so the socket boundary
|
|
# (`strip_engine_internals` in `build_payload`) drops them from the
|
|
# user-facing payload and re-files the cross-node subset under
|
|
# `internals`. Merge that whole carve-out back in before resolving so
|
|
# `strip_internals` can re-apply the hide-from-prompt filter AND
|
|
# `with_resolver_tables` can re-attach the `$var.AXIS` / `$var.N`
|
|
# tables. Pre-fix only `__wp_internal_flags__` was merged, so
|
|
# `$outfit.SHOES` rendered "" across the socket even though the
|
|
# identical read resolved one node upstream (the combine surface,
|
|
# which runs before the socket strips the table).
|
|
render_ctx.update(context.internals or {})
|
|
resolve_ctx = with_resolver_tables(strip_internals(render_ctx), render_ctx)
|
|
reads: list = []
|
|
resolved = resolve_variables(template, resolve_ctx, reads=reads)
|
|
negative = assemble_negative(
|
|
render_ctx, reads, negative_template or "",
|
|
lambda text: resolve_variables_raw(text, resolve_ctx),
|
|
)
|
|
return io.NodeOutput(resolved, negative)
|
|
|
|
|
|
def assemble_negative(render_ctx, reads, negative_template, resolve_raw) -> str:
|
|
"""The negative output: the negatives of the variables the positive
|
|
template rendered (follow usage), deduped tag by tag, placed into the
|
|
negative template at `$negatives`. Variables in the negative template add
|
|
their text but never their own negatives."""
|
|
entries = negatives.entries_for_reads(render_ctx, reads)
|
|
collected = negatives.join_unique(str(e.get("text", "")) for e in entries)
|
|
return negatives.render_negative(negative_template, collected, resolve_raw)
|