astrbot_plugin_imagestudio/backend/services/character_workflow_service.py

343 行
15 KiB
Python

"""Deterministic execution helpers for character workflows."""
from __future__ import annotations
from typing import Any
from backend.dto.common import ApiError, ErrorCode
from backend.dto.profile_dto import CharacterProfile
from backend.dto.workflow_dto import ProfileWorkflow, WorkflowNode
from backend.services.workflow_service import WorkflowService
from backend.services.workflow_progression_service import (
CHARACTER_GRADUAL_PROGRESSION_VERSION,
character_execution_metadata,
character_gradual_progression,
)
from core.identity import RequestContext
from engines.character_flow import canonical_gradual_node_key
CHARACTER_WORKFLOW_TYPES = {"character_fast", "character_gradual", "quick_profile"}
class CharacterWorkflowService:
def __init__(self, workflow_service: WorkflowService):
self.workflow_service = workflow_service
def execute_workflow(self, workflow_id: str, ctx: RequestContext) -> dict[str, Any]:
detail = self.workflow_service.get_workflow(workflow_id)
workflow = detail.workflow
if workflow.workflow_type not in CHARACTER_WORKFLOW_TYPES:
raise ApiError(
ErrorCode.VALIDATION_FAILED,
"workflow is not a character workflow",
{"workflow_id": workflow_id, "workflow_type": workflow.workflow_type},
)
profile = self.workflow_service.profile_repository.get(workflow.profile_id)
if profile is None:
raise ApiError(ErrorCode.RESOURCE_NOT_FOUND, "profile not found", {"profile_id": workflow.profile_id})
active_nodes = [
node
for node in sorted(detail.nodes, key=lambda item: item.sort_order)
if node.branch_name == workflow.current_branch
]
if not active_nodes:
raise ApiError(ErrorCode.VALIDATION_FAILED, "workflow has no executable nodes", {"workflow_id": workflow_id})
if workflow.status == "completed":
return {
"workflow": workflow.to_dict(),
"nodes": [node.to_dict() for node in active_nodes],
"completed_node_ids": [
node.node_id for node in active_nodes if node.status in {"completed", "selected"}
],
"queued_node_ids": [node.node_id for node in active_nodes if node.status == "queued"],
"prepared_node_ids": [],
"idempotent_replay": True,
}
if workflow.workflow_type != "character_gradual":
return self._execute_unsequenced(workflow, active_nodes, profile, ctx)
previous_execution = workflow.metadata.get("character_execution")
if isinstance(previous_execution, dict):
# execute-character is the confirmation of the synthetic import
# entry. Replaying it must not prepare or unlock another node.
progression = character_gradual_progression(
workflow,
active_nodes,
branch_name=workflow.current_branch,
)
if progression is not None and progression.import_confirmed:
return _execution_result(
workflow,
active_nodes,
progression=progression,
idempotent_replay=True,
)
# Persist the import confirmation before touching profile_brief. The
# central progression guard then makes that first backend node (and
# only that node) writable.
initial_execution = {
**(previous_execution if isinstance(previous_execution, dict) else {}),
"executor": "character_workflow",
"workflow_type": workflow.workflow_type,
"import_confirmed": True,
"progression_version": CHARACTER_GRADUAL_PROGRESSION_VERSION,
}
running_workflow = self.workflow_service.update_workflow(
workflow_id,
{
"status": "running" if workflow.status != "running" else workflow.status,
"current_node_id": active_nodes[0].node_id,
"metadata": {
**workflow.metadata,
"character_execution": initial_execution,
},
},
ctx,
allow_character_execution_update=True,
)
progression = character_gradual_progression(
running_workflow,
active_nodes,
branch_name=running_workflow.current_branch,
)
if progression is None:
raise ApiError(
ErrorCode.VALIDATION_FAILED,
"workflow does not support gradual progression",
{"workflow_id": workflow_id, "workflow_type": workflow.workflow_type},
)
prepared_nodes = list(active_nodes)
current = progression.current_node
if current is not None:
prompt_context = [
node.prompt or ""
for node in active_nodes
if node.node_id in progression.completed_node_ids and node.prompt
]
prepared = self.workflow_service.update_node(
current.node_id,
{
"status": current.status,
"prompt": _node_prompt(profile, current, running_workflow, prompt_context),
"negative_prompt": _negative_prompt(profile, current),
"metadata": {
**current.metadata,
"execution": {
"executor": "character_workflow",
"workflow_type": workflow.workflow_type,
"context_prompts": len(prompt_context),
"prepared_only": current.status == "draft",
"progression_version": CHARACTER_GRADUAL_PROGRESSION_VERSION,
},
},
},
ctx,
)
prepared_nodes = [prepared if node.node_id == prepared.node_id else node for node in active_nodes]
progression = character_gradual_progression(
running_workflow,
prepared_nodes,
branch_name=running_workflow.current_branch,
) or progression
current_node_id = progression.current_node.node_id if progression.current_node else prepared_nodes[-1].node_id
execution = character_execution_metadata(
running_workflow,
progression,
existing=initial_execution,
)
execution["queued_node_ids"] = [node.node_id for node in prepared_nodes if node.status == "queued"]
execution["prepared_node_ids"] = [current_node_id] if progression.current_node is not None else []
updated_workflow = self.workflow_service.update_workflow(
workflow_id,
{
"status": "completed" if progression.all_completed else "running",
"current_node_id": current_node_id,
"metadata": {
**running_workflow.metadata,
"character_execution": execution,
},
},
ctx,
allow_character_execution_update=True,
)
return _execution_result(updated_workflow, prepared_nodes, progression=progression)
def _execute_unsequenced(
self,
workflow: ProfileWorkflow,
active_nodes: list[WorkflowNode],
profile: CharacterProfile,
ctx: RequestContext,
) -> dict[str, Any]:
previous_execution = workflow.metadata.get("character_execution")
if workflow.status == "running" and isinstance(previous_execution, dict):
return {
"workflow": workflow.to_dict(),
"nodes": [node.to_dict() for node in active_nodes],
"completed_node_ids": [
node.node_id for node in active_nodes if node.status in {"completed", "selected"}
],
"queued_node_ids": [node.node_id for node in active_nodes if node.status == "queued"],
"prepared_node_ids": [node.node_id for node in active_nodes if node.status == "draft"],
"idempotent_replay": True,
}
running_workflow = workflow
if workflow.status != "running":
running_workflow = self.workflow_service.update_workflow(
workflow.workflow_id,
{"status": "running"},
ctx,
)
prepared_nodes: list[WorkflowNode] = []
prompt_context: list[str] = []
for node in active_nodes:
prepared = self.workflow_service.update_node(
node.node_id,
{
"status": node.status,
"prompt": _node_prompt(profile, node, running_workflow, prompt_context),
"negative_prompt": _negative_prompt(profile, node),
"metadata": {
**node.metadata,
"execution": {
"executor": "character_workflow",
"workflow_type": workflow.workflow_type,
"context_prompts": len(prompt_context),
"prepared_only": node.status == "draft",
},
},
},
ctx,
)
prepared_nodes.append(prepared)
if prepared.prompt:
prompt_context.append(prepared.prompt)
queued_nodes = [node for node in prepared_nodes if node.status == "queued"]
completed_nodes = [node for node in prepared_nodes if node.status in {"completed", "selected"}]
editable_nodes = [node for node in prepared_nodes if node.status == "draft"]
next_node = next(
(node for node in prepared_nodes if node.status not in {"completed", "selected", "skipped"}),
prepared_nodes[-1],
)
updated_workflow = self.workflow_service.update_workflow(
workflow.workflow_id,
{
"status": "running" if editable_nodes or queued_nodes else "completed",
"current_node_id": next_node.node_id,
"metadata": {
**running_workflow.metadata,
"character_execution": {
"executor": "character_workflow",
"workflow_type": workflow.workflow_type,
"completed_node_ids": [node.node_id for node in completed_nodes],
"queued_node_ids": [node.node_id for node in queued_nodes],
"prepared_node_ids": [node.node_id for node in editable_nodes],
},
},
},
ctx,
)
return {
"workflow": updated_workflow.to_dict(),
"nodes": [node.to_dict() for node in prepared_nodes],
"completed_node_ids": [node.node_id for node in completed_nodes],
"queued_node_ids": [node.node_id for node in queued_nodes],
"prepared_node_ids": [node.node_id for node in editable_nodes],
}
def _node_prompt(
profile: CharacterProfile,
node: WorkflowNode,
workflow: ProfileWorkflow,
prompt_context: list[str],
) -> str:
parts = [
profile.name,
profile.base_prompt,
profile.appearance_prompt,
profile.style_prompt,
_step_instruction(workflow.workflow_type, node.node_key),
node.prompt,
]
if prompt_context:
parts.append(f"consistency reference: {prompt_context[-1]}")
return ", ".join(_unique_text(parts))
def _negative_prompt(profile: CharacterProfile, node: WorkflowNode) -> str:
return ", ".join(_unique_text([profile.negative_prompt, node.negative_prompt, "low quality, blurry"]))
def _step_instruction(workflow_type: str, node_key: str) -> str:
quick_profile = {
"questionnaire": "normalized questionnaire and quick profile archive input",
"preview": "saved quick profile prompts ready for structured setting split",
}
fast = {
"quick_avatar": "square avatar, face, hair, eyes, expression, keep identity consistent",
"quick_bust": "half-body character reference, clothing, posture, keep identity consistent",
"quick_stand": "full-body standing character illustration, complete outfit and proportions",
"quick_sheet": "character design sheet, multiple views, outfit details and material callouts",
}
gradual = {
"profile_brief": "core identity and visual anchors",
"prompt_plan": "prompt matrix and inheritance rules for every image stage",
"material_bind": "reference material binding rules for style, composition, clothing, shoes, and accessories",
"headshot": "headshot, face, hair, eyes, expression",
"upper_body": "upper body, outfit, posture, silhouette, preserve the confirmed face identity",
"lower_body": "lower body, waist, hips, legs, footwear and consistent body proportions",
"full_body": "full body, complete outfit and proportions",
"turnaround": "front, side and back turnaround views with identical identity and costume structure",
"design_sheet": "design sheet with expressions, poses, hands, accessories and reusable callouts",
"style_lock": "high-resolution restoration, identity consistency and final rendering rules",
"profile_card": "validate accepted assets and seal the immutable character profile card",
}
tables = {
"character_fast": fast,
"character_gradual": gradual,
"quick_profile": quick_profile,
}
table = tables.get(workflow_type, {})
resolved_key = canonical_gradual_node_key(node_key) if workflow_type == "character_gradual" else node_key
return table.get(resolved_key, resolved_key.replace("_", " "))
def _unique_text(values: list[Any]) -> list[str]:
result: list[str] = []
seen: set[str] = set()
for value in values:
text = str(value or "").strip()
if text and text not in seen:
result.append(text)
seen.add(text)
return result
def _execution_result(
workflow: ProfileWorkflow,
nodes: list[WorkflowNode],
*,
progression,
idempotent_replay: bool = False,
) -> dict[str, Any]:
current_node_id = progression.current_node.node_id if progression.current_node else None
result = {
"workflow": workflow.to_dict(),
"nodes": [node.to_dict() for node in nodes],
"completed_node_ids": list(progression.completed_node_ids),
"queued_node_ids": [node.node_id for node in nodes if node.status == "queued"],
"prepared_node_ids": [current_node_id] if current_node_id else [],
"locked_node_ids": [
node.node_id
for node in nodes
if not progression.is_accessible(node)
],
"progression": progression.to_dict(),
}
if idempotent_replay:
result["idempotent_replay"] = True
return result