Core Concepts¶
Understanding the key concepts behind Vibe AIGC.
The Intent-Execution Gap¶
Traditional AI content generation suffers from a fundamental problem: the Intent-Execution Gap. Users have complex, multi-dimensional creative visions, but current single-shot models require precise prompt engineering to achieve desired results.
Vibe AIGC addresses this by:
- Accepting high-level intent (Vibes) instead of detailed prompts
- Automatically decomposing intent into executable steps
- Adapting dynamically based on execution feedback
Vibe¶
A Vibe is a high-level representation of creative intent:
from vibe_aigc import Vibe
vibe = Vibe(
description="Main creative goal", # What you want
style="aesthetic preferences", # How it should feel
constraints=["limit1", "limit2"], # Boundaries
domain="content type", # Context
metadata={"key": "value"} # Additional info
)
Unlike prompts, Vibes are:
- Multi-dimensional: Capture style, constraints, and context
- Composable: Can be refined through feedback
- Abstract: Don't specify implementation details
MetaPlanner¶
The MetaPlanner is the central orchestrator that:
- Receives a Vibe from the user
- Uses LLM reasoning to decompose it into tasks
- Generates a hierarchical WorkflowPlan
- Manages execution and adaptation
from vibe_aigc import MetaPlanner
planner = MetaPlanner()
plan = await planner.plan(vibe) # Planning only
result = await planner.execute(vibe) # Plan + execute
WorkflowPlan¶
A WorkflowPlan is the executable decomposition of a Vibe:
from vibe_aigc import WorkflowPlan, WorkflowNode
# Plans contain hierarchical nodes
plan = WorkflowPlan(
id="plan-001",
source_vibe=vibe,
root_nodes=[node1, node2, node3]
)
WorkflowNode¶
WorkflowNodes are individual tasks in the workflow:
from vibe_aigc import WorkflowNode, WorkflowNodeType
node = WorkflowNode(
id="research",
type=WorkflowNodeType.ANALYZE,
description="Research the topic",
dependencies=["setup"], # Must complete first
children=[sub_node1] # Hierarchical decomposition
)
Node Types¶
| Type | Purpose |
|---|---|
ANALYZE |
Information gathering and analysis |
GENERATE |
Content creation |
TRANSFORM |
Modify existing content |
VALIDATE |
Quality checks |
COMPOSITE |
Groups of related tasks |
Execution Flow¶
graph TD
A[User Vibe] --> B[MetaPlanner]
B --> C[WorkflowPlan]
C --> D[WorkflowExecutor]
D --> E{Node Type}
E -->|Independent| F[Parallel Execution]
E -->|Dependent| G[Sequential Execution]
F --> H[Results]
G --> H
H --> I{Success?}
I -->|Yes| J[Complete]
I -->|No| K[Adaptive Replan]
K --> B
Feedback Loops¶
The system supports multiple feedback mechanisms:
- Progress Callbacks: Real-time execution updates
- Adaptive Replanning: Automatic recovery from failures
- Checkpointing: Save/restore execution state
Parallel Execution¶
Independent nodes execute concurrently:
Nodes with dependencies wait:
Next Steps¶
- Workflows - Creating and managing workflows
- Visualization - Visualizing execution
- Checkpoints - Saving and resuming state