2.2 KiB
2.2 KiB
Text-Driven Prompting Workflow Guide
This workflow streamlines prompt management and eliminates manual weight tweaking by utilizing dedicated custom nodes for layered control and real-time validation.
Node Overview
1. Text Line Prompt (Text-Driven)
Manages a list of prompt assets or concepts line by line.
- Bi-directional Sync: Select a line via the dropdown menu to modify its multiplier or active state. The raw text area updates automatically in real-time.
- Muting Function: Deactivating a line prepends
//to comment it out. These lines are completely skipped during backend execution.
2. Multi Prompts (Text-Driven)
Allows managing multiple independent prompt blocks simultaneously within a single node.
- Layered Control: Each prompt block can be independently weighted with a multiplier or toggled active/inactive, streamlining the blending of styles, characters, and environments.
3. Prompt Token Counter (Text-Driven)
An input-only visualization node designed specifically for Stable Diffusion 1.5 and SDXL to estimate the total token count of the incoming combined prompt.
- Real-time Color Signals:
- Green (≤ 65 tokens): Safe zone. The entire prompt fits comfortably within the primary context chunk.
- Yellow (66–75 tokens): Warning zone. Approaching the absolute hardware limit of a single CLIP text encoder chunk.
- Red (≥ 76 tokens): Danger zone. The prompt will be sliced into multiple chunks (sections).
- Chunk Slicing Risk: Tokens forced into the second section or later have a significantly lower probability of being referenced accurately during rendering, which can dilute attention weights, skew character concepts, or lead to image breakdown.
- Approximation Note: Since a strictly exact calculation requires a massive dictionary (~49k words), this node utilizes a highly optimized, high-speed approximation algorithm. The displayed count serves as a close reference estimate.
4. Join Strings (Text-Driven)
Concatenates strings from multiple text-driven nodes using a customizable delimiter (e.g., a comma), serving as the bridge to consolidate your structured prompt layers into a single continuous stream.