google-labs-jules[bot]
956e6846ef
The previous implementation of normalization in `_apply_direct_scaling` applied normalization per-token immediately after scaling. This effectively cancelled out the applied weight, rendering the emphasis feature non-functional when normalization was enabled. This change fixes the logic to: 1. Calculate the global mean of the conditioning tensor before scaling. 2. Apply weights to all tokens. 3. Normalize the entire tensor to restore the original global mean. This correctly implements A1111-style 'EmphasisOriginal' behavior where relative weights are preserved while maintaining global magnitude.
A1111 Prompt Node (God Node)
A custom ComfyUI node that implements A1111-style prompt handling with proper isolation and emphasis math.
Installation
Requires lark for full nested syntax support:
pip install lark
✅ Working Features
Core Features
- Hard Chunking (The Sandbox): Tokens split into 75-token chunks with padding, preventing concept bleeding
- Direct Scaling (Anti-Burn): Uses
z * weightinstead of Comfy's interpolation, avoiding artifacts at high weights - BREAK Support: Fully isolated context windows
- Emphasis:
(text:1.2),(text),[text]
Scheduling (Fully Recursive)
| Syntax | Meaning |
|---|---|
[cat:dog:0.5] |
Switch from "cat" to "dog" at 50% |
[cat:dog:10] |
Switch at step 10 (requires steps input) |
[glasses:0.5] |
Add "glasses" at 50% |
[glasses:10] |
Add at step 10 |
[hat::0.7] |
Remove "hat" at 70% |
[hat::15] |
Remove at step 15 |
Nested syntax supported:
[honovy, exsys:chen bin, [as109|fkey], [sweetonedollar:11]:0.4]
Alternation
[white|black]- Blends concepts (50/50 average)[A|B|C]- Equal blend of 3 options (33% each)
Model Support
- SDXL: Dual CLIP (clip_l + clip_g) with independent weight scaling
- SD1.5: Full support
Inputs
| Input | Type | Description |
|---|---|---|
| clip | CLIP | The CLIP model |
| text | STRING | Prompt with A1111 syntax |
| steps | INT | Total sampling steps (for step-based scheduling) |
| normalization | BOOL | Enable EmphasisOriginalNoNorm |
🚧 Known Limitations
- Token Counter: Display pending frontend integration
- True Alternation: Uses blending, not per-step switching (would need sampler hooks)
- Without Lark: Nested syntax won't parse; falls back to simple regex
Usage Examples
# Basic emphasis
a (beautiful:1.3) landscape
# BREAK for isolation
artist name BREAK character name, wearing a hat
# Step-based scheduling (with steps=28)
[mountains:ocean:14] at sunset
# Nested scheduling with alternation
1girl, [honovy:chen bin, [as109|fkey]:0.4]
# Combined
(epic:1.2) [forest:city:0.3] BREAK detailed background