V 1.0.0 - New readme

2024-11-22 13:20:53 +01:00
parent a868e88907
commit 2fb83d2964
5 changed files with 337 additions and 1 deletions
@@ -298,4 +298,153 @@ Remember: Fine-tune these settings based on your specific use case and desired o
 #### Hands, but aesthetic scorer automatically ignored the second:
 <img src="refiner_block_hands2.jpg" width="500px">

+<hr>
+
+# <ins>DiT Prompt Purifier Node:</ins>
+
+<img src="prompt_dit_cleaner.jpg" width="500px">
+
+### Overview:
+The DiT Prompt Purifier node automatically cleans and optimizes prompts for modern Diffusion Transformer (DiT) architectures by removing incompatible elements, weights, and structural components that may interfere with newer models.
+
+### Key Features:
+
+- Automatic model version detection
+- Character length limitation
+- Architecture-specific prompt purification
+- Selective cleaning based on model Architecture
+- Incompatible weight and structure removal
+
+### Cleaning Operations
+
+#### Removes:
+- Prompt weights (e.g., `(element:1.2)`)
+- Legacy break commands (`BREAK`)
+- Nested parentheses structures
+- Old-style prompt formatting
+
+#### Controls:
+- `max_length`: Character limit (0 = unlimited)
+- Model-specific purification toggles
+- Automatic architecture detection by loaded checkpoint
+
+### Architecture Controls:
+
+#### Standard Models (no need to purify)
+- `purify_sd1`: SD1.x prompts
+- `purify_sd2`: SD2.x prompts
+- `purify_sdxl`: SDXL prompts
+- `purify_sd3`: SD3 prompts
+
+#### Modern DiT Models (have to purify)
+- `purify_stablecascade`: Stable Cascade
+- `purify_flux`: Flux models
+- `purify_kwaikolors`: KwaiKolors
+- `purify_hunyuan`: Hunyuan
+
+#### Specialized Models (no need to purify)
+- `purify_turbo`: Turbo models
+- `purify_playground`: Playground
+- `purify_pony`: Pony models
+- `purify_lcm`: LCM variants
+- `purify_lightning`: Lightning
+- `purify_hyper`: Hyper models
+- `purify_pixartsigma`: PixArt-Sigma
+
+### Use Cases:
+
+#### Basic Prompt Cleaning:
+```plaintext
+Input: "(high quality:1.2), (detailed:1.4) BREAK (masterpiece)"
+Output: "high quality, detailed, masterpiece"
+```
+
+#### Length Control:
+```plaintext
+Settings:
+- max_length: 50
+- purify_flux: ON
+
+Input: "Long prompt with excessive description and weights..."
+Output: "Truncated prompt within 50 characters but keep the whole last word..."
+```
+
+### Configuration Examples:
+
+#### Flux Optimization:
+```plaintext
+Settings:
+- purify_flux: ON
+- max_length: 0
+
+Benefits:
+- DiT-compatible structure
+- Weight removal
+- Clean formatting
+```
+
+#### Mixed Model Pipeline:
+```plaintext
+Settings:
+- Keep non-target models OFF
+- Enable specific model purification
+- Set appropriate length limit
+
+Benefits:
+- Model-specific optimization
+- Controlled cleaning
+- Maintained compatibility
+```
+
+### Benefits:
+
+#### Compatibility:
+- Ensures DiT architecture compliance
+- Removes problematic structures
+- Maintains old (A1111 compatible) prompt essence
+
+### Best Practices:
+
+1. **Model Selection**
+   - Enable purification for target model
+   - Disable for non-relevant architectures
+   - Verify model detection
+
+2. **Length Management**
+   - Set appropriate max_length
+   - Use 0 for unlimited length
+   - Monitor output for truncation
+
+3. **Cleaning Strategy**
+   - Enable relevant model switches
+   - Check debug output
+   - Verify cleaned prompt
+
+### Recommended Settings:
+
+#### DiT Models:
+```plaintext
+- set purify_<model_type>: ON
+- max_length: 0 (or set if the result too noisy)
+- other model type purify switches: OFF
+```
+
+#### Legacy Models:
+```plaintext
+Keep settings:
+- purify switches: OFF
+- max_length: 0
+- Maintains original structure
+```
+
+### Tips for Usage:
+
+1. Monitor debug output for cleaning verification
+2. Adjust length limits based on model requirements
+3. Enable only relevant DiT model purification
+4. Test cleaned prompts before batch processing
+5. Keep original prompts for reference
+
+Remember: DiT prompt purification is essential for optimal performance with modern architectures while maintaining prompt effectiveness.
+
 <hr>
@@ -395,6 +395,140 @@ This node automatically saves images along with metadata, generated by the full

 <hr>

+# <ins>Network Tag Cleaner Node:</ins>
+
+<img src="prompt_network_cleaner.jpg" width="250px">
+
+### Overview:
+The Network Tag Cleaner node provides intelligent management of network adapter tags in prompts, supporting both manual and automatic cleaning modes for various model architectures.
+
+### Supported Network Types:
+- Embeddings (`embedding:name`)
+- LoRA (`<lora:name>`)
+- LyCORIS (`<lycoris:name>`)
+- Hypernetworks (`<hypernetwork:name>`)
+
+### Operation Modes:
+
+#### Manual Mode:
+When `auto_remover` is set to OFF:
+- Full control over tag removal
+- Selective cleaning based on model architecture
+- Requires manual configuration
+
+#### Auto Mode:
+When `auto_remover` is set to ON:
+- Intelligent compatibility checking
+- Automatic removal of only incompatible tags
+- Preserves all matching architecture tags
+
+### Control Switches:
+
+#### Network Type Controls:
+- `remove_embedding`: Toggle embedding removal
+- `remove_lora`: Toggle LoRA tag removal
+- `remove_lycoris`: Toggle LyCORIS tag removal
+- `remove_hypernetwork`: Toggle hypernetwork removal
+
+#### Architecture-Specific Controls:
+- `remove_only_sd1`: SD1.x network tags
+- `remove_only_sd2`: SD2.x network tags
+- `remove_only_sdxl`: SDXL network tags
+- `remove_only_sd3`: SD3 network tags
+- `remove_only_stablecascade`: Stable Cascade tags
+- `remove_only_turbo`: Turbo model tags
+- `remove_only_flux`: Flux model tags
+- `remove_only_kwaikolors`: KwaiKolors tags
+- `remove_only_hunyuan`: Hunyuan tags
+- `remove_only_playground`: Playground tags
+- `remove_only_pony`: Pony tags
+- `remove_only_lcm`: LCM tags
+- `remove_only_lightning`: Lightning tags
+- `remove_only_hyper`: Hyper tags
+- `remove_only_pixartsigma`: PixArt-Sigma tags
+
+### Use Cases:
+
+#### Manual Cleaning Mode:
+```plaintext
+Scenario 1: SDXL-Only Workflow
+Settings:
+- auto_remover: OFF
+- remove_lora: ON
+- remove_only_sd1: ON
+- remove_only_sd2: ON
+- remove_only_sdxl: OFF
+
+Result: Removes all SD1 and SD2 LoRA tags while keeping SDXL LoRAs
+```
+
+#### Auto Cleaning Mode:
+```plaintext
+Scenario 1: Using Flux Checkpoint
+Settings:
+- auto_remover: ON
+- remove_lora: ON
+- remove_lycoris: ON
+
+Result: 
+- Keeps Flux-compatible network tags
+- Removes SD1/SD2/SDXL and all other non-flux network tags
+- Preserves model-specific optimizations
+```
+
+```plaintext
+Scenario 2: Using SDXL Checkpoint
+Settings:
+- auto_remover: ON
+- remove_lora: ON
+
+Result:
+- Maintains SDXL LoRA tags
+- Removes all other incompatible network tags
+- Ensures workflow compatibility
+```
+
+### Benefits:
+
+#### Workflow Optimization:
+- Prevents incompatible network usage
+- Reduces generation errors
+- Streamlines prompt management
+
+#### Error Prevention:
+- Eliminates architecture mismatches
+- Prevents memory issues
+- Reduces failed generations
+
+#### User Convenience:
+- Automatic compatibility checking
+- Simple toggle controls
+- Clear visual feedback
+
+#### Advanced Usage:
+```plaintext
+Settings:
+- auto_remover: ON
+- All network types enabled
+- Model-specific switches as needed
+
+Benefits:
+- Comprehensive tag management
+- Maximum compatibility
+- Automated workflow optimization
+```
+
+### Recommendations:
+
+1. Start with auto mode for general use
+2. Switch to manual mode for specific needs
+3. Regular prompt verification
+4. Keep model architecture in mind
+
+Remember: The cleaner node ensures optimal network compatibility while maintaining prompt integrity and workflow efficiency.
+
+<hr>
+
 # <ins>Network Tag Loader:</ins>

 <img src="networks_loader.jpg" width="250px">
@@ -431,4 +565,57 @@ This node manages additional networks such as Lora, Lycoris, and hypernetworks f
 - `Enhanced Control`: Allows fine-tuning of model behavior using additional networks
 - `Prompt Customization`: Adjusts prompts dynamically using Lora and Lycoris keywords for more refined image generation
 - `Network Compatibility`: Supports multiple network types, providing flexibility in mixing styles and behaviors
- `Workflow Integration`: Seamlessly integrates with existing models and workflows in ComfyUI for consistent results
+- `Workflow Integration`: Seamlessly integrates with existing models and workflows in ComfyUI for consistent results
+
+<hr>
+
+# <ins>Primere Embedding Handler:</ins>
+
+<img src="prompt_embedding.jpg" width="500px">
+
+A ComfyUI node that automatically converts Automatic1111-style textual inversion embedding syntax to ComfyUI-compatible format.
+
+### Description:
+
+This node allows you to use Automatic1111-style embedding syntax in your prompts while working in ComfyUI. It automatically detects embedding keywords and converts them to the proper ComfyUI format.
+
+### How It Works:
+
+#### Input Format (Automatic1111 style):
+- Simple keywords like `Autumn` or `Summer`
+- No special prefix needed
+- Must match exactly with embedding filename (case-sensitive) without extension
+
+#### Output Format (ComfyUI style):
+- Converted to `embedding:Autumn` or `embedding:Summer`
+- Properly formatted for ComfyUI processing
+- Only converts if matching embedding file exists
+
+### Example:
+
+Input prompt:
+```
+This prompt contains Autumn. The embedding name is case sensitive: summer just a word, but Summer is embedding.
+```
+
+Output prompt:
+```
+This prompt contains embedding:Autumn. The embedding name is case sensitive: summer just a word, but embedding:Summer is embedding.
+```
+
+### Features:
+
+- Seamless conversion between Automatic1111 and ComfyUI embedding formats
+- Case-sensitive matching to prevent unwanted conversions
+- Only converts existing embeddings
+- Works with both positive and negative prompts
+- Real-time conversion during workflow execution
+
+### Benefits:
+
+- Use Automatic1111 prompts directly in ComfyUI without manual conversion
+- Maintain compatibility with existing prompt collections
+- Reduce errors from incorrect embedding syntax
+- Save time on prompt formatting
+
+<hr>