EricRollei-Local_LLM_Prompt_Enhancer/docs/QUICK_REFERENCE.md

Quick Reference Guide - Eric's Prompt Enhancers

Finding the Nodes

After restarting ComfyUI, all nodes appear under:

Add Node → Eric Prompt Enhancers

You'll find 5 nodes:

1. Video Prompt Expander
2. Video Prompt Expander (Advanced)
3. Image-to-Video Prompt Expander
4. Image-to-Image Prompt Expander
5. Text-to-Image Prompt Enhancer ⭐ NEW

---

Text-to-Image Prompt Enhancer (NEW)

Platform Selection

Your Image Model	Select Platform
FLUX.1-dev/schnell	flux
Stable Diffusion XL	sd_xl
Pony Diffusion v6	pony
Illustrious XL	illustrious
Meissonic/MeissonFlow	chroma
Qwen Image	qwen_image
Qwen Image Edit	qwen_image_edit
Wan Image Models	wan_image

Platform Requirements

Pony Diffusion (CRITICAL!)

Positive MUST start with: score_9, score_8_up, score_7_up
Use underscores: long_hair NOT "long hair"
Negative MUST include: score_6, score_5, score_4

Illustrious XL

Positive MUST start with: masterpiece, best quality
Use underscores in tags
Detailed character appearance

All Others

Follow platform guidelines automatically
Node handles formatting

Quick Workflow

[Text-to-Image Prompt Enhancer]
  - text_prompt: "your simple description"
  - target_platform: (match your image model)
  - camera_angle: (optional, or "auto")
  - lighting_source: (optional, or "auto")
  - time_of_day: (optional, or "auto")
  ↓
positive_prompt → [CLIP Text Encode] → [KSampler]
negative_prompt → [CLIP Text Encode] → [KSampler]

Control Options

Set to:

• "auto" = LLM decides what fits best
• "random" = Random choice for variety
• "none" = Omit this control
• Specific value = Force this setting

Available Controls:

• camera_angle: Eye level, low angle, high angle, bird's eye, etc.
• composition: Rule of thirds, centered, golden ratio, etc.
• lighting_source: Natural, studio, golden hour, moonlight, etc.
• lighting_quality: Soft, dramatic, volumetric, etc.
• time_of_day: Dawn, noon, golden hour, dusk, night, etc.
• weather: Clear, foggy, rainy, snowy, etc.
• art_style: Photorealistic, anime, oil painting, etc.
• color_mood: Vibrant, muted, warm tones, etc.
• detail_level: Standard, highly detailed, minimalist

Wildcards for Variety

camera_angle: random
lighting_source: random
weather: random
color_mood: random

Perfect for batch generation with variety!

---

Image-to-Image Prompt Expander

Platform Selection

Same as Text-to-Image, but focused on image editing:

• flux, sd_xl, wan22, hunyuan_image, qwen_image, qwen_image_edit

Quick Workflow

[Load Image]
  ↓
[Image-to-Image Prompt Expander]
  - image: (from Load Image)
  - change_request: "what to change"
  - target_platform: (match your model)
  - use_vision_model: true
  ↓
positive_prompt → [Your Image Gen Node]
negative_prompt → [Your Image Gen Node]

Vision Model Setup

LM Studio:

• Load a vision model (Llama 3.2 Vision, LLaVA)
• Start server
• vision_endpoint: http://localhost:1234/v1

Ollama:

• Pull vision model: ollama pull llama3.2-vision
• vision_endpoint: http://localhost:11434

---

Video Prompt Expander

Expansion Tiers

• auto: LLM analyzes and chooses tier
• basic: ~50-100 words, simple
• enhanced: ~100-200 words, detailed
• advanced: ~200-350 words, professional
• cinematic: ~300-500 words, film-quality

Presets

• cinematic: Professional film, dramatic lighting
• surreal: Dreamlike, otherworldly
• action: High-energy, dynamic
• stylized: Artistic, creative
• noir: Dark, moody, dramatic
• random: Random preset each time

Quick Workflow

[Video Prompt Expander]
  - basic_prompt: "your video idea"
  - preset: cinematic
  - expansion_tier: auto
  - mode: text-to-video
  ↓
positive_prompt_1 → [Your Video Gen Node]
negative_prompt → [Your Video Gen Node]

---

Image-to-Video Prompt Expander

Quick Workflow

[Load Image]
  ↓
[Image-to-Video Prompt Expander]
  - image: (from Load Image)
  - motion_description: "what should move/happen"
  - preset: cinematic
  - use_vision_model: true
  ↓
positive_prompt → [Your Video Gen Node]
negative_prompt → [Your Video Gen Node]

---

Common Settings (All Nodes)

LLM Backend

LM Studio:

llm_backend: lm_studio
api_endpoint: http://localhost:1234/v1
model_name: llama3 (or your loaded model)

Ollama:

llm_backend: ollama
api_endpoint: http://localhost:11434
model_name: llama3

Temperature

• 0.5-0.6: Precise, consistent
• 0.7: Balanced (recommended)
• 0.8-1.0: More creative
• 1.0+: Very creative, less predictable

Keywords

Positive Keywords:

"LoRA_trigger, cinematic_lighting, specific_style"

Comma-separated terms to include

Negative Keywords:

"watermark, blurry, low quality, distorted"

Comma-separated terms to avoid

---

Platform Examples

Flux Example

Input: "a woman in a garden"
Platform: flux
Output: "masterpiece, best quality, highly detailed photograph 
of a woman in a lush garden, golden hour lighting, professional 
photography, in the style of Annie Leibovitz, 8k uhd"

Pony Example

Input: "anime girl with cat ears"
Platform: pony
Output: "score_9, score_8_up, score_7_up, 1girl, cat_ears, 
animal_ears, long_hair, detailed_face, garden, best_quality"
Negative: "score_6, score_5, score_4, worst quality, low quality"

SDXL Example

Input: "cyberpunk city street"
Platform: sd_xl
Output: "masterpiece, best quality, highly detailed, cyberpunk 
city street at night, neon lights, wet pavement, professional"

Chroma Example

Input: "marketplace scene"
Platform: chroma
Output: "high quality, intricate, masterpiece: A bustling 
marketplace scene with vendors arranging colorful fruits, 
customers browsing stalls, hanging fabric awnings creating 
dappled shadows, dynamic composition, detailed..."

---

Troubleshooting

Nodes Not Appearing

1. Restart ComfyUI completely
2. Check console for errors
3. Verify __init__.py has no errors
4. Look under "Eric Prompt Enhancers" category

LLM Connection Failed

1. Is LM Studio/Ollama running?
2. Check API endpoint URL
3. Test with simple prompt
4. Check console for detailed error

Poor Output Quality

1. Wrong platform selected?
2. Temperature too high/low?
3. Model not appropriate for task?
4. Add more specific controls
5. Include positive/negative keywords

Pony/Illustrious Issues

1. Check output has required tokens
2. Verify underscores in tags
3. Make sure negative has score tags
4. Platform must be set to "pony" or "illustrious"

---

Tips for Best Results

Text-to-Image

1. ✅ Match platform to your image model
2. ✅ Set at least lighting or art style
3. ✅ Use "auto" for most controls initially
4. ✅ Add LoRA triggers to positive keywords
5. ✅ Always connect negative prompt

Image-to-Image

1. ✅ Use vision model for best results
2. ✅ Be specific in change requests
3. ✅ Match platform to target model
4. ✅ Set key aesthetic controls
5. ✅ Test without vision first if issues

Video Prompts

1. ✅ Use "auto" tier to start
2. ✅ Match preset to video style
3. ✅ Add LoRA triggers as keywords
4. ✅ Generate variations for options
5. ✅ Save successful prompts to file

---

File Saving

Enable save_to_file: true to save:

• Enhanced prompts
• All settings used
• Metadata
• Timestamps

Files saved to:

• Video: output/video_prompts/
• Image: output/txt2img_prompts/
• Img2Img: output/img2img_prompts/

---

Documentation Files

• README.md - Overview and installation
• TXT2IMG_GUIDE.md - Complete text-to-image guide with examples
• IMG2IMG_GUIDE.md - Image-to-image guide
• UPDATE_V16_ERIC.md - Version 1.6 update notes
• CONFIGURATION.md - LLM setup details
• LM_STUDIO_SETUP.md - LM Studio setup
• QUICKSTART.md - Getting started quickly

---

Need More Help?

1. Read full guides: See TXT2IMG_GUIDE.md for detailed examples
2. Check platform requirements: Each platform has specific needs
3. Console logs: Check ComfyUI console for detailed errors
4. Test simple prompts: Start basic, then add complexity
5. Experiment: Try different settings to learn what works

Happy prompting!