Initial commit
This commit is contained in:
@@ -0,0 +1,443 @@
|
||||
# FLUX 1.1 Pro Model Optimization Guide
|
||||
|
||||
This reference provides technical details for using the FLUX 1.1 Pro model on Replicate via the MCP server.
|
||||
|
||||
## Model Information
|
||||
|
||||
**Model identifier**: `black-forest-labs/flux-1.1-pro`
|
||||
**Provider**: Replicate (accessed via installed Replicate MCP server)
|
||||
**Documentation**: https://replicate.com/black-forest-labs/flux-1.1-pro
|
||||
|
||||
## Using the Replicate MCP Server
|
||||
|
||||
The Replicate MCP server is already installed and provides direct access to create predictions. Use the `mcp__replicate__create_predictions` tool to generate images.
|
||||
|
||||
### Required Parameters
|
||||
|
||||
```typescript
|
||||
{
|
||||
version: "black-forest-labs/flux-1.1-pro", // Model identifier
|
||||
input: {
|
||||
prompt: string, // Your Midjourney-style prompt
|
||||
aspect_ratio: "16:9", // Default widescreen format
|
||||
output_format: "png", // PNG for highest quality
|
||||
safety_tolerance: number // Optional: 1-5, default is 2
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
### MCP Server Tool Usage
|
||||
|
||||
Call the tool with these parameters:
|
||||
```json
|
||||
{
|
||||
"version": "black-forest-labs/flux-1.1-pro",
|
||||
"input": {
|
||||
"prompt": "[Your enhanced Midjourney-style prompt here]",
|
||||
"aspect_ratio": "16:9",
|
||||
"output_format": "png"
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
The tool will return a prediction object with status and output URL when complete.
|
||||
|
||||
## FLUX 1.1 Pro Specifications
|
||||
|
||||
### Supported Aspect Ratios
|
||||
- **1:1** - Square (1024×1024)
|
||||
- **16:9** - Widescreen (1360×768) ← **Recommended default**
|
||||
- **21:9** - Ultrawide (1536×640)
|
||||
- **3:2** - Classic photo (1216×832)
|
||||
- **2:3** - Portrait (832×1216)
|
||||
- **4:5** - Portrait (960×1200)
|
||||
- **5:4** - Landscape (1200×960)
|
||||
- **9:16** - Vertical (768×1360)
|
||||
- **9:21** - Vertical ultrawide (640×1536)
|
||||
|
||||
**Default**: 16:9 for cinematic, editorial-quality outputs
|
||||
|
||||
### Output Formats
|
||||
- **png** - Highest quality, lossless ← **Recommended**
|
||||
- **jpg** - Smaller file size, some quality loss
|
||||
- **webp** - Modern format, good compression
|
||||
|
||||
### Safety Tolerance
|
||||
Range: 1-5 (default: 2)
|
||||
- **1**: Most restrictive
|
||||
- **2**: Default, balanced ← **Recommended**
|
||||
- **5**: Least restrictive
|
||||
|
||||
For Midjourney-style artistic content, use default (2) or higher (3-4).
|
||||
|
||||
## FLUX 1.1 Pro Strengths
|
||||
|
||||
### What FLUX Does Exceptionally Well
|
||||
|
||||
1. **Photorealism**
|
||||
- Near-perfect photographic quality
|
||||
- Accurate physical materials and textures
|
||||
- Realistic lighting and shadows
|
||||
- Natural human anatomy and proportions
|
||||
|
||||
2. **Text Rendering**
|
||||
- Accurate text in images (unlike most AI models)
|
||||
- Legible typography and signage
|
||||
- Proper letter spacing and formatting
|
||||
|
||||
3. **Complex Compositions**
|
||||
- Multiple subjects with accurate spatial relationships
|
||||
- Detailed backgrounds without quality loss
|
||||
- Consistent style across entire image
|
||||
|
||||
4. **Color Accuracy**
|
||||
- Precise color reproduction
|
||||
- Accurate skin tones
|
||||
- Faithful material colors (less stylized than Midjourney)
|
||||
|
||||
5. **Fine Details**
|
||||
- Intricate textures and patterns
|
||||
- Sharp focus with excellent clarity
|
||||
- High resolution output (up to 1536px on longest side)
|
||||
|
||||
### Optimizing for FLUX's Strengths
|
||||
|
||||
**Leverage photorealism:**
|
||||
```
|
||||
"ultra-realistic portrait, photographic quality, accurate skin texture with visible pores,
|
||||
natural lighting, highly detailed, 8k resolution"
|
||||
```
|
||||
|
||||
**Include text when needed:**
|
||||
```
|
||||
"vintage neon sign reading 'OPEN', glowing red letters, weathered metal frame,
|
||||
urban nighttime photography, highly detailed"
|
||||
```
|
||||
|
||||
**Request complex scenes:**
|
||||
```
|
||||
"bustling Tokyo street at night, multiple neon signs with Japanese text, crowds of people,
|
||||
reflections on wet pavement, cinematic street photography, rich detail throughout"
|
||||
```
|
||||
|
||||
**Specify accurate colors:**
|
||||
```
|
||||
"sunset over ocean, accurate warm orange and pink tones, natural color gradient,
|
||||
photographic color accuracy, no oversaturation"
|
||||
```
|
||||
|
||||
## Adapting Midjourney Prompts for FLUX
|
||||
|
||||
### Key Adaptations
|
||||
|
||||
1. **Be More Explicit About Artistic Treatment**
|
||||
|
||||
Midjourney automatically adds artistic elevation; FLUX is more literal.
|
||||
|
||||
❌ Midjourney approach: "portrait of a woman"
|
||||
✅ FLUX adaptation: "editorial portrait of a woman, professional color grading, cinematic lighting, artistic photography"
|
||||
|
||||
2. **Specify Color Grading Clearly**
|
||||
|
||||
❌ Too vague: "beautiful colors"
|
||||
✅ Specific: "warm color grading with orange and teal tones, cinematic color treatment"
|
||||
|
||||
3. **Request Composition Explicitly**
|
||||
|
||||
❌ Implicit: "dramatic photo"
|
||||
✅ Explicit: "low angle composition, dramatic perspective, carefully framed, rule of thirds"
|
||||
|
||||
4. **Add "Cinematic" for Elevation**
|
||||
|
||||
Include "cinematic" for Midjourney-quality elevation:
|
||||
- "cinematic photography"
|
||||
- "cinematic lighting"
|
||||
- "cinematic composition"
|
||||
- "cinematic color grading"
|
||||
|
||||
5. **Emphasize Professional Quality**
|
||||
|
||||
Add explicit quality markers:
|
||||
- "professional photography"
|
||||
- "editorial quality"
|
||||
- "award-winning"
|
||||
- "museum quality"
|
||||
- "expert color grading"
|
||||
|
||||
### FLUX-Specific Prompt Enhancements
|
||||
|
||||
**For Midjourney-level polish, add:**
|
||||
|
||||
```
|
||||
Base prompt: "Woman in red dress in field"
|
||||
|
||||
Enhanced for FLUX:
|
||||
"Woman in flowing red dress standing in golden wheat field, editorial fashion photography,
|
||||
cinematic composition with subject centered, golden hour lighting, warm color grading with
|
||||
rich earth tones, shot on Canon R5 85mm f/1.4, shallow depth of field, professional color
|
||||
grading, highly detailed, award-winning photography"
|
||||
```
|
||||
|
||||
**Quality enhancement formula:**
|
||||
```
|
||||
[Subject and scene] + [professional category] + [lighting specifics] +
|
||||
[color treatment] + [technical specs] + [quality markers]
|
||||
```
|
||||
|
||||
## Comparison: FLUX vs Midjourney
|
||||
|
||||
### Where FLUX Excels Over Midjourney
|
||||
- ✅ More accurate photorealism
|
||||
- ✅ Better text rendering
|
||||
- ✅ More literal interpretation (pro and con)
|
||||
- ✅ Higher detail retention in complex scenes
|
||||
- ✅ More accurate colors and materials
|
||||
|
||||
### Where Midjourney Has Edge
|
||||
- 🎨 Automatic artistic elevation
|
||||
- 🎨 More "creative" interpretation
|
||||
- 🎨 Stronger default stylization
|
||||
- 🎨 Better at abstract/artistic concepts
|
||||
|
||||
### Bridging the Gap
|
||||
|
||||
To achieve Midjourney's artistic quality with FLUX:
|
||||
|
||||
1. **Always include style descriptors**: "editorial", "cinematic", "artistic"
|
||||
2. **Specify color treatment**: Don't rely on automatic enhancement
|
||||
3. **Request composition**: Explicitly describe framing and perspective
|
||||
4. **Add quality terms**: "professional", "expert", "award-winning"
|
||||
5. **Reference artistic styles**: Mention photographers, movements, or films
|
||||
|
||||
## Prompt Engineering Strategies for FLUX
|
||||
|
||||
### Strategy 1: Technical Photography Anchor
|
||||
|
||||
Ground the prompt in photographic reality:
|
||||
```
|
||||
"shot on [camera/lens], [aperture], [lighting setup], [film stock or processing]"
|
||||
```
|
||||
|
||||
Example:
|
||||
```
|
||||
"shot on Hasselblad H6D, 80mm f/2.8, studio softbox lighting, professional color grading"
|
||||
```
|
||||
|
||||
### Strategy 2: Cinematic Reference
|
||||
|
||||
Reference films or cinematographers for immediate style:
|
||||
```
|
||||
"cinematography inspired by [film/cinematographer], [specific visual qualities]"
|
||||
```
|
||||
|
||||
Example:
|
||||
```
|
||||
"cinematography inspired by Blade Runner 2049, orange and teal color grading, volumetric fog,
|
||||
dramatic lighting"
|
||||
```
|
||||
|
||||
### Strategy 3: Artistic Movement Context
|
||||
|
||||
Place the image in an artistic tradition:
|
||||
```
|
||||
"[art movement] photography, [key characteristics], [notable practitioner]"
|
||||
```
|
||||
|
||||
Example:
|
||||
```
|
||||
"contemporary portrait photography, intimate and natural, inspired by Annie Leibovitz,
|
||||
editorial quality"
|
||||
```
|
||||
|
||||
### Strategy 4: Layered Quality Modifiers
|
||||
|
||||
Stack quality terms strategically:
|
||||
```
|
||||
[technical quality] + [artistic quality] + [professional category]
|
||||
```
|
||||
|
||||
Example:
|
||||
```
|
||||
"highly detailed, 8k resolution + cinematic composition, expert color grading +
|
||||
award-winning editorial photography"
|
||||
```
|
||||
|
||||
### Strategy 5: Environmental Storytelling
|
||||
|
||||
Rich environmental details enhance the Midjourney aesthetic:
|
||||
```
|
||||
[subject] in [detailed environment], [atmospheric conditions], [mood through details]
|
||||
```
|
||||
|
||||
Example:
|
||||
```
|
||||
"vintage car on coastal highway, scattered clouds in blue sky, late afternoon golden light,
|
||||
sense of nostalgia and freedom"
|
||||
```
|
||||
|
||||
## Common FLUX Optimization Patterns
|
||||
|
||||
### Portrait Photography
|
||||
```
|
||||
[Subject] [expression/pose], editorial portrait photography, [lighting direction and quality],
|
||||
[color treatment], shot on [camera] [lens] [aperture], [background treatment],
|
||||
professional color grading, highly detailed, [quality marker]
|
||||
```
|
||||
|
||||
### Landscape/Environment
|
||||
```
|
||||
[Location], [time of day], [weather/atmosphere], cinematic landscape photography,
|
||||
[composition], [lighting quality], shot on [camera] [lens], [color grading],
|
||||
[artistic reference], 8k resolution, highly detailed
|
||||
```
|
||||
|
||||
### Product Photography
|
||||
```
|
||||
[Product] on [surface/setting], editorial product photography, [lighting setup],
|
||||
[background], shot on [camera] [lens], [material emphasis], professional color grading,
|
||||
tack sharp focus, commercial photography quality
|
||||
```
|
||||
|
||||
### Architectural Photography
|
||||
```
|
||||
[Structure], [perspective], architectural photography, [time/lighting], [mood],
|
||||
shot on [camera] [tilt-shift lens], [compositional elements], professional color grading,
|
||||
highly detailed, [quality marker]
|
||||
```
|
||||
|
||||
### Fashion Editorial
|
||||
```
|
||||
[Subject] in [clothing], [pose/movement], high fashion editorial, [setting], [lighting],
|
||||
[aesthetic], shot on [camera] [lens], [artistic reference], [color treatment],
|
||||
award-winning fashion photography
|
||||
```
|
||||
|
||||
## Workflow Recommendations
|
||||
|
||||
### Step 1: Understand User Intent
|
||||
Clarify what the user wants to create - subject, mood, style, purpose.
|
||||
|
||||
### Step 2: Select Prompt Pattern
|
||||
Choose appropriate pattern from midjourney-style-guide.md based on genre.
|
||||
|
||||
### Step 3: Build Layered Prompt
|
||||
Construct prompt following the 5-layer structure from style guide.
|
||||
|
||||
### Step 4: Apply FLUX Adaptations
|
||||
Enhance with explicit quality terms, color grading, and technical specs.
|
||||
|
||||
### Step 5: Verify Quality
|
||||
Check against the style guide checklist (40-75 words, all elements present).
|
||||
|
||||
### Step 6: Generate via MCP Server
|
||||
Use `mcp__replicate__create_predictions` with optimized parameters.
|
||||
|
||||
### Step 7: Offer Variations
|
||||
Provide 2-3 alternative prompts emphasizing different aspects if requested.
|
||||
|
||||
## Parameter Selection Guide
|
||||
|
||||
### When to Use Different Aspect Ratios
|
||||
|
||||
**16:9 (Default)**:
|
||||
- Cinematic scenes
|
||||
- Landscapes
|
||||
- Editorial spreads
|
||||
- Professional photography
|
||||
|
||||
**1:1 (Square)**:
|
||||
- Social media posts
|
||||
- Profile pictures
|
||||
- Product shots
|
||||
- Symmetrical compositions
|
||||
|
||||
**2:3 / 4:5 (Portrait)**:
|
||||
- Fashion photography
|
||||
- Portrait photography
|
||||
- Magazine covers
|
||||
- Vertical social media
|
||||
|
||||
**3:2 (Classic Photo)**:
|
||||
- Traditional photography
|
||||
- Print photography
|
||||
- Gallery prints
|
||||
|
||||
**21:9 (Ultrawide)**:
|
||||
- Panoramic landscapes
|
||||
- Epic cinematic shots
|
||||
- Wide establishing shots
|
||||
|
||||
### Safety Tolerance Selection
|
||||
|
||||
**Level 2 (Default)**: Most Midjourney-style content
|
||||
**Level 3-4**: Artistic nude photography, mature themes (when appropriate)
|
||||
**Level 1**: Family-friendly, corporate, conservative contexts
|
||||
|
||||
## Advanced Techniques
|
||||
|
||||
### Multi-Element Composition
|
||||
For complex scenes, structure hierarchically:
|
||||
```
|
||||
[Primary subject] + [secondary elements] + [background] + [atmosphere] + [technical specs]
|
||||
```
|
||||
|
||||
Example:
|
||||
```
|
||||
"Elderly fisherman mending nets in foreground, weathered fishing boat in middle ground,
|
||||
misty harbor in background, early morning fog, golden hour lighting, cinematic composition,
|
||||
shot on ARRI Alexa 35mm, warm color grading, highly detailed, documentary photography style"
|
||||
```
|
||||
|
||||
### Color Palette Control
|
||||
Specify complete color schemes:
|
||||
```
|
||||
[Primary color] with [secondary colors], [color relationship], [color mood]
|
||||
```
|
||||
|
||||
Example:
|
||||
```
|
||||
"deep teal ocean with warm orange sunset, complementary color scheme, rich color saturation,
|
||||
cinematic color grading"
|
||||
```
|
||||
|
||||
### Mood Through Details
|
||||
Build atmosphere through specific environmental details:
|
||||
```
|
||||
[Subject] + [small environmental details] + [lighting that reinforces mood] + [color that reinforces mood]
|
||||
```
|
||||
|
||||
Example:
|
||||
```
|
||||
"Woman reading by window, dust particles visible in sunbeams, soft afternoon light,
|
||||
warm golden tones, sense of peaceful solitude, intimate and contemplative"
|
||||
```
|
||||
|
||||
## Troubleshooting Common Issues
|
||||
|
||||
### Issue: Output Too Literal, Not Artistic Enough
|
||||
**Solution**: Add "cinematic", "editorial", "artistic photography", "professional color grading"
|
||||
|
||||
### Issue: Colors Too Flat
|
||||
**Solution**: Specify "rich color saturation", "tonal depth", "expert color grading", reference film stock
|
||||
|
||||
### Issue: Composition Too Centered
|
||||
**Solution**: Explicitly request "rule of thirds", "dynamic composition", "carefully framed"
|
||||
|
||||
### Issue: Lighting Too Uniform
|
||||
**Solution**: Describe complete lighting setup with direction, quality, and motivation
|
||||
|
||||
### Issue: Missing Midjourney Polish
|
||||
**Solution**: Add quality stack: "highly detailed + professional color grading + award-winning + editorial quality"
|
||||
|
||||
## Best Practices Summary
|
||||
|
||||
1. ✅ Always include "cinematic" or "editorial" for artistic elevation
|
||||
2. ✅ Specify color grading explicitly (don't rely on defaults)
|
||||
3. ✅ Use technical photography terms to anchor realism
|
||||
4. ✅ Reference artistic styles or practitioners when appropriate
|
||||
5. ✅ Keep prompts 40-75 words for optimal results
|
||||
6. ✅ Layer quality modifiers strategically
|
||||
7. ✅ Describe lighting with direction and quality
|
||||
8. ✅ Request specific composition techniques
|
||||
9. ✅ Use 16:9 aspect ratio for Midjourney-style cinematic quality
|
||||
10. ✅ Choose PNG output format for maximum quality
|
||||
Reference in New Issue
Block a user