Sign upAI Image Refinement Pipeline
Step 1: Input Processing
Your input prompt is bundled within the default prompt.
default_prompt: "Create me an image of [user_prompt]"
Step 4: User Selection
You select specific attributes from each section to refine your image.
✓ Lighting: bright, soft
✓ Style: realistic
✓ Background: simple
✓ Style: realistic
✓ Background: simple
Step 2: LLM Enhancement
GPT-4 analyzes your prompt and generates 3 new attribute sections with 5 options each.
"Create 3 new attribute sections for: cat"
Step 5: Iteration
You can repeat steps 2-4 up to 10 times to further refine your image.
🔄 Steps 2-4 repeat
Max 10 iterations
3 gems per cycle
Step 3: Attribute Generation
GPT-4 returns structured JSON with section names and attribute options.
Lighting: bright, soft, dramatic
Style: realistic, artistic, cartoon
Background: simple, complex, blurred
Style: realistic, artistic, cartoon
Background: simple, complex, blurred
Step 6: Generation
Your selections are combined into a detailed prompt and sent to DALL-E 3.
"Generate me an image of cat with:
• Lighting: bright, soft
• Style: realistic
• Background: simple"
• Lighting: bright, soft
• Style: realistic
• Background: simple"
🎨 → DALL-E 3 API (10 gems)