Task 1: Semantic Style Frame Creation

Objective: Generate a stylized keyframe that semantically aligns a reference style with a target subject.

Current Approach:

• Use Flux Kontext as primary solution
• Implement Claude agent to prompt for subject and style extraction
• Remix reference image into target subject while maintaining style consistency

Future Iterations:

• Fine-tune Flux Kontext with open weights for direct processing
• Explore OmniGen2 as alternative solution
• Current testing shows Flux Kontext as best performing option

Task 2: Semantic Image Alignment with Keyframe

Objective: Precisely edit the generated style frame to match the exact keyframe positioning and composition.

Current State:

• Existing image model provides decent baseline
• Issue: Gets overpowered when using ControlNet Union

Implementation Options:

1. Style Guide Implementation: Single-shot semantic fitting without manual ControlNet selection
2. Video Model Transition: Use GPT 4o video model to semantically align frames through transition analysis

Requirements: