fix: LLMClient remove explicit provider and validate HF token starts with 'hf_' for auto-routing

fix: explicitly set provider='hf-inference' in LLMClient to bypass auto-router error

fix: explicitly set provider='hf-inference' in LLMClient to bypass auto-router error

fix: normalize empty HF token to None to fix auto-router error in InferenceClient

fix: switch IMAGE_PROVIDER from fal-ai (paid) to hf-inference (free)

fix: switch IMAGE_MODEL_ID to FLUX.1-schnell (supports text-to-image with fal-ai)

style: increase caption font size and overlay opacity for better readability

fix: handle gemini-2.5-flash thinking model JSON output (strip think tags, use response_mime_type, increase max_tokens)

update gemini models: text=gemini-2.5-flash, image=gemini-2.0-flash-preview-image-generation

show actual error detail in UI for debugging

add gemini api logic

add gemini api config

feat(app): enhance comic generation with improved error handling and user guidance

feat(caption): enhance caption handling in images and improve output formatting

feat(comic): switch to FLUX API and generate validated 4-scene storyboards

Refactor LLMClient to use Hugging Face Inference API, removing local model loading and CUDA dependencies. Updated requirements to include huggingface_hub. Enhanced logging for API readiness.

Update file permissions and refactor app logic for comic generation. Improved error handling in inference function and enhanced logging for model loading. Cleaned up code formatting and removed dotenv dependency from config.

Add initial project structure for AI Comic Generation