# filename: flux_consistent_integration_plan.txt

## Phase 1: Abstraction & Unification Strategy

- `FluxImageGenerator` uses `gradio_client` to interact with a remote Hugging Face space.
- `IPAdapterRunner` is a local generation pipeline using diffusers + IP-Adapter.
- Both support prompt-based image generation with customizable parameters (prompt, steps, cfg, seed, etc).

### Goal
Create a **unified wrapper** class or interface that allows flexible switching between:
- **Online generation** (via Flux Gradio Space)
- **Local consistent generation** (via IP-Adapter)

---

## Phase 2: Interface Unification Blueprint

### Abstract base class: `ImageGeneratorInterface`

```python
class ImageGeneratorInterface:
    def generate(self, prompt: str, **kwargs) -> list[Image.Image]:
        raise NotImplementedError("Subclasses must implement 'generate'")

Phase 3: Adapter Classes
Adapter 1: FluxClientAdapter(ImageGeneratorInterface)

Wraps FluxImageGenerator using its generate_image() method, returns PIL image list.
Adapter 2: IPAdapterLocalAdapter(ImageGeneratorInterface)

Wraps IPAdapterRunner, uses generate_text2img() or others, returns PIL image list.
Phase 4: Switcher Logic

Implement a GeneratorSwitcher or factory function that selects an adapter based on:

    User preference (e.g., "local" vs "remote")

    Availability (e.g., fallback to remote if CUDA unavailable)

    Scenario (e.g., "consistent generation" requires local IP-Adapter)

Phase 5: Shared Prompt Configuration

Create a shared PromptConfig dataclass:

@dataclass
class PromptConfig:
    prompt: str
    negative_prompt: str = ""
    steps: int = 35
    cfg_scale: float = 7.0
    seed: int = 42
    sampler: str = "DPM++ 2M Karras"
    strength: float = 0.7
    enhance_prompt_option: bool = False

This ensures both adapters consume a common configuration object, simplifying switching and reproducibility.
Phase 6: Optional Prompt Enhancers

Support optional enhancements:

    Mistral/Nemo prompt enhancements (for FluxClient)

    IP-based guidance (for IPAdapterLocal)

Phase 7: Next Steps...

    Draft interfaces & adapters for both pipelines

    Normalize return types to always return list of PIL.Image.Image

    Add support for preview grids via image_grid()

    Consider hybrid fallback/merging strategy

➡️ Interested in getting a draft of the ImageGeneratorInterface and first adapter implementation?