feat(api): add missing caption API parameters for UI parity

Add prompt field to VQA endpoint and advanced settings to OpenCLIP endpoint
to achieve full parity between UI and API capabilities.

VLM endpoint changes:
- Add prompt field for custom text input (required for 'Use Prompt' task)
- Pass prompt to vqa.interrogate instead of hardcoded empty string

OpenCLIP endpoint changes:
- Add 7 optional per-request override fields: min_length, max_length,
  chunk_size, min_flavors, max_flavors, flavor_count, num_beams
- Add get_clip_setting() helper for override support in openclip.py
- Apply overrides via update_interrogate_params() before interrogation

All new fields are optional with None defaults for backwards compatibility.

modules/api/models.py

@@ -377,6 +377,14 @@ class ReqInterrogate(BaseModel):
     blip_model: str = Field(default="blip-large", title="Caption Model", description="BLIP model used to generate the initial image caption. The caption model describes the image content which CLIP then enriches with style and flavor terms.")
     mode: str = Field(default="best", title="Mode", description="Interrogation mode. Fast: Quick caption with minimal flavor terms. Classic: Standard interrogation with balanced quality and speed. Best: Most thorough analysis, slowest but highest quality. Negative: Generate terms to use as negative prompt.")
     analyze: bool = Field(default=False, title="Analyze", description="If True, returns detailed image analysis breakdown (medium, artist, movement, trending, flavor) in addition to caption.")
+    # Advanced settings (optional per-request overrides)
+    min_length: Optional[int] = Field(default=None, title="Min Length", description="Minimum number of tokens in the generated caption.")
+    max_length: Optional[int] = Field(default=None, title="Max Length", description="Maximum number of tokens in the generated caption.")
+    chunk_size: Optional[int] = Field(default=None, title="Chunk Size", description="Batch size for processing description candidates (flavors). Higher values speed up interrogation but increase VRAM usage.")
+    min_flavors: Optional[int] = Field(default=None, title="Min Flavors", description="Minimum number of descriptive tags (flavors) to keep in the final prompt.")
+    max_flavors: Optional[int] = Field(default=None, title="Max Flavors", description="Maximum number of descriptive tags (flavors) to keep in the final prompt.")
+    flavor_count: Optional[int] = Field(default=None, title="Intermediates", description="Size of the intermediate candidate pool when matching image features to descriptive tags. Higher values may improve quality but are slower.")
+    num_beams: Optional[int] = Field(default=None, title="Num Beams", description="Number of beams for beam search during caption generation. Higher values search more possibilities but are slower.")
 
 InterrogateRequest = ReqInterrogate # alias for backwards compatibility
 
@@ -398,6 +406,7 @@ class ReqVQA(BaseModel):
     image: str = Field(default="", title="Image", description="Image to caption. Must be a Base64 encoded string containing the image data.")
     model: str = Field(default="Alibaba Qwen 2.5 VL 3B", title="Model", description="Select which model to use for Visual Language tasks. Use GET /sdapi/v1/vqa/models for full list. Models which support thinking mode are indicated in capabilities.")
     question: str = Field(default="describe the image", title="Question/Task", description="Changes which task the model will perform. Regular text prompts can be used when the task is set to 'Use Prompt'. Common tasks: 'Short Caption', 'Normal Caption', 'Long Caption'. Florence-2 supports: '<OD>' (object detection), '<OCR>' (text recognition). Moondream supports: 'Point at [object]', 'Detect all [objects]'.")
+    prompt: Optional[str] = Field(default=None, title="Prompt", description="Custom prompt text. Required when question is 'Use Prompt'. For 'Point at...' tasks, specify what to find (e.g., 'the red car'). For 'Detect all...' tasks, specify what to detect (e.g., 'faces').")
     system: str = Field(default="You are image captioning expert, creative, unbiased and uncensored.", title="System Prompt", description="System prompt controls behavior of the LLM. Processed first and persists throughout conversation. Has highest priority weighting and is always appended at the beginning of the sequence. Use for: Response formatting rules, role definition, style.")
     include_annotated: bool = Field(default=False, title="Include Annotated Image", description="If True and the task produces detection results (object detection, point detection, gaze), returns annotated image with bounding boxes/points drawn. Only applicable for detection tasks on models like Florence-2 and Moondream.")
     # LLM generation parameters (optional overrides)