Expand description
One trait per task, many engines per trait.
An engine is a named, swappable implementation of a task — exactly the
role a codec plays in ffmpeg. Engines are registered in an
crate::registry::EngineRegistry and selected by name.
Structs§
- AsrOptions
- Depth
Options - Options for a depth run.
- Depth
Output - Image → objects (Diana). What a depth engine returns: a dense map plus what is needed to place it back on the source image.
- Detect
Options - Detection options (Diana).
- Engine
Info - Metadata every engine exposes for discovery (
ffai engines). - OcrOptions
- TtsOptions
- VlmOptions
- Options for a VLM call (Argus).
- VlmPrompt
- An ordered, interleaved multimodal prompt —
text <img> text <img> text.
Enums§
- Decoding
- How a VLM decoder picks each next token.
- Engine
Status - Honesty marker shown in
ffai engines— stubs are visible, not hidden. - Task
- The tasks
FFaiknows about (the “stream types” of the toolkit). - VlmPart
- One piece of a multimodal prompt.
Traits§
- AsrEngine
- Speech → text (Mercury).
- Depth
Engine - Monocular depth estimation.
- Detect
Engine - OcrEngine
- Image → text (Carmenta).
- TtsEngine
- Text → speech (Mercury).
- VlmEngine
- Image/video → description (Argus).