An image content part used in multimodal ALLM.Message{:content} lists.
See spec §35.6.
Layer A — pure serializable data. :image (an %ALLM.Image{}) is the only
required field. :detail is a closed-set atom in [:auto, :low, :high]
matching OpenAI's vision-detail wire field; :auto is the default and
matches OpenAI's documented default. :metadata defaults to an empty map.
Validation
ALLM.Validate.message/1 accepts any %ImagePart{} in a content list per
Decision #11 (Phase 14.4). Layer A does not enforce :detail membership —
callers passing an unknown atom (e.g. :medium) round-trip cleanly through
ETF but JSON decoding will surface
{:_unknown, :atom_decode_failed} per the Serializer rescue contract
(Phase 13).
Serializability
ETF round-trip via :erlang.term_to_binary/1 is total. JSON round-trip via
ALLM.Serializer follows the standard __type__-tagged wire shape; the
embedded %ALLM.Image{} round-trips through its own encoder.
Summary
Functions
Build an %ImagePart{} from an %ALLM.Image{} and optional keyword opts.
Types
@type detail() :: :auto | :low | :high
@type t() :: %ALLM.ImagePart{detail: detail(), image: ALLM.Image.t(), metadata: map()}
Functions
@spec new( ALLM.Image.t(), keyword() ) :: t()
Build an %ImagePart{} from an %ALLM.Image{} and optional keyword opts.
Accepts :detail (one of :auto | :low | :high; default :auto) and
:metadata (a map).
Examples
iex> img = ALLM.Image.from_url("https://example.com/cat.png")
iex> ALLM.ImagePart.new(img).detail
:auto
iex> img = ALLM.Image.from_url("https://example.com/cat.png")
iex> ALLM.ImagePart.new(img, detail: :high).detail
:high