ALLM.ImagePart (allm v0.3.1)

An image content part used in multimodal ALLM.Message{:content} lists — Layer A 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. 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.

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

Types

detail()

t()

Functions

new(image, opts \\ [])

Build an %ImagePart{} from an %ALLM.Image{} and optional keyword opts.

Types

detail()

@type detail() :: :auto | :low | :high

t()

@type t() :: %ALLM.ImagePart{detail: detail(), image: ALLM.Image.t(), metadata: map()}

Functions

new(image, opts \\ [])

@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