Skip to content

vllm_omni.model_extras.magi2 ¶

Request fields supported by the MAGI-2 Preview pipeline.

MAGI2_EXTRA_BODY_PARAMS `module-attribute` ¶

MAGI2_EXTRA_BODY_PARAMS = frozenset(
    {
        "seconds",
        "resolution",
        "image_path",
        "output_width",
        "output_height",
        "deterministic",
    }
)

MAGI2_EXTRA_OUTPUT_PARAMS `module-attribute` ¶

MAGI2_EXTRA_OUTPUT_PARAMS: frozenset[str] = frozenset()

get_magi2_video_generation_defaults ¶

get_magi2_video_generation_defaults(
    extra_body: Mapping[str, Any] | None = None,
) -> VideoGenerationDefaults

magi2_preserves_reference_image_size ¶

magi2_preserves_reference_image_size(
    *, model: str | None, revision: str | None = None
) -> bool

MAGI-2 owns checkpoint-aligned reference resize and padding.