Enum ContentBlock 

pub enum ContentBlock {
    Text {
        text: String,
    },
    ImageBase64 {
        data: String,
        media_type: String,
    },
    ImageUrl {
        url: String,
        detail: String,
    },
    VideoPath {
        path: String,
        fps: Option<f32>,
        max_frames: Option<u32>,
    },
    VideoUrl {
        url: String,
        fps: Option<f32>,
        max_frames: Option<u32>,
    },
    VideoBase64 {
        data: String,
        media_type: String,
        fps: Option<f32>,
        max_frames: Option<u32>,
    },
    AudioPath {
        path: String,
        sample_rate: Option<u32>,
    },
    AudioUrl {
        url: String,
        sample_rate: Option<u32>,
    },
    AudioBase64 {
        data: String,
        media_type: String,
        sample_rate: Option<u32>,
    },
}

A content block in a multimodal message.

The image variants (ImageBase64, ImageUrl) are fully wired on the native Qwen2.5-VL backend. The video variants (VideoPath, VideoUrl, VideoBase64) are defined on the public request surface so higher-level tooling can express Qwen2.5-VL video-understanding payloads, but the native backend returns crate::InferenceError::UnsupportedMode for them until the video-tokenization path lands. Remote multimodal providers (Anthropic, Google Vertex) accept them through the protocol handlers today.

Variants

Text

Plain text content.

Fields

text: String

ImageBase64

Base64-encoded image.

Fields

data: String

Base64-encoded image data.

media_type: String

MIME type (e.g., “image/png”, “image/jpeg”).

ImageUrl

Image from URL.

Fields

url: String

URL of the image.

detail: String

Detail level for image processing (“auto”, “low”, “high”).

VideoPath

Video loaded from a local filesystem path. Qwen2.5-VL samples the clip at fps frames/sec (default: backend-chosen) and caps at max_frames to respect context budgets.

Fields

path: String

fps: Option<f32>

max_frames: Option<u32>

VideoUrl

Video accessible over HTTP(S). Semantics as ContentBlock::VideoPath.

Fields

url: String

fps: Option<f32>

max_frames: Option<u32>

VideoBase64

Base64-encoded video bytes. Prefer VideoPath when possible; inline base64 is expensive to round-trip.

Fields

data: String

media_type: String

fps: Option<f32>

max_frames: Option<u32>

AudioPath

Audio loaded from a local filesystem path. Used for audio-understanding models (Gemma 4 small variants, Gemini).

Fields

path: String

sample_rate: Option<u32>

Optional explicit sample-rate hint. Most backends will resample internally; this is a best-effort declaration.

AudioUrl

Audio accessible over HTTP(S).

Fields

url: String

sample_rate: Option<u32>

AudioBase64

Base64-encoded audio bytes.

Fields

data: String

media_type: String

sample_rate: Option<u32>

Implementations

impl ContentBlock

pub fn is_video(&self) -> bool

Return true if this block carries video data (any encoding). Used by backends that need to refuse video inputs until the tokenization path is wired.

pub fn is_audio(&self) -> bool

Return true if this block carries audio data (any encoding). Used by backends that need to refuse audio inputs until the tokenization path is wired. Gemma 4 small variants and Gemini accept audio; everything else in CAR rejects with UnsupportedMode.

Trait Implementations

impl Clone for ContentBlock

fn clone(&self) -> ContentBlock

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for ContentBlock

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for ContentBlock

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for ContentBlock

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.