Crate ollama_vision 

ollama-vision

Robust Ollama vision model toolkit for image tagging and captioning.

Features

• Image tagging with a 7-strategy response parser that handles every common LLM output format (JSON arrays, code blocks, <think> tags, JSON objects, numbered lists, comma-separated text)
• Image captioning with automatic <think> block stripping
• Works with any Ollama vision model (llava, minicpm-v, llama3.2-vision, etc.)
• Base64 API for in-memory images (no file I/O required)

Quick Start

use ollama_vision::{OllamaVisionConfig, TagOptions, CaptionOptions};
use std::path::Path;

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let config = OllamaVisionConfig::with_model("llava");
    let client = reqwest::Client::new();

    // Tag an image
    let tags = ollama_vision::tag_image(
        &client, &config,
        Path::new("photo.jpg"),
        &TagOptions::default(),
    ).await?;
    println!("Tags: {:?}", tags);

    // Caption an image
    let caption = ollama_vision::caption_image(
        &client, &config,
        Path::new("photo.jpg"),
        &CaptionOptions::default(),
    ).await?;
    println!("Caption: {}", caption);

    Ok(())
}

Parsing Robustness

The tag parser handles all common LLM response formats:

use ollama_vision::parse_tags;

// Pure JSON array
assert!(parse_tags(r#"["portrait", "fantasy"]"#).is_ok());

// With <think> blocks from reasoning models
assert!(parse_tags(r#"<think>analyzing...</think>["portrait"]"#).is_ok());

// Markdown code blocks
assert!(parse_tags("```json\n[\"portrait\"]\n```").is_ok());

// JSON objects with "tags" key
assert!(parse_tags(r#"{"tags": ["portrait"]}"#).is_ok());

// Comma-separated fallback
assert!(parse_tags("portrait, fantasy, dark").is_ok());

Re-exports

pub use captioner::caption_image;

pub use captioner::caption_image_base64;

pub use captioner::CaptionError;

pub use tagger::tag_image;

pub use tagger::tag_image_base64;

pub use tagger::TagError;

pub use types::CaptionOptions;

pub use types::GenerateOptions;

pub use types::OllamaVisionConfig;

pub use types::TagOptions;

Modules

captioner

parser

LLM response parser — delegates to llm-output-parser crate.

tagger

types

Enums

ParseError

Errors returned by output parsers.

Functions

parse_tags

Parse an LLM response into a cleaned list of strings.

strip_think_tags

Strip all <think>...</think> and <thinking>...</thinking> blocks from text.