public/scripts/extensions/caption/index.js

import { ensureImageFormatSupported, getBase64Async, getFileExtension, isTrueBoolean, saveBase64AsFile } from '../../utils.js';
import { getContext, getApiUrl, doExtrasFetch, extension_settings, modules, renderExtensionTemplateAsync } from '../../extensions.js';
import { appendMediaToMessage, chat_metadata, eventSource, event_types, getRequestHeaders, saveChatConditional, saveSettingsDebounced, substituteParamsExtended } from '../../../script.js';
import { getMessageTimeStamp } from '../../RossAscends-mods.js';
import { SECRET_KEYS, secret_state } from '../../secrets.js';
import { getMultimodalCaption } from '../shared.js';
import { textgen_types, textgenerationwebui_settings } from '../../textgen-settings.js';
import { SlashCommandParser } from '../../slash-commands/SlashCommandParser.js';
import { SlashCommand } from '../../slash-commands/SlashCommand.js';
import { ARGUMENT_TYPE, SlashCommandArgument, SlashCommandNamedArgument } from '../../slash-commands/SlashCommandArgument.js';
import { commonEnumProviders } from '../../slash-commands/SlashCommandCommonEnumsProvider.js';
import { callGenericPopup, Popup, POPUP_TYPE } from '../../popup.js';
import { debounce_timeout, MEDIA_DISPLAY, MEDIA_SOURCE, MEDIA_TYPE, SCROLL_BEHAVIOR } from '../../constants.js';
export { MODULE_NAME };

const MODULE_NAME = 'caption';

const PROMPT_DEFAULT = 'What\'s in this image?';
const TEMPLATE_DEFAULT = '[{{user}} sends {{char}} a picture that contains: {{caption}}]';

/**
 * Migrates old extension settings to the new format.
 * Must keep this function for compatibility with old settings.
 */
function migrateSettings() {
    if (extension_settings.caption.local !== undefined) {
        extension_settings.caption.source = extension_settings.caption.local ? 'local' : 'extras';
    }

    delete extension_settings.caption.local;

    if (!extension_settings.caption.source) {
        extension_settings.caption.source = 'extras';
    }

    if (extension_settings.caption.source === 'openai') {
        extension_settings.caption.source = 'multimodal';
        extension_settings.caption.multimodal_api = 'openai';
        extension_settings.caption.multimodal_model = 'gpt-4-turbo';
    }

    if (!extension_settings.caption.multimodal_api) {
        extension_settings.caption.multimodal_api = 'openai';
    }

    if (!extension_settings.caption.multimodal_model) {
        extension_settings.caption.multimodal_model = 'gpt-4-turbo';
    }

    if (!extension_settings.caption.prompt) {
        extension_settings.caption.prompt = PROMPT_DEFAULT;
    }

    if (!extension_settings.caption.template) {
        extension_settings.caption.template = TEMPLATE_DEFAULT;
    }

    if (!extension_settings.caption.show_in_chat) {
        extension_settings.caption.show_in_chat = false;
    }
}

/**
 * Sets an image icon for the send button.
 */
async function setImageIcon() {
    try {
        const sendButton = $('#send_picture .extensionsMenuExtensionButton');
        sendButton.addClass('fa-image');
        sendButton.removeClass('fa-hourglass-half');
    }
    catch (error) {
        console.log(error);
    }
}

/**
 * Sets a spinner icon for the send button.
 */
async function setSpinnerIcon() {
    try {
        const sendButton = $('#send_picture .extensionsMenuExtensionButton');
        sendButton.removeClass('fa-image');
        sendButton.addClass('fa-hourglass-half');
    }
    catch (error) {
        console.log(error);
    }
}

/**
 * Wraps a caption with a message template.
 * @param {string} caption Raw caption
 * @returns {Promise<string>} Wrapped caption
 */
async function wrapCaptionTemplate(caption) {
    let template = extension_settings.caption.template || TEMPLATE_DEFAULT;

    if (!/{{caption}}/i.test(template)) {
        console.warn('Poka-yoke: Caption template does not contain {{caption}}. Appending it.');
        template += ' {{caption}}';
    }

    let messageText = substituteParamsExtended(template, { caption: caption });

    if (extension_settings.caption.refine_mode) {
        messageText = await Popup.show.input(
            'Review and edit the generated caption:',
            'Press "Cancel" to abort the caption sending.',
            messageText,
            { rows: 8, okButton: 'Send' });

        if (!messageText) {
            throw new Error('User aborted the caption sending.');
        }
    }

    return messageText;
}

/**
 * Appends caption to an existing message.
 * @param {ChatMessage} message Message data
 * @param {number} mediaIndex Index of the image to caption
 * @returns {Promise<void>}
 */
async function captionExistingMessage(message, mediaIndex) {
    if (!Array.isArray(message?.extra?.media) || message.extra.media.length === 0) {
        return;
    }

    if (mediaIndex === undefined || isNaN(mediaIndex) || mediaIndex < 0 || mediaIndex >= message.extra.media.length) {
        mediaIndex = 0;
    }

    const mediaAttachment = message.extra.media[mediaIndex];

    if (!mediaAttachment || !mediaAttachment.url || mediaAttachment.type === MEDIA_TYPE.AUDIO) {
        return;
    }

    if (mediaAttachment.type === MEDIA_TYPE.VIDEO && !isVideoCaptioningAvailable()) {
        throw new Error('Captioning videos is not supported for the current source.');
    }

    const imageData = await fetch(mediaAttachment.url);
    const blob = await imageData.blob();
    const fileName = mediaAttachment.url.split('/').pop().split('?')[0] || 'image.jpg';
    const file = new File([blob], fileName, { type: blob.type });
    const caption = await getCaptionForFile(file, null, true);

    if (!caption) {
        console.warn('Failed to generate a caption for the image.');
        return;
    }

    const wrappedCaption = await wrapCaptionTemplate(caption);

    const messageText = String(message.mes).trim();

    if (!messageText) {
        message.extra.inline_image = false;
        message.mes = wrappedCaption;
        mediaAttachment.title = wrappedCaption;
        mediaAttachment.captioned = true;
    } else {
        message.extra.inline_image = true;
        mediaAttachment.append_title = true;
        mediaAttachment.title = wrappedCaption;
        mediaAttachment.captioned = true;
    }
}

/**
 * Sends a captioned message to the chat.
 * @param {string} caption Caption text
 * @param {string} image Image URL
 * @param {string} mimeType Image MIME type
 * @returns {Promise<void>}
 */
async function sendCaptionedMessage(caption, image, mimeType) {
    const messageText = await wrapCaptionTemplate(caption);

    const context = getContext();

    /** @type {MediaAttachment} */
    const mediaAttachment = {
        url: image,
        type: MEDIA_TYPE.getFromMime(mimeType) || MEDIA_TYPE.IMAGE,
        title: messageText,
        captioned: true,
        source: MEDIA_SOURCE.CAPTIONED,
    };
    /** @type {ChatMessage} */
    const message = {
        name: context.name1,
        is_user: true,
        send_date: getMessageTimeStamp(),
        mes: messageText,
        extra: {
            media: [mediaAttachment],
            media_display: MEDIA_DISPLAY.GALLERY,
            media_index: 0,
            inline_image: !!extension_settings.caption.show_in_chat,
        },
    };
    chat_metadata['tainted'] = true;
    context.chat.push(message);
    const messageId = context.chat.length - 1;
    await eventSource.emit(event_types.MESSAGE_SENT, messageId);
    context.addOneMessage(message);
    await eventSource.emit(event_types.USER_MESSAGE_RENDERED, messageId);
    await context.saveChat();
    setTimeout(() => context.scrollOnMediaLoad(), debounce_timeout.short);
}

/**
 * Generates a caption for an image using a selected source.
 * @param {string} base64Img Base64 encoded image without the data:image/...;base64, prefix
 * @param {string} fileData Base64 encoded image with the data:image/...;base64, prefix
 * @param {string} externalPrompt Caption prompt
 * @returns {Promise<{caption: string}>} Generated caption
 */
async function doCaptionRequest(base64Img, fileData, externalPrompt) {
    switch (extension_settings.caption.source) {
        case 'local':
            return await captionLocal(base64Img);
        case 'extras':
            return await captionExtras(base64Img);
        case 'horde':
            return await captionHorde(base64Img);
        case 'multimodal':
            return await captionMultimodal(fileData, externalPrompt);
        default:
            throw new Error('Unknown caption source.');
    }
}

/**
 * Generates a caption for an image using Extras API.
 * @param {string} base64Img Base64 encoded image without the data:image/...;base64, prefix
 * @returns {Promise<{caption: string}>} Generated caption
 */
async function captionExtras(base64Img) {
    if (!modules.includes('caption')) {
        throw new Error('No captioning module is available.');
    }

    const url = new URL(getApiUrl());
    url.pathname = '/api/caption';

    const apiResult = await doExtrasFetch(url, {
        method: 'POST',
        headers: {
            'Content-Type': 'application/json',
            'Bypass-Tunnel-Reminder': 'bypass',
        },
        body: JSON.stringify({ image: base64Img }),
    });

    if (!apiResult.ok) {
        throw new Error('Failed to caption image via Extras.');
    }

    const data = await apiResult.json();
    return data;
}

/**
 * Generates a caption for an image using a local model.
 * @param {string} base64Img Base64 encoded image without the data:image/...;base64, prefix
 * @returns {Promise<{caption: string}>} Generated caption
 */
async function captionLocal(base64Img) {
    const apiResult = await fetch('/api/extra/caption', {
        method: 'POST',
        headers: getRequestHeaders(),
        body: JSON.stringify({ image: base64Img }),
    });

    if (!apiResult.ok) {
        throw new Error('Failed to caption image via local pipeline.');
    }

    const data = await apiResult.json();
    return data;
}

/**
 * Generates a caption for an image using a Horde model.
 * @param {string} base64Img Base64 encoded image without the data:image/...;base64, prefix
 * @returns {Promise<{caption: string}>} Generated caption
 */
async function captionHorde(base64Img) {
    const apiResult = await fetch('/api/horde/caption-image', {
        method: 'POST',
        headers: getRequestHeaders(),
        body: JSON.stringify({ image: base64Img }),
    });

    if (!apiResult.ok) {
        throw new Error('Failed to caption image via Horde.');
    }

    const data = await apiResult.json();
    return data;
}

/**
 * Generates a caption for an image using a multimodal model.
 * @param {string} base64Img Base64 encoded image with the data:image/...;base64, prefix
 * @param {string} externalPrompt Caption prompt
 * @returns {Promise<{caption: string}>} Generated caption
 */
async function captionMultimodal(base64Img, externalPrompt) {
    let prompt = externalPrompt || extension_settings.caption.prompt || PROMPT_DEFAULT;

    if (!externalPrompt && extension_settings.caption.prompt_ask) {
        const customPrompt = await callGenericPopup('Enter a comment or question:', POPUP_TYPE.INPUT, prompt, { rows: 4 });
        if (!customPrompt) {
            throw new Error('User aborted the caption sending.');
        }
        prompt = String(customPrompt).trim();
    }

    const caption = await getMultimodalCaption(base64Img, prompt);
    return { caption };
}

/**
 * Handles the image selection event.
 * @param {Event} e Input event
 * @param {string} prompt Caption prompt
 * @param {boolean} quiet Suppresses sending a message
 * @returns {Promise<string>} Generated caption
 */
async function onSelectImage(e, prompt, quiet) {
    if (!(e.target instanceof HTMLInputElement)) {
        return '';
    }

    const file = e.target.files[0];
    const form = e.target.form;

    if (!file || !(file instanceof File)) {
        form && form.reset();
        return '';
    }

    const caption = await getCaptionForFile(file, prompt, quiet);
    form && form.reset();
    return caption;
}

/**
 * Gets a caption for an image file.
 * @param {File} file Input file
 * @param {string} prompt Caption prompt
 * @param {boolean} quiet Suppresses sending a message
 * @returns {Promise<string>} Generated caption
 */
async function getCaptionForFile(file, prompt, quiet) {
    try {
        if (file.type.startsWith('video/') && !isVideoCaptioningAvailable()) {
            throw new Error('Video captioning is not available for the current source.');
        }

        setSpinnerIcon();
        const context = getContext();
        const fileData = await getBase64Async(await ensureImageFormatSupported(file));
        const extension = getFileExtension(file);
        const base64Data = fileData.split(',')[1];
        const { caption } = await doCaptionRequest(base64Data, fileData, prompt);
        if (!quiet) {
            const imagePath = await saveBase64AsFile(base64Data, context.name2, '', extension);
            await sendCaptionedMessage(caption, imagePath, file.type);
        }
        return caption;
    }
    catch (error) {
        const errorMessage = error.message || 'Unknown error';
        toastr.error(errorMessage, 'Failed to caption');
        console.error(error);
        return '';
    }
    finally {
        setImageIcon();
    }
}

function onRefineModeInput() {
    extension_settings.caption.refine_mode = $('#caption_refine_mode').prop('checked');
    saveSettingsDebounced();
}

/**
 * Callback for the /caption command.
 * @param {object} args Named parameters
 * @param {string} prompt Caption prompt
 */
async function captionCommandCallback(args, prompt) {
    const quiet = isTrueBoolean(args?.quiet);
    const messageId = args?.mesId ?? args?.id;
    const index = Number(args?.index ?? 0);

    if (!isNaN(Number(messageId))) {
        /** @type {ChatMessage} */
        const message = getContext().chat[messageId];
        if (Array.isArray(message?.extra?.media) && message.extra.media.length > 0) {
            try {
                const mediaAttachment = message.extra.media[index] || message.extra.media[0];
                if (!mediaAttachment || !mediaAttachment.url) {
                    toastr.error('The specified message does not contain an image.');
                    return '';
                }
                if (mediaAttachment.type === MEDIA_TYPE.AUDIO) {
                    toastr.error('The specified media is an audio file. Captioning audio files is not supported.');
                    return '';
                }
                if (mediaAttachment.type === MEDIA_TYPE.VIDEO && !isVideoCaptioningAvailable()) {
                    toastr.error('The specified media is a video. Captioning videos is not supported for the current source.');
                    return '';
                }
                const fetchResult = await fetch(mediaAttachment.url);
                const blob = await fetchResult.blob();
                const fileName = mediaAttachment.url.split('/').pop().split('?')[0] || 'image.jpg';
                const file = new File([blob], fileName, { type: blob.type });
                return await getCaptionForFile(file, prompt, quiet);
            } catch (error) {
                toastr.error('Failed to get image from the message. Make sure the image is accessible.');
                return '';
            }
        }
    }

    return new Promise(resolve => {
        const input = document.createElement('input');
        input.type = 'file';
        input.accept = 'image/*,video/*';
        input.onchange = async (e) => {
            const caption = await onSelectImage(e, prompt, quiet);
            resolve(caption);
        };
        input.oncancel = () => resolve('');
        input.click();
    });
}

/**
 * Checks if video captioning is available for the current source.
 * @returns {boolean} True if video captioning is supported for the current source.
 */
function isVideoCaptioningAvailable() {
    if (extension_settings.caption.source !== 'multimodal') {
        return false;
    }

    return ['google', 'vertexai', 'zai'].includes(extension_settings.caption.multimodal_api);
}

jQuery(async function () {
    function addSendPictureButton() {
        const sendButton = $(`
        <div id="send_picture" class="list-group-item flex-container flexGap5">
            <div class="fa-solid fa-image extensionsMenuExtensionButton"></div>
            <span data-i18n="Generate Caption">Generate Caption</span>
        </div>`);

        $('#caption_wand_container').append(sendButton);
        $(sendButton).on('click', () => {
            const hasCaptionModule = (() => {
                const settings = extension_settings.caption;

                // Handle non-multimodal sources
                if (settings.source === 'extras' && modules.includes('caption')) return true;
                if (settings.source === 'local' || settings.source === 'horde') return true;

                // Handle multimodal sources
                if (settings.source === 'multimodal') {
                    const api = settings.multimodal_api;
                    const altEndpointEnabled = settings.alt_endpoint_enabled;
                    const altEndpointUrl = settings.alt_endpoint_url;

                    // APIs that support reverse proxy
                    const reverseProxyApis = {
                        'openai': SECRET_KEYS.OPENAI,
                        'mistral': SECRET_KEYS.MISTRALAI,
                        'google': SECRET_KEYS.MAKERSUITE,
                        'vertexai': SECRET_KEYS.VERTEXAI,
                        'anthropic': SECRET_KEYS.CLAUDE,
                        'xai': SECRET_KEYS.XAI,
                    };

                    if (reverseProxyApis[api]) {
                        if (secret_state[reverseProxyApis[api]] || settings.allow_reverse_proxy) {
                            return true;
                        }
                    }

                    const chatCompletionApis = {
                        'openrouter': SECRET_KEYS.OPENROUTER,
                        'groq': SECRET_KEYS.GROQ,
                        'cohere': SECRET_KEYS.COHERE,
                        'aimlapi': SECRET_KEYS.AIMLAPI,
                        'moonshot': SECRET_KEYS.MOONSHOT,
                        'nanogpt': SECRET_KEYS.NANOGPT,
                        'chutes': SECRET_KEYS.CHUTES,
                        'electronhub': SECRET_KEYS.ELECTRONHUB,
                        'zai': SECRET_KEYS.ZAI,
                    };

                    if (chatCompletionApis[api] && secret_state[chatCompletionApis[api]]) {
                        return true;
                    }

                    const textCompletionApis = {
                        'ollama': textgen_types.OLLAMA,
                        'llamacpp': textgen_types.LLAMACPP,
                        'ooba': textgen_types.OOBA,
                        'koboldcpp': textgen_types.KOBOLDCPP,
                        'vllm': textgen_types.VLLM,
                    };

                    if (textCompletionApis[api] && altEndpointEnabled && altEndpointUrl) {
                        return true;
                    }

                    if (textCompletionApis[api] && !altEndpointEnabled && textgenerationwebui_settings.server_urls[textCompletionApis[api]]) {
                        return true;
                    }

                    // Custom API doesn't need additional checks
                    if (api === 'custom' || api === 'pollinations') {
                        return true;
                    }
                }

                return false;
            })();

            if (!hasCaptionModule) {
                toastr.error('Choose other captioning source in the extension settings.', 'Captioning is not available');
                return;
            }

            $('#img_file').trigger('click');
        });
    }
    function addPictureSendForm() {
        const imgInput = document.createElement('input');
        imgInput.type = 'file';
        imgInput.id = 'img_file';
        imgInput.accept = 'image/*,video/*';
        imgInput.hidden = true;
        imgInput.addEventListener('change', (e) => onSelectImage(e, '', false));
        const imgForm = document.createElement('form');
        imgForm.id = 'img_form';
        imgForm.appendChild(imgInput);
        imgForm.hidden = true;
        $('#form_sheld').append(imgForm);
    }
    async function switchMultimodalBlocks() {
        await addRemoteEndpointModels();
        const isMultimodal = extension_settings.caption.source === 'multimodal';
        if (!extension_settings.caption.multimodal_model) {
            const dropdown = $('#caption_multimodal_model');
            const options = dropdown.find(`option[data-type="${extension_settings.caption.multimodal_api}"]`);
            extension_settings.caption.multimodal_model = String(options.first().val());
        }
        $('#caption_multimodal_block').toggle(isMultimodal);
        $('#caption_prompt_block').toggle(isMultimodal);
        $('#caption_multimodal_api').val(extension_settings.caption.multimodal_api);
        $('#caption_multimodal_model').val(extension_settings.caption.multimodal_model);
        $('#caption_multimodal_block [data-type]').each(function () {
            const type = $(this).data('type');
            const types = type.split(',');
            $(this).toggle(types.includes(extension_settings.caption.multimodal_api));
        });
    }
    async function addSettings() {
        const html = await renderExtensionTemplateAsync('caption', 'settings', { TEMPLATE_DEFAULT, PROMPT_DEFAULT });
        $('#caption_container').append(html);
    }

    async function addRemoteEndpointModels() {
        async function processEndpoint(api, url) {
            const dropdown = document.getElementById('caption_multimodal_model');
            if (!(dropdown instanceof HTMLSelectElement)) {
                return;
            }
            if (extension_settings.caption.source !== 'multimodal' || extension_settings.caption.multimodal_api !== api) {
                return;
            }
            const options = Array.from(dropdown.options);
            const response = await fetch(url, {
                method: 'POST',
                headers: getRequestHeaders({ omitContentType: true }),
            });
            if (!response.ok) {
                return;
            }
            const modelIds = await response.json();
            if (Array.isArray(modelIds) && modelIds.length > 0) {
                modelIds.sort().forEach((modelId) => {
                    if (!modelId || typeof modelId !== 'string' || options.some(o => o.value === modelId)) {
                        return;
                    }
                    const option = document.createElement('option');
                    option.value = modelId;
                    option.textContent = modelId;
                    option.dataset.type = api;
                    dropdown.add(option);
                });
            }
        }

        await processEndpoint('openrouter', '/api/openrouter/models/multimodal');
        await processEndpoint('aimlapi', '/api/backends/chat-completions/multimodal-models/aimlapi');
        await processEndpoint('pollinations', '/api/backends/chat-completions/multimodal-models/pollinations');
        await processEndpoint('nanogpt', '/api/backends/chat-completions/multimodal-models/nanogpt');
        await processEndpoint('chutes', '/api/backends/chat-completions/multimodal-models/chutes');
        await processEndpoint('electronhub', '/api/backends/chat-completions/multimodal-models/electronhub');
        await processEndpoint('mistral', '/api/backends/chat-completions/multimodal-models/mistral');
        await processEndpoint('xai', '/api/backends/chat-completions/multimodal-models/xai');
    }

    await addSettings();
    addPictureSendForm();
    addSendPictureButton();
    setImageIcon();
    migrateSettings();
    await switchMultimodalBlocks();

    $('#caption_refine_mode').prop('checked', !!(extension_settings.caption.refine_mode));
    $('#caption_allow_reverse_proxy').prop('checked', !!(extension_settings.caption.allow_reverse_proxy));
    $('#caption_prompt_ask').prop('checked', !!(extension_settings.caption.prompt_ask));
    $('#caption_auto_mode').prop('checked', !!(extension_settings.caption.auto_mode));
    $('#caption_source').val(extension_settings.caption.source);
    $('#caption_prompt').val(extension_settings.caption.prompt);
    $('#caption_template').val(extension_settings.caption.template);
    $('#caption_refine_mode').on('input', onRefineModeInput);
    $('#caption_source').on('change', async () => {
        extension_settings.caption.source = String($('#caption_source').val());
        await switchMultimodalBlocks();
        saveSettingsDebounced();
    });
    $('#caption_prompt').on('input', () => {
        extension_settings.caption.prompt = String($('#caption_prompt').val());
        saveSettingsDebounced();
    });
    $('#caption_template').on('input', () => {
        extension_settings.caption.template = String($('#caption_template').val());
        saveSettingsDebounced();
    });
    $('#caption_allow_reverse_proxy').on('input', () => {
        extension_settings.caption.allow_reverse_proxy = $('#caption_allow_reverse_proxy').prop('checked');
        saveSettingsDebounced();
    });
    $('#caption_prompt_ask').on('input', () => {
        extension_settings.caption.prompt_ask = $('#caption_prompt_ask').prop('checked');
        saveSettingsDebounced();
    });
    $('#caption_auto_mode').on('input', () => {
        extension_settings.caption.auto_mode = !!$('#caption_auto_mode').prop('checked');
        saveSettingsDebounced();
    });
    $('#caption_ollama_pull').on('click', (e) => {
        const selectedModel = extension_settings.caption.multimodal_model;
        const staticModels = { 'ollama_current': textgenerationwebui_settings.ollama_model, 'ollama_custom': extension_settings.caption.ollama_custom_model };
        const presetModel = staticModels[selectedModel] || selectedModel;
        e.preventDefault();
        $('#ollama_download_model').trigger('click');
        $('.popup .popup-input').val(presetModel);
    });
    $('#caption_multimodal_api').on('change', async () => {
        const api = String($('#caption_multimodal_api').val());
        extension_settings.caption.multimodal_api = api;
        extension_settings.caption.multimodal_model = '';
        await switchMultimodalBlocks();
        saveSettingsDebounced();
    });
    $('#caption_multimodal_model').on('change', () => {
        extension_settings.caption.multimodal_model = String($('#caption_multimodal_model').val());
        saveSettingsDebounced();
    });
    $('#caption_altEndpoint_url').val(extension_settings.caption.alt_endpoint_url).on('input', () => {
        extension_settings.caption.alt_endpoint_url = String($('#caption_altEndpoint_url').val());
        saveSettingsDebounced();
    });
    $('#caption_altEndpoint_enabled').prop('checked', !!(extension_settings.caption.alt_endpoint_enabled)).on('input', () => {
        extension_settings.caption.alt_endpoint_enabled = !!$('#caption_altEndpoint_enabled').prop('checked');
        saveSettingsDebounced();
    });
    $('#caption_show_in_chat').prop('checked', !!(extension_settings.caption.show_in_chat)).on('input', () => {
        extension_settings.caption.show_in_chat = !!$('#caption_show_in_chat').prop('checked');
        saveSettingsDebounced();
    });
    $('#caption_ollama_custom_model').val(extension_settings.caption.ollama_custom_model || '').on('input', () => {
        extension_settings.caption.ollama_custom_model = String($('#caption_ollama_custom_model').val()).trim();
        saveSettingsDebounced();
    });
    $('#caption_refresh_models').on('click', async () => {
        extension_settings.caption.multimodal_model = '';
        await switchMultimodalBlocks();
        saveSettingsDebounced();
    });

    const onMessageEvent = async (/** @type {number} */ messageId) => {
        if (!extension_settings.caption.auto_mode) {
            return;
        }

        const message = getContext().chat[messageId];
        if (Array.isArray(message?.extra?.media) && message.extra.media.length > 0) {
            for (let mediaIndex = 0; mediaIndex < message.extra.media.length; mediaIndex++) {
                const mediaAttachment = message.extra.media[mediaIndex];
                if (mediaAttachment.type === MEDIA_TYPE.VIDEO && !isVideoCaptioningAvailable()) {
                    continue;
                }
                if (mediaAttachment.type === MEDIA_TYPE.AUDIO) {
                    continue;
                }
                // Skip already captioned images and non-uploaded (generated, etc.) images
                if (mediaAttachment.source !== MEDIA_SOURCE.UPLOAD || mediaAttachment.captioned) {
                    continue;
                }
                try {
                    await captionExistingMessage(message, mediaIndex);
                } catch (e) {
                    console.error(`Auto-captioning failed for message ID ${messageId}, media index ${mediaIndex}`, e);
                    continue;
                }
            }
        }
    };

    eventSource.on(event_types.MESSAGE_SENT, onMessageEvent);
    eventSource.on(event_types.MESSAGE_FILE_EMBEDDED, onMessageEvent);

    $(document).on('click', '.mes_img_caption', async function () {
        const animationClass = 'fa-fade';
        const messageBlock = $(this).closest('.mes');
        const mediaContainer = $(this).closest('.mes_media_container');
        const messageMedia = mediaContainer.find('.mes_img, .mes_video');
        if (messageMedia.hasClass(animationClass)) return;
        messageMedia.addClass(animationClass);
        try {
            const messageId = Number(messageBlock.attr('mesid'));
            const mediaIndex = Number(mediaContainer.attr('data-index'));
            const data = getContext().chat[messageId];
            await captionExistingMessage(data, mediaIndex);
            appendMediaToMessage(data, messageBlock, SCROLL_BEHAVIOR.KEEP);
            await saveChatConditional();
        } catch (e) {
            console.error('Message image recaption failed', e);
            toastr.error(e.message || 'Unknown error', 'Failed to caption');
        } finally {
            messageMedia.removeClass(animationClass);
        }
    });

    SlashCommandParser.addCommandObject(SlashCommand.fromProps({
        name: 'caption',
        callback: captionCommandCallback,
        returns: 'caption',
        namedArgumentList: [
            new SlashCommandNamedArgument(
                'quiet', 'suppress sending a captioned message', [ARGUMENT_TYPE.BOOLEAN], false, false, 'false',
            ),
            SlashCommandNamedArgument.fromProps({
                name: 'mesId',
                description: 'get image from a message with this ID',
                typeList: [ARGUMENT_TYPE.NUMBER],
                enumProvider: commonEnumProviders.messages(),
            }),
            SlashCommandNamedArgument.fromProps({
                name: 'index',
                description: 'index of the image in the message to caption (starting from 0)',
                typeList: [ARGUMENT_TYPE.NUMBER],
                enumProvider: commonEnumProviders.messageMedia(),
            }),
        ],
        unnamedArgumentList: [
            new SlashCommandArgument(
                'prompt', [ARGUMENT_TYPE.STRING], false,
            ),
        ],
        helpString: `
            <div>
                Caption an image with an optional prompt and passes the caption down the pipe.
            </div>
            <div>
                Only multimodal sources support custom prompts.
            </div>
            <div>
                Provide a message ID to get an image from a message instead of uploading one.
            </div>
            <div>
                Set the "quiet" argument to true to suppress sending a captioned message, default: false.
            </div>
        `,
    }));

    document.body.classList.add('caption');
});