Library and element tools

All calls use the authenticated OAuth account. Listing returns metadata; listing a model or media item does not prove it is ready for generation.

​

Personal library

​

list_personal_models

All inputs are optional: pageNumber × pageSize must not exceed 1000. Results are newest first and contain items, hasMore, pageNumber, pageSize. Each item includes modelId, nullable name, usage, isPublic, and nullable createdTime (ISO 8601). The tool returns no model files or URLs. Supported usages are chat, tweet, portrait, portraitSim, textSpeech, imageVideoHigh, imageVideoBest, imageVideoUltra, talkingAvatarUltra, reStyle, promptCameraMovement, reEverything, reShot.

​

list_personal_medias

All inputs are optional: Results are newest first and contain items, hasMore, nullable nextCursorId, and pageSize. Continue with previousId: nextCursorId while keeping filters unchanged. An INVALID_CURSOR error requires restarting from the first page. Each item contains mediaId, nullable modelId, contentId, title, quality, createdTime (ISO 8601), plus mediaType and fileType. The tool returns no media URLs; retrieve completed generation outputs with get_generation_result.

Supported mediaTypes

none, chat, tweet, image, upscale, faceSwap, song, imageVideo, videoVideo, portraitMaster, businessImage, textSpeech, vton, simVtonPoor, simProduct, ideaText, textOptimization, mixVideo, textVideo, imageRemoveBackground, imageEnlarge, audioVideo, videoLipSync, videoTalkingAvatar, clothingPreprocess, voiceDesign, imageEdit, cloneImage, cloneVideo, cameraMovement, mixVideoV2, videoUpscale, mixVideoV2ScriptOptimize, mixVideoV3Storyboard, mixVideoV3, followMovement, subtitleVideoRender, bgmVideo, revideoReferenceImage, extendVideo, productPreprocess, reStyle, reEverything, reProduct, reBackground, reShot, reMotion, textMusic.

​

Character models

​

create_model

Creates one portrait SIM model from exactly one owned image sourceMediaId or canonical uploadedImage: { "url": "..." } from media preparation. Account model limits and preprocessing still apply. The receipt contains modelId, taskId, status, and terminal.

​

get_model_status

Required: nonblank modelId, at most 200 characters. Reads your own active portrait SIM model and its root model task. Returns modelId, taskId, status, terminal, ready, and nullable error. ready requires successful completion. Terminal failures expose sanitized MODEL_FAILED details. Retrying this read does not create another model. get_generation_result reads content tasks; use get_model_status for model readiness. Model errors include MODEL_NOT_FOUND (unavailable or not owned), MODEL_POLICY_REJECTED (privacy/model-limit policy), MODEL_READ_FAILED (retry the same read), and MODEL_SUBMISSION_UNCONFIRMED (reconcile before any new submission).

​

Elements

Elements can be discovered from your library or the community, created from uploads or existing media/models, and bound in generation through elementIds. The categories are Character, Product, Background, Style, Others. IDs are positive signed 64-bit decimal strings.

​

list_elements

Optional inputs: scope (personal, default; or community), category, fileType (image, audio, video), pageNumber (positive signed 32-bit integer, default 1), pageSize (1–50, default 20). Returns items, hasMore, pageNumber, pageSize. Summaries contain elementId, isDerived, ownedByCaller, canManage, name, nullable category, description, previewUrl, plus isPublic, fileType, and optional width/height. Preview URLs are temporary. Lists contain ordinary elements. Derived media/model elements are excluded from list results; their returned IDs can still be read and referenced.

​

get_element

Required: elementId. Reads an owned or public accessible element, including derived elements. Returns the summary plus resources. Each resource contains fileType, temporary url, and optional previewUrl, width, height, durationMs.

​

create_element

Required: nonblank name and exactly one source: mediaId, modelId, images, audios, or videos. Optional: category (default Others), string description, isPublic (uploaded ordinary elements only; default false). Upload arrays must be nonempty and contain objects with only the canonical url from your own import_file or complete_file_upload. The server resolves media metadata. Media/model-derived elements inherit source ownership and visibility, may reuse an existing derived element, and reject an isPublic override. Derived elements can be read and used in generation, but cannot be edited, published, hidden, or deleted with MCP management tools. Returns a write receipt: elementId, isDerived, ownedByCaller, canManage, name, category, nullable description, isPublic. Use get_element for fresh resource URLs. Moderation/processing can continue after creation, so a receipt does not guarantee ready generation references.

​

update_element

Required: elementId and at least one updated field. Optional update fields: nonblank name, category, string description, and at most one nonempty uploaded images, audios, or videos array in the same shape as creation. Only owned ordinary elements are manageable. An empty description clears it. A resource array replaces the complete resource set. Null/no-op updates are rejected. Returns the same public write-receipt fields as creation.

​

set_element_visibility

Required: elementId, boolean isPublic. Only owned ordinary elements can be changed; private visibility requires account eligibility. Returns elementId, isDerived, ownedByCaller, canManage, isPublic.

​

delete_element

Required: elementId. Deletes an owned ordinary element. Returns elementId, isDerived, ownedByCaller, canManage: false, deleted: true. Element failures include INVALID_ARGUMENTS, ELEMENT_NOT_ACCESSIBLE, ELEMENT_NOT_MANAGEABLE, ELEMENT_VISIBILITY_REJECTED, and ELEMENT_READ_FAILED. Reads can be retried; retries of mutations need reconciliation first.