asset_import
ChatGPTDownload a user-selected image, video, audio or 3D attachment, validate it and save an owned Nova reference. Maximum 64 MB. Voice samples can feed audioImport. Never persist the temporary download_url.
Overview
Create AI images, short videos, and creative variations from a brief or your reference media. Explore visual directions, animate product images, and refine the result in a reusable Nova Creative workflow. Reopen saved projects, change selected steps, and continue from the same editable workflow. Review available models and estimated usage, confirm generation in the Nova card, follow progress, see previews, and download finished files with a link back to your project. Requires a connected Nova Creative account. Generation uses the models and usage available to that account.
Tools
asset_list
ChatGPTList the connected account's owned Assets, including files uploaded directly to Nova from a computer. These files do not need to be attached to chat again. Call this before asking for missing chat attachments. Return stable asset ids/URLs and names, not image bytes. Use the returned references in incremental workflow operations. Remote MCP cannot read files still only on the user's disk; select/drop a batch into Nova first.
asset_upload
ChatGPTSave one user-selected local image, video, audio or 3D reference from base64 bytes, at most 16 MB. The client reads the selected file; the remote server never reads disk paths. No generation or charge. Use the returned owned asset in incremental workflow operations.
generation_catalog
ChatGPTList configured image, video, audio, 3D and lyrics providers, model ids and supported parameters. No provider secrets or paid calls. Use gpu_templates and gpu_catalog for GPU models. For video presentations with animated slides/text or explicit Remotion requests, read workflow_node_reference with query=remotionVideo; its fixed scene renderer is a workflow node, not a model in this catalog.
gpu_avatar_runtime
ChatGPTRead talking-avatar runtime readiness on an owned GPU. Shared cards cannot be reconfigured.
gpu_avatar_runtime_install
ChatGPTProvision one supported avatar runtime on an owned GPU only after explicit approval of its large dependencies and ongoing GPU time. Requires workflows.generate and confirm_install=true. Returns a job handle; no generation.
gpu_avatar_runtime_status
ChatGPTPoll one avatar-runtime installation job on an owned GPU. No new installation or generation.
gpu_catalog
ChatGPTRead GPU model ids, download sizes, VRAM requirements and license metadata; no credentials or download URLs. kind: image/video/llm or empty.
gpu_nodes
ChatGPTList owned/available GPU nodes, live memory/workloads and optional installed model ids. Set an explicit online gpuNodeId before reviewing Avatar/Comfy workflows. No rental or installation.
gpu_server_model_install
ChatGPTAdd a catalog model to the exact current rental after approval of its download size and hourly rate. Requires workflows.generate, confirm_install=true and the current credits/hour. Poll gpu_server_state; never imply downloading is free GPU time.
gpu_server_model_remove
ChatGPTRemove this catalog model/files from the exact owned rental after explicit confirmation. Does not remove media or stop the server. Requires confirm_remove=true.
gpu_server_quotes
ChatGPTRead GPU tiers, card types, disk/idle options and hourly rates for existing-server review and settings. GPU rental is unavailable through this plugin. Does not rent or charge.
gpu_server_settings
ChatGPTUpdate this exact rental's idle shutdown only after explicit confirmation. idle_off_minutes=0 disables auto-off and continues billing until stopped. Use options from gpu_server_quotes; requires confirm_change=true.
gpu_server_start
ChatGPTRent one owned GPU after the user approves the quoted card, disk, idle shutdown, prepaid first hour and hourly rate. Read gpu_server_state and gpu_server_quotes first. Requires workflows.generate and confirm_spend=true; changed card or a price above the ceiling is rejected before charging. Does not install models or generate. Poll gpu_server_state after submission or an uncertain response; never blindly repeat the start.
gpu_server_state
ChatGPTRead My server ids, readiness, model progress, charges and idle shutdown. Optionally inspect an exact owned managed_rental_id to verify completed versus teardown-pending provider deletion. off is not permission to rent again.
gpu_server_terminate
ChatGPTPermanently delete the exact owned cloud GPU server and its attached disk. Read gpu_server_state for expected_managed_rental_id and get explicit approval of permanent deletion before setting confirm_stop=true. This is not a pause; local GPUs and Nova media are preserved. Requires workflows.write. Charges stop only after provider deletion completes; verify the rental with gpu_server_state. Stale targets are rejected.
gpu_template_import
ChatGPTImport a user-selected ComfyUI API-format graph as a NEW shared catalog template (designer/admin, workflows.write). Returns the real template id, inputs and defaults for workflow authoring. Never overwrites templates, reads disk paths, rents a GPU, downloads weights or generates. Discover existing gpu_templates first; use the returned id and an explicit gpuNodeId, then preflight generation separately.
gpu_templates
ChatGPTRead Comfy template ids, required bundles, VRAM, input handles, runtime voice/reference sockets and parameter defaults. kind: image/video or empty.
nova_capabilities
ChatGPTRead this deployment's feature version, media/upload limits, GPU role access and generation grant. Personal tokens cannot generate. With workflows.generate, use workflow_generate after the user's chat approval of a fresh preflight; the native card is optional. Use before relying on new features.
nova_status
ChatGPTCheck the authenticated Nova Creative MCP connection. Does not call external providers.
object_editor_asset
ChatGPTUI-only: load an owned editor image or mask through the MCP bridge. Image bytes are private UI metadata; no browser cookies or cross-origin pixel access are required.
object_editor_open
ChatGPTOpen the interactive Nova Studio image editor in the chat's side panel for replace/change/remove requests. The user selects any object or layer on the image or in the layer tree, describes changes and generates a preview after reviewing usage. Reuses an existing image editor, or incrementally adds one for an owned source_asset_id (can select the sole saved image output). Does not analyze, generate or charge on opening. Unanalyzed images offer Analyze layers with a usage review. Use this instead of only patching a generator prompt for object-specific changes.
object_editor_run_outputs
ChatGPTUI-only: retrieve this editor's durable scene and render outputs and submitted edit snapshot from an owned run. Layer details remain private UI metadata.
object_editor_upload
ChatGPTUI-only: validate and save the user's PNG/JPEG/WebP replacement reference in the connected Nova account. Does not generate or charge.
voice_search
ChatGPTFind voices for an elevenLabs node (text to speech with any ElevenLabs voice): the public Voice Library (about 20,000 community voices in every language), ElevenLabs default voices (source=default) or the user's own voices (source=mine, only with their personal ElevenLabs key). Describe the voice in query and narrow it with language, gender, age, accent or use_case. Put the chosen voiceId and name into the node's data.voiceId and data.voiceName. Read-only, no charges.
workflow_apply_operations
ChatGPTAtomically apply 1 to 200 incremental addNode, updateNode, removeNode, addEdge, or removeEdge operations. Whole-graph replacement is not supported.
workflow_create
ChatGPTCreate a new Nova workflow. The backend validates node types, ports, cycles, security and ownership.
workflow_duplicate
ChatGPTSave an independent copy of a workflow including ALL current node contents and generation results, without copying run history. Reuses durable assets; no uploads or generation required. Use for Save as / duplicate requests instead of rebuilding nodes or archiving outputs manually.
workflow_generate
ChatGPTStart generation after the user explicitly approves the saved workflow's scope and estimated usage in chat. Requires a fresh workflow_preflight and workflows.generate OAuth scope; no card or private confirmation token is needed. Uses the connected account's credits or configured credentials. Unknown-cost reviews are rejected. Reuse request_id when retrying; one reviewed snapshot can execute only once. Follow the returned run id for progress and results.
workflow_get
ChatGPTLoad one user-owned workflow, including its graph and updatedAt concurrency token.
workflow_list
ChatGPTList saved workflows owned by the selected Nova user, newest first.
workflow_node_reference
ChatGPTRead the authoritative Nova node and port reference. Use before creating or wiring unfamiliar node types. For video presentations with animated slides, text or statistics, or an explicit Remotion request, query remotionVideo for selection guidance, workflow placement, scene plans and media inputs. Write the plan with the current model and use mode=plan. Generic video/ad requests do not automatically need Remotion; use videoGenerator for new realistic action and editing nodes for simple edits.
workflow_preflight
ChatGPTValidate an owned saved workflow and review its exact configuration, provider-call bound and estimated Nova credits. Does not generate or charge. After the user approves this scope and estimate, call workflow_generate with the preflight id and exact estimatedCredits, or use the optional Nova card. Expires after five minutes; editing the project invalidates it.
workflow_rename
ChatGPTAtomically rename a workflow owned by the selected Nova user without changing its graph or results.
workflow_run
ChatGPTStart the reviewed generation after the user's explicit confirmation in the Nova card. Uses the connected account's existing credits or configured credentials. The UI supplies the private confirmation_token. Reuse request_id when retrying this preflight to avoid another charge.
workflow_run_cancel
ChatGPTRequest cancellation of an owned queued/running generation. Already completed provider calls may have incurred charges. Preserve existing files. Check status until the run reaches a terminal state.
workflow_run_results
ChatGPTGet owned generated image, video, audio and 3D files with preview/download links, plus the saved Nova project. Links expire after fifteen minutes; call again to refresh them. Partial results remain accessible without regenerating. Use workflow_get with include_results=true for saved node values such as text.
workflow_run_status
ChatGPTGet a saved run's progress and available image, video, audio and 3D files. Runs continue when the chat closes. No new generation or charges. Only the owning Nova account can read the run.
workflow_skill_reference
ChatGPTSearch trusted Nova workflow skills used by AI Chat. Use when diagnosing a workflow quality issue or applying a known repair such as seamless video continuity.
Capabilities
WritesInteractive
Example Prompts
Click any prompt to copy it.
