complete_asset_upload
ChatGPTValidate and archive your uploaded file, free. Requires assets.upload. Retry the same upload_id after transient errors; do not create a new session on timeout. Only ready references may be used for generation. Keep original reference URLs on generation retries.
confirm_video_clone_script
ChatGPTGenerate one ad-clone video after script approval and authorization under the selected progression mode. Automatic mode needs no additional spending prompt; confirm-each-time needs approval of a fresh quote. Ask for the mode if unset. Automatic mode does not replace full-script approval. Requires generation.create. Supply the current script_version and the approved script (1-20000 characters). Overrides allow only resolution/aspect_ratio. Model and duration stay fixed by the project. Use a UUID; retries must keep project, script/version, overrides and key unchanged. Same key recovers the existing task; a different key intentionally creates another paid video. On SCRIPT_VERSION_CONFLICT fetch and show the new script, then obtain approval again. On REQUEST_IN_PROGRESS retry the same request later; on UPSTREAM_SUBMISSION_UNCERTAIN do not use a new key. Open show_generation_job immediately for the returned job ID; let supported cards poll and end the turn. Without live cards, use bounded get_generation_job checks. The project retains its original analysis script.
create_asset_upload
ChatGPTCreate a free single-file upload session. Requires assets.upload. Image/video generation, motion imitation and video cloning require references in this environment's Loova storage. For external media, obtain the actual file and upload it here; do not pass the external URL to generation, even if it plays in a browser. Verified existing Loova media can be reused. Use a UUID idempotency key. Supply all three file fields or omit all for web upload. With local file access, send the raw file bytes using the returned upload method, URL and headers; do not use multipart or put file bytes/paths in MCP calls. Never print upload credentials. Without local file access, show upload_page_url and ask the user to upload there and return. Repeat the SAME create request to renew expired credentials without changing the file. After a PUT timeout, call complete_asset_upload before retrying the upload. After transfer, call complete_asset_upload/get_asset_upload until ready and use reference.url, NOT the PUT upload.url. Upload does not authorize paid generation; quote and follow the selected progression mode.
create_video_clone_project
ChatGPTAnalyze an ad reference and prepare an editable script. Requires generation.create. Supply one reference video, 1-3 product images and optionally one person image, ALL already uploaded to this environment's Loova storage. Use ready reference.url values from complete_asset_upload/get_asset_upload or verified existing Loova media. External URLs (including playable video/CDN/other OSS links) are rejected with REFERENCE_UPLOAD_REQUIRED before analysis; upload those files via create_asset_upload first. Select an image-reference video model from list_generation_models(mode=img2vid). Parameters may specify duration, resolution, aspect_ratio, and out_number=1; model and duration are fixed at creation. Product analysis/script preparation has no separate Credits charge. This tool never starts video generation. Use a UUID key and keep the SAME full URLs, model, parameters and instructions on retries. Query get_video_clone_project until the script is ready; show its summary and complete script to the user. Then quote_video_clone, obtain script approval and follow the selected spending-confirmation mode before confirm_video_clone_script. Automatic progression does not skip creative/script review.
generate_image
ChatGPTRequest one image under the user's selected progression mode using Loova's generation service. Use list_generation_models for allowed parameters and quote_generation to check access and estimate cost. Automatic mode submits user-requested tasks without repeated spending confirmation or a preset Credits cap. Confirm-each-time requires approval of model, parameters and cost before submission. Ask for the mode if unset. The main backend selects the provider and applies its normal billing/benefit rules. billing_preference=unlimited requires eligible Unlimited and never falls back to Credits; default preserves normal billing, including trials. Quote with the same preference and show queue wait; follow the selected mode. Preserve billing_preference on retries. Do not switch to paid default without new user confirmation. Optional references use [{"type":"image" or "video", "url":"https://..."}]. Select and quote img2img when references are present; otherwise use txt2img. Check model reference_types. References must be in this environment's Loova storage, using ready reference.url values from complete_asset_upload/get_asset_upload or verified existing Loova media. External storage URLs are rejected with REFERENCE_UPLOAD_REQUIRED; upload those files via create_asset_upload first. URLs must be HTTPS without credentials or fragments; browser playback is not sufficient. Reuse the SAME reference types, full URLs (including signatures) and order on retries. Requires generation.create. Prompt: 1–4000 characters. Key: 8–128 alphanumeric/underscore/hyphen; use a UUID and reuse the SAME key, model, parameters and prompt on retries. Open show_generation_job immediately after submission; let supported cards poll and end the turn. For clients without live cards, use bounded get_generation_job checks. REQUEST_IN_PROGRESS means retry later with the same key. On UPSTREAM_SUBMISSION_UNCERTAIN, contact Loova; do not generate again with a new key.
generate_motion_imitation
ChatGPTGenerate one motion-imitation video under the user's selected progression mode. Requires generation.create. Use list_generation_models(mode=motion_imitation) for models and configured parameters. Supply one character image and one motion video already uploaded to this environment's Loova storage. Use ready reference.url values from complete_asset_upload/get_asset_upload; external storage URLs are rejected with REFERENCE_UPLOAD_REQUIRED. Upload external files through create_asset_upload before submission. Quote motion_imitation with reference_video_duration first. Automatic mode needs no further spending confirmation; confirm-each-time must show the estimate and wait for approval. Ask for the mode if unset. The backend measures actual video duration, selects the provider and applies existing product billing. Optional prompt adds instructions (maximum 4000 characters). Do not pass quote-only duration to generation. Use a UUID key. Retries must keep the SAME model, full URLs, parameters, prompt and key. On REQUEST_IN_PROGRESS retry later with the same request; on UPSTREAM_SUBMISSION_UNCERTAIN do not use a new key. Open show_generation_job immediately after submission; let supported cards poll and end the turn. For clients without live cards, use bounded get_generation_job checks. Uploading files does not authorize paid generation.
generate_video
ChatGPTRequest one video under the user's selected progression mode using Loova's generation service. Use list_generation_models for allowed parameters and quote_generation to check access and estimate cost. Automatic mode submits user-requested tasks without repeated spending confirmation or a preset Credits cap. Confirm-each-time requires approval of model, parameters and cost before submission. Ask for the mode if unset. The main backend selects the provider and applies its normal billing/benefit rules. billing_preference=unlimited requires eligible Unlimited and never falls back to Credits; default preserves normal billing, including trials. Quote with the same preference and show queue wait; follow the selected mode. Preserve billing_preference on retries. Do not switch to paid default without new user confirmation. Optional references use [{"type":"image" or "video" or "audio", "url":"https://..."}]. Select and quote img2vid when references are present; otherwise use txt2vid. Check model reference_types. References must be in this environment's Loova storage, using ready reference.url values from complete_asset_upload/get_asset_upload or verified existing Loova media. External storage URLs are rejected with REFERENCE_UPLOAD_REQUIRED; upload those files via create_asset_upload first. URLs must be HTTPS without credentials or fragments; browser playback is not sufficient. Reuse the SAME reference types, full URLs (including signatures) and order on retries. Requires generation.create. Prompt: 1–4000 characters. Key: 8–128 alphanumeric/underscore/hyphen; use a UUID and reuse the SAME key, model, parameters and prompt on retries. Open show_generation_job immediately after submission; let supported cards poll and end the turn. For clients without live cards, use bounded get_generation_job checks. REQUEST_IN_PROGRESS means retry later with the same key. On UPSTREAM_SUBMISSION_UNCERTAIN, contact Loova; do not generate again with a new key.
get_asset_upload
ChatGPTRead your upload status and fresh reference URL, free. Requires assets.read. Does not upload or validate files. After web upload, call this to continue the workflow. Poll only a bounded number of times, or ask the user to return when upload is complete. New generation requests may use fresh URLs; retries must keep the original reference and key.
get_credits
ChatGPTRead the current Loova product Credits balance, not API Credits. Requires credits.read. Free and read-only.
get_generation_job
ChatGPTRead 1–20 own image/video jobs. Requires generation.read; free. Pass job_ids, even for one task. Duplicates are removed in input order. Each jobs item contains a task or status=unavailable and a safe error code. Result links follow product storage retention; unsigned domain links have no expires_at field. credits_cost is the main backend's recorded task cost, not a refund confirmation. billing_method is the recorded payment method; zero cost alone does not mean Unlimited. queue_wait_seconds is remaining scheduled wait, not provider completion time. Respect poll_after_seconds. job_id is the Loova product task record ID, not a provider task ID. After show_generation_job in a UI-capable client, let the card poll; do not duplicate polling. Use this directly for text-only clients, unavailable card updates, or explicit status requests.
get_profile
ChatGPTRead the current authenticated Loova account profile. Requires profile.read. Free and read-only.
get_video_clone_project
ChatGPTRead your MCP ad-clone project, analysis summary, editable script/version and latest job IDs. Requires generation.read; free. Never starts or confirms generation. Poll a bounded number of times. Open show_generation_job for returned video job IDs; supported cards handle result polling. Use bounded get_generation_job checks only without live cards or for explicit status requests. A completed analysis is not a completed video.
list_generation_models
ChatGPTList all enabled MCP-supported image/video models and their options/defaults, regardless of your plan. A listing is not permission to use a model. Quotes check access; submission rechecks access and billing. Use img2img or img2vid when supplying references; check reference_types for accepted media. Use motion_imitation for a character image plus a motion reference video; check required_inputs. unlimited lists active entitlement parameter alternatives, not guaranteed eligibility or free access to all settings. Obtain an Unlimited quote for the exact settings; defaults are never changed automatically. Requires generation.read; free. Use quote_generation for an estimate.
quote_generation
ChatGPTRead a current price estimate for one image or video. Requires generation.quote; free. The estimate is not saved, does not reserve a price and is not a generation credential. Actual billing and eligible benefits follow the main backend's existing Agent generation rules. Default quotes read-only check available Credits after eligible free trials; no Credits or trials are reserved or consumed. Creation rechecks billing, since balance and benefits can change after a quote. Always quote, then follow the conversation's progression mode: automatic submits user-requested tasks without another spending prompt or preset Credits cap; confirm-each-time shows estimated_credits, model and normalized parameters and waits for approval. Ask for the mode if not yet selected. confirmation_required accepts the user's active automatic-mode authorization, not only per-task approval. Use parameters from list_generation_models. Quotes additionally accept reference_video_duration for video-reference pricing and has_reference_video=true for video-to-image references. Do not pass these quote-only fields to generation. For motion_imitation, reference_video_duration is required for the estimate; execution measures the video. billing_preference=default preserves normal billing, including eligible trials; it does not select Unlimited. Use unlimited only for ordinary image/video modes. It requires an eligible entitlement and never falls back to Credits. Show the estimated wait and zero Credits quote; eligibility and queue are rechecked at creation. Use the SAME billing_preference when generating. A quote does not reserve entitlement or concurrency. PLAN_REQUIRED and INSUFFICIENT_CREDITS include purchase guidance when available. Show the returned action URL, not an invented link; neither error requires OAuth reauthorization.
quote_video_clone
ChatGPTEstimate one video for a script-ready ad-clone project. Requires generation.quote; free. Optional overrides are resolution and aspect_ratio only. Uses product pricing and ad-clone feature multiplier. Show the complete script, version and estimate and obtain script approval. Automatic progression covers spending authorization without a preset Credits cap or another cost-confirmation prompt; confirm-each-time also requires approval of the estimate. Ask for the progression mode if unset. confirmation_required does not require repeated spending approval in active automatic mode. The quote is not saved or price-locked; actual billing follows the main backend's rules. Checks available Credits after eligible free trials without consuming or reserving either. For PLAN_REQUIRED or INSUFFICIENT_CREDITS, show the returned action URL; do not retry automatically.
show_generation_job
ChatGPTDisplay 1–20 existing Loova jobs in a read-only result grid where supported. Pass job_ids returned by generation or supplied by the user, even for one job. Requires generation.read; free. Duplicates are removed in input order. Does not create, retry or charge for generation. The card refreshes via get_generation_job. Clients without UI receive the same structured task status and result links as text. Call immediately after generation returns job IDs, even while queued/running. In UI-capable clients with card-side tool calls, hand off polling to the card and end the turn. Do not also poll, schedule background checks, or repeatedly call this tool. Tool success alone does not confirm client rendering. If UI support is unknown, state that automatic updates depend on the client; if unavailable, use bounded get_generation_job checks.