Create from
your conversation.

Animate an owned image into a video clip.

Restyle an owned image while keeping its subject.

Cancel your own queued or running job. Budget spent is not returned; credits are refunded by the existing path.

Re-frame an owned image from another camera position.

Relight an owned image.

Read text in a voice cloned from owned audio.

Promote a pending upload after bytes land. Idempotent; refuses foreign tickets.

Create a new empty project to file generated work into. Cannot rename, archive or delete.

Create a single-use presigned PUT for your own bytes. No spend; confirm after PUT.

Read a public model input schema for an exact tool and model ID or slug. Documentation only; cannot execute or upload.

Read one executable tool/model input schema, pricing and polling contract. Documentation only.

Dub an owned video into another language.

Edit an owned image from an instruction.

Outpaint an owned image to a larger frame or new aspect ratio.

Continue an owned video past its current end.

Create a 3D model from a text prompt.

Create an image from a text prompt.

Compose an instrumental track from a description.

Create a sound effect from a description.

Compose a song with vocals from a description.

Read text aloud as speech.

Create a video clip from a text prompt.

Read your account ID, name, email and verification status. Requires explicit account consent.

Read an owned asset and an approved private URL if configured. Does not fetch or upload media.

Mint a short-lived private download URL for an owned asset. Returns null when storage cannot sign.

Read this connection’s scopes and lifetime credit budget. grantExpiresAt (also expiresAt) is the grant expiry; accessTokenExpiresAt is the short-lived access-token expiry. Refresh rotates tokens without extending the grant. This is not an admin permission grant.

Read spendable credits using existing account entitlements, plus the lifetime consent budget and reserved amount. The submission transaction is authoritative.

Read one of your wallet transactions by its internal ID.

Read status and credit cost of your own job.

Read approved output links for your own job. Links may grant temporary access; treat them as private.

Read metadata for one of your projects.

Read recorded subscription status and period. Does not calculate entitlements, renew or cancel.

Price a tool/model/parameters triple with the same resolver dispatch charges through. No hold, no job.

Read one of your top-up orders by internal ID. Provider references are excluded.

Create a 3D model from an owned image.

Replace a masked region of an owned image.

Match the mouth in an owned video to owned audio.

Read published application IDs, names, categories and tool keys. No execution or internal configuration.

Discover public model/tool mappings and user credit prices. Optional exact filters; no generation. Prices are catalogue values, not a quote.

Read your saved chat room IDs and titles from the latest 100-room snapshot. Total counts the snapshot. Client-stored session references are excluded.

Read your chat session IDs and titles. Message bodies and token-priced chat operations are unavailable.

Read IDs, last activity and unread counts of conversations you belong to, from the latest 100-conversation snapshot. No participants or message bodies. Nothing is marked read.

Read a bounded page of your wallet transactions. Does not create a wallet.

Read your existing credit ledger. Amounts and statuses are recorded values, not a billing estimate.

Read your existing subscription payment documents. No provider calls or settlement.

Read your non-deleted job history, optionally within an owned project. No guest or foreign jobs.

Read notification IDs, titles and status. Message bodies and delivery metadata are excluded; nothing is marked read.

Read your persona IDs, names and categories. System prompts and private configuration are excluded.

Read current published plans from the existing billing catalogue. Display prices are not a checkout quote.

Read owned project asset metadata with approved private URLs. Project-scoped; refuses foreign projects.

Read names and status of your non-archived projects.

Read your saved prompt IDs and titles. Prompt bodies and parameters are excluded.

Read a bounded page of your support ticket metadata. No messages, attachments, staff information or changes.

Discover executable studio operations and models from the live DB catalogue. Same executability predicate as dispatch.

Read your recorded top-up orders. Does not verify a provider payment or grant credits.

Read currently sellable fixed packages using existing billing configuration. Custom amount quoting and purchase are unavailable.

Read your own direct-upload assets and whether each ticket still awaits bytes. Never includes foreign uploads.

Drive an owned character image with the motion of an owned clip.

Recommend a runnable tool/model for an operation or a described goal, with alternates and credit costs, from the live DB catalogue. Reads only; no hold and no job.

Rank configured speech models for a voice-over style. Per-voice options live in the chosen model input schema, which describe_tool_model returns.

Cut the subject out of an owned image.

Cut the subject out of an owned video.

Repair and sharpen faces in an owned image.

Re-render an owned video in another style.

Retry your own failed job within a credit ceiling and idempotency key. Reports the retried job; replays never re-spend.

File an owned completed asset into an owned project. Sets projectId only.

Submit up to 5 priced studio steps under one idempotency key and credit ceiling. Fails whole on unpriceable steps. Output dependencies require completed prior output; asynchronous dependencies return partial results without dispatching dependent work.

Spend credits to generate using an explicitly selected public tool/model. Requires prior user consent, a per-call credit ceiling and a stable retry key. Never retry ambiguous failures with a new key.

Return an ordered plan of task tools for a described goal, including only steps with a runnable model. A plan, not a submission: nothing is dispatched and nothing is spent.

Move the face from one owned image onto another.

Replace the face in an owned video.

Make an owned portrait speak or sing.

Put media on your account from a public https link or inline base64, and get an asset ID usable immediately by the editing tools. No spend. Use this instead of create_upload unless you can PUT raw bytes yourself.

Increase the resolution of an owned image.

Increase the resolution of an owned video.

Dry-run schema coercion and pricing for a tool/model without spending. No hold, no job.

Wait for your own job to reach a terminal state, up to a bounded timeout, and return its approved output links. completed=false means it is still running — call again.