Dare

Estimate image credits using the same model params and references as generate_image. Prompt is optional. Does not submit generation or spend credits.

Quote hosted transcription of owned audio/video; includes duration-based provider and extraction costs.

Quote scene analysis for an owned video without starting work or spending credits. Uses the same fixed credit price as billing and reports current balance and affordability. The balance is checked again and credits charged before analysis starts.

Create one standalone speech audio file from text with Seed Audio 1.0 or ElevenLabs. This paid action spends the connected account's Dare credits. Before calling it, choose a supported voice, call estimate_audio_cost with the same input, and get the user's approval. For one narrator, use the complete script in one request; distinct speakers can use separate requests and voice IDs. Seed Audio supports @Audio1–@Audio3 reference voices. Returns an owned job ID to poll with media_job_status, and saves the completed audio to the library. Never automatically play the output.

Create up to six independent speech audio files in one paid batch. Each accepted request spends the connected account's Dare credits. Before calling it, quote every request with estimate_audio_cost, show the total, and get the user's approval. Stable caller indices map requests to results. Preserve successful job IDs on partial failure, poll every accepted ID with media_job_status, and do not retry the entire batch or resubmit accepted jobs.

Burn captions into an owned video. Reuse clean narration timing with alignment:{jobId,offsetSeconds:0} from a completed align_audio job on this account; no new speech analysis is performed or charged. Use only when speech timing is unchanged apart from that offset. Alternatively supply transcript for script audit + forced alignment; omit both to use hosted Scribe. Transcript and alignment are mutually exclusive. Settings include fontFamily, fontSizePx, pixel margins, colors, outline/shadow, word/character/line limits, gap/duration limits, hold and text case. Preserves audio through picture-only rendering. Poll media_job_status for the asset, words, audit, and actual rendered phrase cues with measured line breaks and SRT/VTT. SRT/VTT carries text/timing, not ASS styling or active-word highlight animation. Up to five minutes. Requires the media-render container.

Create one or more images from a text prompt with a supported Dare image model. This paid action spends the connected account's Dare credits. Before calling it, use models_get to validate model-specific params, call estimate_image_cost with the same params and references, and get the user's approval. Optional references must be owned uploads, products, characters or completed generations; address them as @ref1, @ref2 and so on. Prompts default to verbatim. Returns generation IDs to poll with jobs_wait. Do not resubmit while any returned ID is queued or processing.

Submit up to six paid video requests. Each request has model params, references and count. Use @ref1 etc. for references. A successful call may charge credits. Keep every returned generation ID. An identical request with active jobs returns those same IDs without a new submission or charge. Use jobs_wait to poll. Never resubmit queued or processing jobs. Ask the user before a new paid attempt after a terminal failure. Results use zero-based request indices; a submission failure stops later entries.

Read a workflow recipe without executing it. Omit preset to list commands. Use /video-editing for a finished video from a supplied script, beat sheet or footage, including motion graphics, audio and captions. Use /marketing-adapt-video to adapt an existing reference ad.

Wait up to 5 seconds for the specified owned generation jobs. complete=false means the original jobs remain active. Call jobs_wait again with pendingIds. Never call a generation tool while an ID is queued or processing. Polling never submits work or spends credits. A refund exists only when status=failed and billingStatus=refunded.

List public provider voices with gender, languages, accent, delivery, descriptions, available preview URLs and metadata sources. Missing attributes are unknown. Filter exactly by gender/language code/accent; search matches words across names and descriptions. No playback or charge. Use model=seed_audio or text2speech_v2; paginate with next_cursor. Saved character voices use voice_type=element and voice_id=character ID.

After a successful upload PUT, save it to your library. Returns the upload ID to use in {kind:upload,id} references. Poll media_get until the asset status is succeeded before referencing it.

Read an owned library upload's status, metadata and URLs.

Read an owned speech/transcription/alignment/caption job. Completed media includes an upload reference; poll media_get until its asset is ready before reuse. No automatic playback.

List your uploaded library assets, newest first, with a pagination cursor.

Reserve an image, video or audio upload. Supply the actual file size in bytes. PUT the file to uploadUrl with its Content-Type, then call media_confirm. No ZIP archives or local-path reading.

Inspect a model's accepted params and reference limits. Use these exact params with generate_image, generate_video_batch or generate_audio. Unsupported model IDs fail.

Save a named product and factual description from an owned image reference. Reuse the returned ID as {kind:product,id} in generation references.

List your saved products with images and descriptions.

Read owned generation jobs with explicit terminal and billing status. queued or processing means the original job remains active. Reuse pendingIds with jobs_wait. A refund exists only when status=failed and billingStatus=refunded. This tool never submits work or spends credits.

Transcribe owned audio/video with hosted Scribe v2 or Whisper, up to five minutes. Returns a job ID; poll media_job_status for transcript, timed words, speaker labels where available, and word-level SRT/VTT. Does not require local Whisper.

Start paid scene analysis using Ad Studio's analyzer. Call estimate_video_analysis_cost first. Source is an owned uploaded or generated video. Credits are charged against your balance before work starts and refunded on failure. Returns an analysis ID; poll video_analysis_status. Requires the media-render container. Longer videos can reduce observation accuracy.

Read an owned scene analysis. Completed output contains structured scenes with source timing, visuals, dialogue, audio, graphics and uncertainties, from Ad Studio's source analyzer.

Get a public voice's description, gender, languages, accent, delivery, preview URL and metadata sources. Missing values are unknown. IDs belong to the selected provider catalog.