Developers
Building on the Sume API: keys and hosts, jobs and runs, polling, webhooks, idempotency, errors and rate limits, the SDK, the CLI, and MCP auth.
Music API metadata field: tag a track job with your own ids
Sume's music request has an optional metadata field, stored on the job and not sent to the provider. Use it to match tracks to your own records.
Netflix subtitle limit: 42 characters per line, and max_chars
Netflix's English timed text spec allows 42 characters per line and two lines. Sume's caption design.phrasing.max_chars accepts 4 to 60, so 42 fits.
Subtitle reading speed: check 20 characters per second before burning
Netflix caps English subtitles at 20 characters per second for adults, 17 for children. Check each cue's rate in a short script before a Sume caption render.
Node fetch with AbortSignal.timeout: poll a Sume job
A Node recipe: submit a Sume job with fetch, bound each call with AbortSignal.timeout, poll status_url, read the result. A local timeout does not cancel it.
OpenAI structured outputs: 5000 properties, 10 levels, vs Sume
OpenAI caps strict schemas at 5000 properties and 10 nesting levels. Sume's output_schema uses the same numbers but counts enums and strings differently.
Python httpx and asyncio: submit and poll a Sume image job
A runnable Python recipe: submit a Sume image job in async mode with an Idempotency-Key, poll with httpx and asyncio, honor retry-after, fetch the artifacts.
Repeat a TTS take: read generation_config and speed from the job
A completed Sume TTS job records its engine, voice, language, output_format, generation_config and speed. Read them back to make the next line sound the same.
Ruby Net::HTTP: submit and poll a Sume job with one key
A Ruby recipe using only Net::HTTP: submit a Sume image job with an Idempotency-Key, set open and read timeouts, poll status_url, and return the artifact URLs.
Seedance 2.5 reference audio alone returns 400: add an image
On Sume's Video Router, reference_audio_urls with no reference image or video returns 400. Pair the audio with an image or clip, within 3 audio and 12 total.
Seedance 2.5 seed, size and provider options: why Sume returns 400
Sume rejects seed, size and non-empty provider.options on seedance-2.5 with 400 instead of dropping them. What to send to repeat a clip or fix the frame size.
Seedance 2.5 sync mode: a 30-second wait, then poll
mode sync on the Video Router blocks at most 30 seconds. A Seedance 2.5 job can outlast it, so read sync.timed_out and poll status_url without resubmitting.
Sonic 3.5 to 3.6 on Sume TTS: change model, keep the voice id
Cartesia says Sonic 3.6 keeps the voice ids of 3.5. On Sume's TTS Router, move a call from sonic-3.5 to sonic-3.6 by changing only the model field.
sonic-preview voice_model_mismatch: use sonic-3.6 for clones
A TTS job on sonic-preview fails with voice_model_mismatch when the voice is a pro voice clone. Send model sonic-3.6 to the Sume TTS Router instead.
soundtrack_shorter_than_spine: loop the music bed in Timeline
Timeline warns soundtrack_shorter_than_spine when the music bed is shorter than the audio. Set soundtrack.loop to true; gain, fade and duck defaults inside.
StepAudio 3 Music preview is free, then retired: keep your API stable
stepaudio-3-music-preview is free for a limited time, then retired for a paid version. How to keep a music integration stable, and which Sume ids stay put.
Sume job response: status_url, result_url, events_url, cancel_url
A Sume submit returns four URLs: status_url to poll, result_url once result_ready is true, events_url for the timeline, cancel_url while cancelable is true.
Sume job status: queue.state, a null position, worker_heartbeat
Why queue.position is null on a Sume job status, what queue.state and worker_heartbeat report, and what a poller should do when a job sits in the queue.
Sume job usage_summary: reserved, captured, refunded, final
Read usage_summary on a Sume job: status reserved, captured or refunded, amounts in micros, the final flag, and why dollars are micros divided by 1,000,000.
Sume job webhook_delivery: attempts, exhausted, redeliveries
Read the webhook_delivery object on a Sume job: status, attempts of 10, last_status_code, manual_redeliveries, and what to do when delivery is exhausted.
Sume SumeMediaFile duration_ms: the 10 percent check and null
In a Sume structured output, a duration_ms must agree with the artifact's recorded length within 10 percent, or the projection fails. A null means not measured.
next_poll_after_seconds vs recommended_poll_interval_seconds
Which delay a Sume poller should sleep: next_poll_after_seconds, recommended_poll_interval_seconds, or retry-after. Null rules, a fallback, and read budgets.
TTS boundary_lead_ms: why sentence slices end 70 ms after a word
Sume TTS sentence segments cut boundary_lead_ms after a sentence's last word: 70 ms by default, 0 to 500. The next segment absorbs the pause, and no gap opens.
TTS language omitted: Spanish text comes out with an English default
On Sume TTS, an omitted language defaults to English at the provider, with a ko or ja fallback only for Hangul or kana text. Set language for all others.
TTS sentence slices have no audio_url: emit_audio needs wav or raw
With segmentation on, Sume TTS only returns a sample-exact audio_url per sentence for wav or raw output. For mp3 you get timings and a warning instead.
TTS speed: slow, normal, fast or generation_config.speed 0.6 to 1.5?
Sume TTS marks the slow, normal and fast speed enum deprecated. Send generation_config.speed from 0.6 to 1.5, plus volume 0.5 to 2 and an emotion string.
TTS call returned processing in sync mode: poll, do not resubmit
Sume TTS in sync mode waits at most 30 seconds. If the job is not terminal, poll status_url, and retry a submit only with the same Idempotency-Key.
TTS voice.id: a UUID or a voi_ library id? What Sume accepts
Sume TTS voice.id takes a voice UUID or a voi_ library id. Any other shape fails with 400 invalid_voice_id before a job is queued or credits are reserved.
Veo 3.1 returns one video per request: how to get 4 variants
Google's Veo 3.1 table says one video per request. To get four variants on Sume, send four requests with four Idempotency-Keys, and watch queue_full.
Which voice does my avatar speak with? Check voice.status is ready
Sume TTS speaks in an avatar's voice when voice.status is ready. List avatars, check voice.status, then send avatar_id or avatar_handle on the TTS request.
Bannerbear sync API 408 after 10 seconds vs Sume mode sync
Bannerbear's sync endpoint answers 408 if the render takes over 10 seconds. Sume's mode sync waits up to 30 seconds, then returns 202 to poll.
C2PA Conformance Explorer: vet your signer before promising credential
Before you promise clients C2PA Content Credentials, look your signing tool up in the C2PA Conformance Explorer. What it lists and how IPTC used it in 2026.
C2PA Interim Trust List frozen: why Content Credentials look untrusted
The C2PA Interim Trust List was frozen on 1 January 2026. What that means when a validator flags credentials, and what to check on files you deliver.
China's implicit AI label: provider name and content ID in metadata
China's implicit AI label is metadata with the provider name and a content ID. Why a re-encode can drop it, and how to check a delivered file with ffprobe.
Run one prompt on five image models via Sume: 200 vs 202 in Python
A Python script fires one prompt at Seedream, FLUX.2, Qwen, Ideogram and Recraft ids on Sume and handles both the 200 image reply and the 202 job envelope.
Why a published Instagram Reel shows media_type VIDEO
A published Reel reads back as media_type VIDEO, so test media_product_type to tell it from other video. Here is the check, plus a file preflight with Sume.
ig_reels_avg_watch_time: cut a shorter Reel with Sume video trim
ig_reels_avg_watch_time is the average time spent playing a Reel. If it sits far below your length, trim a tighter cut with Sume video trim and compare.
Is there an Instagram webhook when a Reel finishes processing?
Instagram's webhook field list has no Reel-finished event, so poll status_code. Sume job webhooks are signed; verify them with a secret that cannot be empty.
Instagram Reels resumable upload (rupload) vs a public video_url
Instagram Reels can publish from a public video_url or a resumable rupload. Resumable is Facebook Login only; check your Sume media link is reachable first.
reels_skip_rate: what the 3-second metric means and how to test hooks
reels_skip_rate is the share of Reel views that skipped in the first 3 seconds. Pull the opening frames and words with Sume video inspect to compare hooks.
Reels views vs crossposted_views vs facebook_views: which to report
Instagram Reels insights list views, crossposted_views and facebook_views separately. Which to report, and how to trace Sume-made variants.
LangGraph 1.2 node timeout: the Sume video job keeps billing
LangGraph 1.2 adds run_timeout and idle_timeout per node. A timeout stops your node, not the Sume job it started: store the job id and re-poll.
LangGraph DeltaChannel: keep Sume artifact URLs in state, not video
LangGraph 1.2 DeltaChannel stores only the per-step delta. Even so, keep Sume media out of state: store the artifact URL and job id and fetch bytes when needed.
LangGraph error_handler after retries: do not resubmit a Sume job
LangGraph 1.2 node error handlers run after retries are exhausted. For a Sume call, the handler should read job status, not submit a second paid request.
LangGraph request_drain and GraphDrained: resume a Sume job later
LangGraph 1.2 can drain a run after the current superstep. If a Sume job is in flight, save its id so the resumed run polls instead of resubmitting.
MiniMax H3-Context-IR returns a prompt, not a video: Sume view
MiniMax's H3-Context-IR endpoint reads text, images, audio and video and returns an enhanced prompt without making a video. What Sume has instead.
MiniMax task statuses vs Sume job statuses: a mapping table
MiniMax reports queued, running, succeeded, failed, cancelled. Sume job status uses queued, processing, completed, failed, canceled. Map them correctly.
Mux Robots webhook events vs Sume job events: a verifier
Mux sends robots.job.{workflow}.{status} on every change. Sume sends only job.completed, job.failed and job.canceled, signed. Verify them in Python.
OpenAI Agents 0.23 MCP listing page limits and Sume tools_list
Agents SDK 0.23.0 adds configurable MCP listing page limits. For a Sume server: call tools_list once, then tools_schema for the one tool you need.