Guides
Sume guides: the API, Formats, agents, models, media tools and pricing, each answer dated and sourced.
Sume agent run webhook: one event per turn, not per clip
A Sume agent run sends one agent.run.terminal webhook per agent turn, never one per generated artifact. Key your handler on the run id.
Recap reel of 12 deal clips: 12 slots is the last single pass
A Sume timeline renders up to 12 slots in one pass. A 12-clip deals recap fits; a 13th slot makes the render chunk. See the limits, the rate, and the cost.
GPT Image 2.5 at 16:9 costs 54 to 56% of a square at the same quality
At 1280x720 GPT Image 2.5 bills $0.004 to $0.142 across five qualities on Sume, about 54 to 56 percent of the 1024x1024 price. Full table.
Commit several Sume Format files in one PUT: a change set
PUT a files list to a Sume Format's contents URL and it is one commit with one version bump. Files you do not name stay as they are. Existing paths need a sha.
Magnific Kling 3.0 Pro 1080p at 450 credits per 5 s vs Sume
Magnific lists Kling 3.0 Pro 1080p at 450 credits per 5 seconds, $0.06 to $0.09 a second. Sume bills Kling 3 at $0.14 a second (audio off) or $0.21 (audio on).
verifyWebhook toleranceSeconds 0 turns off the replay check
In @sume-com/sdk, toleranceSeconds defaults to 300 and 0 skips the timestamp check. Keep the default and use the now seam to test old deliveries.
45-minute webinar into YouTube Shorts: Sume trim takes 30 minutes
Sume video trim accepts a source up to 1800 seconds and cuts up to 900. A 45-minute webinar needs a split first; YouTube Shorts run up to 3 minutes.
Sume MCP consent: granting write always includes read
Sume's MCP consent offers mcp:read (required) and mcp:write (opt-in). A write grant always includes read, and there is no paid scope.
Manus agent calling paid Sume tools: cap spend per call
A Manus agent with a Sume API key can start paid generation. Tell it to send max_spend_usd, dry_run and an idempotency_key on every paid call. Why each matters.
Re-run only the failed rows of a Black Friday bulk queue
One row of a Sume bulk queue failed? Send only the failed rows as a new queue with a new idempotency key. The old key replays the old queue and starts nothing.
Trae MCP timeout: RUN_MCP_TIMEOUT_MS in headers vs Sume jobs_wait
Trae sets HTTP MCP timeouts in the headers field. Keep RUN_MCP_TIMEOUT_MS above Sume's 55-second jobs_wait slice; never resubmit a paid create on timeout.
Platform video limits keep changing: keep a dated table, check by plan
Shorts went from 60 seconds to 3 minutes, TikTok records up to 10 minutes, and Reels allow 20. Keep a dated limits table and test renders against it for free.
How many cuts fit in a 3-minute Short? Timeline caps slots at 200
A 180-second Short built with Sume Timeline can have at most 200 video slots, each at least 0.2 seconds. That averages 0.9 seconds a cut. Here is the math.
Match compose output to the timeline size: no double rescale
Sume timeline compose defaults to 1080x1920. Set output width, height, and fps to the timeline you join it into, so the holiday shot is not rescaled twice.
Reference video analysis on Sume: inspect, ingest or video-analyses?
For a reference video, use Sume video inspect for stills and probe, reference ingest for shots and OCR, and note that video-analyses returns 410 on dest.
Music bed under narration: 'no vocals' and 'no spoken word' clauses
Sume's Music docs say to end the prompt with 'Instrumental, no vocals' and add 'no spoken word' only when a narrator will talk over the track.
Sume idempotency: JSON key order is ignored, array order is not
Resending a Sume job with the same fields in a different JSON key order still returns the original job. Changing an array's order or a value gives a 409.
Raw-to-plated recipe clip: first and last frame on Sume
Pin the raw-ingredients photo as the first frame and the plated dish as the last. Wan 3.0, Omni Flash 1.1 and H3 take both; the 8-second prices are inside.
Timeline refuses more than 8 chained fades: fix a vertical montage
A Sume Timeline montage with nine fades in a row returns too_many_chained_transitions. Insert a hard cut and the 3-minute Short renders. Limits explained.
Slow breathing voiceover: Sume TTS speed 0.6 is the floor
For a calm, slow voiceover set generation_config.speed as low as 0.6 on Sume TTS. Below that the request is rejected, so slow the rest through writing.
Meta One lets Advanced add links to reels: burn a CTA, not a URL
Meta One Advanced adds links in organic posts and reels, so the clip itself only needs a short call to action. Burn it with Sume caption cues for $0.20.
Playlist to Shorts series: audit every file first with video inspect
A playlist becomes a YouTube Shorts series only if it holds only Shorts of 180 s or less. Probe each file with Sume video inspect before adding show features.
Wedding save-the-date clip: first frame or reference photos
Send one engagement photo as a first frame to animate it, or up to 10 reference images to Omni Flash 1.1 to keep the couple's look. 8-second prices inside.
Sume webhook secret fingerprint header: check your secret safely
Every Sume run webhook carries x-sume-webhook-secret-fingerprint. Compare it to the receipt and dashboard fingerprint to catch a wrong secret before verifying.
waitForJob pollInterval is a floor: Sume's hint can only lengthen it
Lowering pollInterval in waitForJob does not poll faster than the server asks. The SDK takes the larger of your interval and next_poll_after_seconds.
Overlay video_fit contain: keep a wide product clip uncropped
In Sume timeline compose overlay, video_fit contain fits a wide 16:9 product clip inside the frame instead of cropping it, with a sale plate over the top.
YouTube thumbnail test set: n=4 on GPT Image 2.5 high may return 202
Four 1280x720 thumbnail takes cost $0.1425 on GPT Image 2.5 high via Sume, and a slow call can return 202 instead of 200. A Python client that handles both.
DSPy Tool.from_mcp_tool with Sume hosted MCP: async notes
dspy.Tool.from_mcp_tool wraps a tool from a live MCP session. What changes with Sume's hosted server: async calls, error results and paid-tool gates.
Trae IDE: add Sume as a remote MCP server with Add Manually
In Trae, open Settings > MCP > Add > Add Manually and paste a JSON entry with Sume's hosted URL and an Authorization header. Steps, trade-offs, key handling.
Vertical workout demo, 15 seconds at 720p: Sume prices
A 15-second 9:16 workout clip at 720p is $1.13 on H3, $1.88 on Wan 3.0, $3.15 on Kling 3 with sound and $8.67 on Seedance 2.5. Omni Flash stops at 10 s.
One output object for every episode: 1080x1920 at 30 fps in Timeline
Pin the same Timeline output block on every Shorts episode: 1080x1920 at 30 fps. Mixed-model clips then do not trigger the fps resample warning.
Trim an ad cutdown from an MCP client: dry run, wait, result
Cut a 15-second holiday ad version from an MCP client with Sume video_trim: dry_run shows the $0.02 estimate, then jobs_wait and jobs_result return the MP4.
Is MiniMax H3 Max 1080p native? Sume's catalog and price
H3 Max renders natively at 480p and 768p; its 1080p is a latent refinement from 768p. On Sume 1080p is $0.20 a second, double the 768p rate.
Sume webhook signature header: why a sume-v2 entry is skipped
A Sume webhook verifier should compare only sume-v1= entries from the comma-separated header and skip other prefixes. Here is a Python check with a test.
Magnific Seedance 2.5 in dollars: 4,400 credits per 4 s vs Sume
Magnific lists Seedance 2.5 at 1,760 credits (720p) and 4,400 (1080p) per 4 seconds. In dollars that is $0.29 to $1.10 a second; Sume bills $0.58 and $1.42.
n above the ceiling returns 400 on Sume images: split the batch
The images schema allows n up to 10, but each model's own range is lower: 4 on most, 1 on Grok Image. Read the descriptor and split the batch. Python helper.
Cherry Studio: add Sume as a remote MCP server and bind it
In Cherry Studio, add a server under Settings > MCP, enable it, check its tools, then bind it to an Agent. A server that is not started cannot be bound.
Podcast trailer clip at 480p: 15 seconds on Wan, H3 or Seedance
A 15-second 480p podcast trailer clip costs $0.94 on Wan 3.0 or MiniMax H3 and $2.11 on Seedance 2.0 Fast on Sume, all with audio.
Preview a vertical clip fast: video inspect seek fast, then cut exact
Sume video inspect can return stills from the keyframe at or before each time, up to about one GOP early, to skim a long clip before an exact trim.
HDR phone clip through Sume video trim: exact mode outputs yuv420p
Sume video trim in exact mode re-encodes with libx264 and yuv420p. Keyframe mode is a stream copy. Check the probe before you promise HDR to a client.
Magnific Seedance 2 Fast and Pro in dollars vs Sume, 4 s at 720p
Magnific lists Seedance 2.0 Fast at 940 credits and Seedance 2 Pro at 1,120 per 4 seconds at 720p. Priced in dollars next to Sume's 4-second 720p rates.
Cut a 10-minute vertical video into Shorts episodes at sentence ends
A 10-minute vertical master needs at least four Shorts episodes of 180 s or less. Find sentence ends with video inspect, cut with video trim: about 18 cents.
Animal shelter adoption reel 9:16: six-second clips priced
A 6-second vertical adoption clip on Sume costs $0.45 on H3, $0.75 on Omni Flash 1.1 or Wan 3.0 and $1.82 on Seedance 2.0 Fast. All four list 9:16.
One parser for Sume run webhooks and the poll response
A Sume run webhook payload is byte-identical to the data object from the poll endpoint, so one parser serves both. Handle outcome ok, degraded and error once.
Captions language field: a speech hint, not a style or font
On Sume video captions, language only hints speech-to-text. It never picks the style or font, so Korean needs a Hangul style such as black-outline.
Sume caption design overrides: one key, one change, bad value 400s
Sume video captions design overrides merge over the style, so one key changes one thing. A bad value fails with 400 at request time, before any render.
Same Idempotency-Key, two Sume members: two jobs, not one
A Sume Idempotency-Key is unique per workspace and per key owner. Two members who send the same string get two jobs. Namespace your keys by what you generate.
Band teaser with your own track: which Sume models take audio
Wan 3.0, H3, H3 Max and Seedance accept an audio reference on Sume; Omni Flash 1.1 and Kling 3 do not. A 12-second 720p teaser is $0.90 to $6.94.