Guides
Sume guides: the API, Formats, agents, models, media tools and pricing, each answer dated and sourced.
urllib3 Retry on POST: retry a Sume submit only with a key
A urllib3 Retry config that retries 429, 502, 503 and 504 on a Sume submit, and a wrapper that refuses to send a POST without an Idempotency-Key.
Validate a Gemini Omni Flash 1.1 request in Python before sending
A 28-line Python check for Sume's gemini-omni-flash-1.1 rules: 3-10 s, 10 reference images, 3 reference videos, no audio off, and edit mode exclusions.
Veo 3.1 last-frame control: which Sume video models take an end frame
Sume does not list Veo. For first-and-last-frame video it lists Seedance, Kling v3 Pro, Wan 3.0, MiniMax H3 and H3 Max and Gemini Omni Flash 1.1.
Verify a Sume Format webhook in Python: empty secret, stale timestamp
A Python check for the Sume format.run.terminal webhook: HMAC-SHA256 over timestamp.raw_body, a 5-minute window, and a refusal to run with an empty secret.
Sume video models by aspect ratio: who takes 9:16, 21:9 and 1:1
Seedance takes 21:9 through 9:16, Kling only 16:9, 9:16 and 1:1, Gemini Omni only 16:9 and 9:16. Full matrix of Sume video models as of 2026-10-08.
AI video API with audio reference input: which Sume models accept it
Seedance (all four), Wan 3.0, MiniMax H3 and H3 Max accept audio references on Sume. Kling, Gemini Omni, Grok Imagine and the motion models do not.
Sentence segments from video inspect: caption lines with no gaps
Sume video inspect with transcribe true can return gapless sentence segments, with silence_split_seconds from 0.2 to 3. Feed them to captions as cues.
AI video with native audio: which Sume models let you turn it off
On Sume, only Kling v3 Pro prices audio separately (+$0.07/s). Gemini Omni and MiniMax always make audio; Grok Imagine and Genjutsu make none. Table inside.
Wan 3.0 on Sume: a 2-second minimum clip and a 30-second max, priced
Wan 3.0 accepts 2 to 30 seconds. On Sume a 2 s clip is $0.125 at 480p and $0.50 at 1080p; 30 s at 1080p is $7.50.
Wan 3.0 reference limits on Sume: 10 images, 5 videos, 5 audio clips
Wan 3.0 on Sume takes up to 10 reference images, 5 reference videos (15 s total, 16 fps or more) and 5 audio clips (15 s total). Here is the full envelope.
How many Sume image rows list mask, background, quality, resolution
mask_url and background list on GPT Image 2.5 only, quality on three row types, resolution on a handful. A count of catalog rows by parameter, with a script.
YouTube bitrate ladder: file size of a 3-minute Short by resolution
YouTube's recommended SDR bitrates turned into file sizes for a 180 second Short, with the arithmetic and a script. This is YouTube guidance, not Sume output.
Zapier MCP costs two tasks a call: batch Sume jobs_wait first
Each Zapier MCP tool call uses two tasks. When an agent uses Sume and Zapier together, wait on all jobs in one jobs_wait, then make one Zapier call.
31-second driver is over Kling Motion Control's 30 s limit
Kling 3.0 Motion Control accepts 1 to 30 seconds at $0.1575 per second. Trim a 31 second driver to 30 and the job costs $4.725.
Drop to 480p: Fabric saves 46.7 percent, H3 lip-sync 37.5 percent
Fabric is $0.10 at 480p against $0.1875 at 720p. H3 Max lip-sync is $0.0625 at 480p against $0.10 at 768p. Savings on 250 clips inside.
Avatar video cost at 4, 8, 15, 30 and 60 seconds on three tiers
Avatar video costs $0.74 to $11.04 on standard, $0.98 to $14.70 on plus and $2.20 to $33.00 on max across the 4 to 60 second window, no product.
Cancel every queued Sume job without cancelling a running one
A Python script that lists queued jobs, checks the cancelable flag, posts the cancel, and handles 409 job_generation_already_started when the race is lost.
caption_no_speech: why a caption job fails on a silent clip
A Sume caption job returns caption_no_speech when the clip has no audible speech. The fix is cues, segments or words, not a retry.
Change one caption color: design.colors.active and what it accepts
A Sume caption design override changes one token for one request. Colors take hex, rgb(), rgba() or transparent; other CSS is rejected at request time.
Check a webhook URL before you give it to Sume: https and public
Sume rejects localhost, private-network and non-HTTPS webhook URLs. A Python preflight that catches all three, and the 2048-character limit, before you submit.
Claude Code headersHelper for a Sume API key: rotate with no edits
Claude Code can run a script for MCP headers on every connection. Use it to feed a Sume API key from a secrets store, so rotation needs no config edit.
Claude Code 25,000-token MCP output cap and Sume batch jobs_result
Claude Code truncates MCP output past 25,000 tokens by default. Sume include_results and batch jobs_result keep wave answers small. Settings and loop.
Claude Code MCP_TOOL_TIMEOUT and the timeout field for Sume jobs_wait
Claude Code has no short default tool timeout, so a stalled Sume jobs_wait can hang. Set a per-server timeout in .mcp.json and let the agent retry slices.
ElevenLabs Free 10,000 credits is 11 minutes of music, no license
ElevenLabs Free gives 10,000 credits a month, about 11 minutes of music, without a commercial license. Sketch there, then ship a track via Sume at $0.125.
Flux, Qwen and Recraft share 13 ratios on Sume: five rows, four prices
Flux 2 Pro and Flex, Qwen Image and Max, and Recraft V4 list the same 13 aspect ratios on Sume. Prices run 2.5 to 9.4 cents; three edit, two are text-only.
Video ad codec table: H.264 everywhere, H.265 only on Pinterest
Which video codec each ad platform accepts, from the vendor pages, and which Sume trim setting keeps or rewrites the codec of your source file.
Gemini CLI httpUrl or url for Sume MCP: which key to use
In Gemini CLI settings, url is the SSE endpoint and httpUrl is HTTP streaming. Sume's endpoint is streamable HTTP, so use httpUrl. Config and checks inside.
Gemini CLI includeTools for Sume MCP: expose reads, keep trust false
Gemini CLI can allowlist MCP tools with includeTools, and trust defaults to false. Use both to give an agent Sume reads without paid generation.
Gemini CLI MCP timeout is 600000 ms; Sume jobs_wait caps at 55 s
Gemini CLI waits up to 10 minutes per MCP request, but Sume cuts jobs_wait at 55 seconds. Set the client timeout low and loop on wait_slice_expired.
Group failed Sume jobs by error category with curl and jq
One curl and jq command that lists the last 100 failed jobs, groups them by error.category, code and retryable, and prints a sample job id for each group.
How many aspect ratios each Sume image model lists, vs Hy's 14
Hy Image 3.5 Preview lists 14 aspect ratios. On Sume, Nano Banana 2.1 and Ideogram list 15, Grok and Flux 13, GPT Image 2.5 only 9. Full counts by row.
Hy Image 3.5 lists a 1.5K tier. No Sume image model does
Hy Image 3.5 Preview offers 1K, 1.5K, 2K and 4K. Sume's tier lists are 512, 1K, 2K, 4K or 1K/2K only. For about 1.5K, GPT rows take 1536x1536 as image_size.
Hy Image 3.5 Preview returns base64 PNG; Sume returns a URL
Moving image code from a base64 PNG response (Hy Image 3.5 Preview on OpenRouter) to Sume means downloading data[].url. A 12-line Python version.
Korean caption fonts: 29 Hangul faces and each style's default
Sume video captions list 29 Hangul fonts under the font field. See which face each style uses by default and what the API rejects.
Korean captions on slam return 400: caption_hangul_text_latin_style
Sending Hangul text to slam, punch or tiktok-green returns 400. Sume refuses rather than burn tofu boxes. Use black-outline or a Hangul style.
Letter size 8.5x11 AI image: 1088x1408 is an exact ratio
GPT Image 2.5 on Sume needs multiples of 16. 1088x1408 matches 8.5x11 exactly (17:22) and passes the pixel box. 2176x2816 is the 2x size.
LinkedIn video ads: the first 10 seconds and a frame check with Sume
LinkedIn says to put your most impactful content in the first 10 seconds. Pull stills at 0 to 9 seconds with Sume video frames and see what a muted viewer sees.
Lyria prompt section markers: [0:00-0:30] Intro on Sume Music
Sume Music has no duration field. You shape a track with section markers like [0:00-0:30] Intro and a stated length in the prompt itself.
Meta feed sound is optional but recommended: add a ducked music bed
Meta says captions and sound are optional but recommended on Facebook Feed video. Add a music bed under a voice with Sume Timeline soundtrack and duck_db.
Music API duration_seconds rejected: set track length in the prompt
Sume Music 1.0 and the Music Router reject duration and duration_seconds. Put the length in the prompt (a 2-minute track) and use time markers.
Music prompt limit: 5000 characters is plenty for a Lyria brief
The Sume Music prompt takes 1 to 5000 characters. A full seven-axis brief with section markers is far shorter, so the limit rarely binds.
Scoring contrasting scenes: 12 BPM apart and a new lead instrument
Sume's music brief rule for contrasting scenes: change the genre family, move tempo at least 12 BPM, and swap the lead instrument. Keep them for one score.
One TTS job per sentence, then a gapless concat: the Sume recipe
Generate narration one sentence at a time with tts_create, then join the takes with timeline audio. Fix one line without redoing the rest.
Pika Starter 900 credits vs Creator 3,150: same credit price
Pika Creator costs 3.5 times Starter and gives 3.5 times the credits. The extra $25 is the commercial license, not a volume discount.
Pinterest video titles show 40 characters: burn the hook in
Pinterest shows the first 40 characters of a video ad title in the feed (30 for CJK) and hides the description. Burn the hook in with Sume captions.
Preview the first frame before an avatar render: Sume vs credits
Synthesia and HeyGen sell avatar video from monthly credit pools. Sume previews the first frame of an avatar scene before you commit to the full render.
Do punch and tiktok-green captions accept design overrides? No
In Sume video captions, punch and tiktok-green do not support design: they render on a path that reads none of the tokens. Use another style to tune.
Retry a Format 402 with the same Idempotency-Key after adding funds
A 402 at create means nothing ran and the key was released. Add funds, then resend with the same Idempotency-Key. A replay costs nothing.