Blog
Sume blog: agent-native video generation, Avatar 1.0 orchestration, and production APIs for builders and coding agents.
One 1920x1080, 24 fps, 20 s master for Amazon, LinkedIn and Threads
A single 20-second 16:9 MP4 at 24 fps sits inside the listed rules for Amazon Sponsored Brands video, LinkedIn video ads and the Threads API. The Sume settings.
Hook matcher rules: when mcp__sume__ names are exact, not regex
Claude Code reads a matcher of letters, digits, underscores and pipes as exact names, and other characters as regex. How to write it for Sume tools.
Claude Code http hook: send paid Sume call records to an audit log
An http hook POSTs hook input to a URL, with headers from allowed environment variables only. Log Sume tool names and idempotency keys with a tiny receiver.
Claude Code prompt hook: 30 s default, a model checks a Sume render
A prompt hook asks a model to review a tool call, with a 30-second default timeout. When that suits a Sume render and when a script check is the better gate.
Claude Code hook: exit 2 or a JSON deny for a paid Sume call?
Exit code 2 blocks and cannot be overridden by JSON; exit 0 with permissionDecision deny carries a reason. A runnable Python guard for Sume's paid tools.
Dry-run 200 holiday renders free with the Sume Timeline plan call
POST /v1/timeline-1.0/plan is unbilled and returns billable_minutes and a cost estimate. A short Python loop totals the bill before you submit a batch.
Vidu Q4 Preview takes 15 reference images: Sume rows take 9 or 10
Vidu Q4 Preview accepts 1 to 15 reference images. On Sume the largest documented image-reference cap is 10 (Wan 3.0, Omni); H3 takes 9.
Split a 15-reference storyboard into two Wan 3.0 jobs on Sume
A 15-image brief does not fit one Sume Wan 3.0 job, which takes 10 references. Split it into two jobs in Python; two 8 s 720p jobs cost $2.00.
How many words fit a 5 to 14.8 s lip-sync segment: measure first
At the 2.8 words a second Sume's avatar planner assumes, 5 to 14.8 s is 14 to 41 words. TTS pace varies: measure the audio before setting duration_seconds.
Group TTS sentences into 5 to 14.8 second lip-sync segments in Python
A short Python planner that groups sentence durations into segments the Sume H3 Max lip-sync route accepts (5 to 14.8 s) and prints the 768p cost of each.
Podcast sponsor clip: restyle captions without a second transcription
Restyle a captioned podcast clip for each holiday platform with source_caption_id. Sume reuses the stored word timings, so speech-to-text does not run again.
LinkedIn wants a text-only SRT; Sume burns captions, so build one
LinkedIn video ad captions must be a text-only SRT. Sume burns captions into pixels and has no SRT export. Build the file from video-inspect segments.
H3 Max lip sync: 480p drafts then a 1080p final for a 6-second line
A 6-second line costs $0.375 at 480p and $1.20 at 1080p on Sume H3 Max lip sync. Drafting pays only past 1.45 expected attempts. The math, with code.
Codex 0.161 defaults to GPT-6.1 Sol: recheck how it calls Sume tools
Codex 0.161.0 makes GPT-6.1 Sol the default model in bundled and Amazon Bedrock catalogs. Re-run a free Sume read and a dry run before trusting paid calls.
bun test for a Sume webhook verifier: five cases to run on Bun 1.4.2
Five bun:test cases for a Sume webhook verifier: fresh, tampered, stale, rotation and an empty secret, plus a pin to Bun 1.4.2 or later. Run on Bun 1.4.0.
input_references by Sume image model: 0, 5, 10 or 16 images
Sume image rows take 0, 5, 10 or 16 reference images. A table for every row, plus how Nano Banana 2.1's 14 and Hy Image 3.5's 20 compare.
Node 26.11 ships Undici 8.11.2: tell a Sume poll timeout from an abort
Catch TimeoutError separately from AbortError when a Sume status poll in Node fetch times out, and know which Sume calls are safe to repeat after that timeout.
Kling lists MCP and CLI tools for batch video; Sume's hosted MCP
Kling's home page mentions MCP and CLI tools for batch generation. Sume's hosted MCP at mcp.sume.com has generate_video; its Kling row is kling-3 only.
Resume a Gemini CLI session after a Sume render: keep the job ids
Gemini CLI 0.64 preview stops deleting resumed session history. How to pick a Sume video job back up after a restart with jobs_list and jobs_wait.
Nano Banana 2.1: Google vs Sume price for one 1K image
Google lists $0.0336 for a 1K Nano Banana 2.1 image; Sume bills about $0.10. Why the numbers differ, and which one your invoice uses.
Check probe.has_audio before captions to avoid caption_no_speech
A silent clip fails Sume speech captions with caption_no_speech. Run video_inspect with frames false, read probe.has_audio, then pick STT captions or text cues.
Griffin-Lite 26 of 54: error bars and a viewer test for avatars
Tavus says 48% of 54 people took Griffin-Lite for real. With 54 viewers the 95% range is 35% to 61%. A Python check and a test plan for Sume avatar clips.
Deno 2.9.7: scope permissions to a Sume key check on GET /v1/me
Run a Sume API key check in Deno 2.9.7 with allow-net limited to api.sume.com and allow-env limited to the key, and see what the permission error looks like.
Codex 0.161: denied reads stay denied, so a Sume upload may fail
Codex 0.161.0 lets approved filesystem escalation widen writes while keeping denied reads. Why a local file upload to Sume can still fail, and how to fix it.
Holiday music bed under a voice: soundtrack duck_db 0 to 20
Sume Timeline takes a soundtrack bed with gain_db, loop, fade_out_seconds up to 10 and duck_db 0 to 20 under a voice spine. Python plan call and errors.
Gemini CLI 0.64 preview moves settings to V2: recheck Sume
The 0.64.0 preview moves settings from V1 to V2. After you upgrade, confirm the Sume hosted MCP server still connects with a free read before any paid call.
Sound-off holiday video: burn your own text cues with Sume captions
For a silent product clip, send cues with text, start and end to POST /v1/video-captions. No speech-to-text runs. Code sample, $0.20 per job, and the limits.
node:test mock.method on fetch: test a Sume status poll offline
Node 26 goes LTS this month. Test a Sume job status poll with node:test and mock.method on fetch: four cases, no network, no key, no third-party test library.
Captioning a 3-minute Short: Sume prices 60 s, so plan 3 jobs
Sume's captions price is $0.20 per job for videos up to 60 seconds. A 180-second Short is three 60-second parts, $0.60 plus trim. What the docs leave out.
LinkedIn ratios 1:1, 4:5, 9:16: Sume video-filter crop fractions
LinkedIn lists 16:9, 1:1, 4:5 and 9:16 video. Centered crop fractions for a 16:9 or 9:16 source, plus a short Python function that prints the video-filter ops.
Is Vidu Q4 Preview on Sume? Check the live video catalog in Python
Sume's documented video catalog lists no Vidu model. A short Python check against GET /v1/videos/models confirms it before you build, plus the closest rows.
Airtable counts failed runs too: a 100-item Sume batch on Free
Airtable counts a run each time a trigger fires, even if actions fail. If each of 100 Sume deliveries triggers one, the Free plan's 100 monthly runs are gone.
n8n test URL or production URL: which goes in a Sume webhook_url
Put the n8n Production URL in Sume's webhook_url. The Test URL only works while the editor is listening, so a holiday batch would burn retries against it.
Nano Banana 2.1 or Pro? The billed gap at 1K, 2K and 4K on Sume
Pro costs 1.25x to 1.9x more than Nano Banana 2.1 per image on Sume. Ratios, tiers and references side by side, with the cases where Pro still earns it.
Runway GWM Worlds 2 is real-time 720p; Sume video jobs are async
Runway's GWM Worlds 2 (Sept 3) streams steerable 720p 24 fps video in real time. Sume video is submit-and-poll jobs; here is when each fits.
Threads API video rules: H.264, AAC, edit lists, and what Sume covers
Meta's Threads page lists MP4 or MOV, H.264 or HEVC, AAC, no edit lists and a front moov atom. Which of these Sume documents, and which you must check.
Vercel's 300 s default: does a 30 s Sume image sync call fit?
Yes: with fluid compute the default is 300 seconds, far above the 30-second Sume image wait. Branch on 200, 202 and 502, and poll video jobs instead.
Hedra Elements library vs Sume avatar handles: reusing one character
Hedra lists Elements as a library of characters, outfits and environments. Sume reuses a character through an avatar handle created once. How the two differ.
Promise.allSettled for many Sume job statuses: isolate one failure
Check many Sume job statuses at once with Promise.allSettled so one 429 or network error is reported on its own row and does not hide the other results.
Creatify Boreal-H3 runs 5 to 15 s; Sume H3 Max lip sync stops at 14.8
Creatify's changelog lists Boreal-H3 at 5 to 15 seconds. Sume's H3 Max lip sync accepts 5 to 14.8 and rejects 15. How to trim a 15.0 s line, and the cost.
Move burned captions: design.placement.anchor_ratio on Sume
Sume caption styles have default heights. Send design.placement.anchor_ratio to move the line, as a fraction of frame height. Which styles accept it.
Sume with Zapier, Make or n8n: no app in the docs, so use HTTP
Sume's docs list no Zapier, Make or n8n app. Start Formats over HTTP, receive results on a webhook_url, and poll as a backup. Limits of each tool.
Vidu Q4 Preview makes 16 s clips: which Sume rows go past 10 s?
Vidu Q4 Preview runs 3 to 16 seconds. On Sume, Wan 3.0 (2 to 30 s) and Seedance 2.5 (4 to 30 s) go past 16; Omni stops at 10 s. Prices for 16 s inside.
Format run usage after a cancel: debited, held, refunded and final
After you cancel a Sume Format run, usage shows the spend. How debited, held, refunded and final differ, and when the number is settled in a holiday batch.
Sume webhook receiver: return 204 for an unknown event, not a 500
Route Sume webhook events through a handler table, answer 204 for any event you do not know, and keep the webhook.test event out of your job table. Node sample.
Cloudflare Workers CPU time vs wall clock for a Sume webhook
A Sume webhook verifier on Workers spends CPU only on the HMAC and body read, not on waiting. See the limits and a WebCrypto verifier.
Cartesia Startup and Scale vs Sume TTS: break-even in characters
Sume TTS costs $0.0475 per 1,000 characters. Cartesia's Startup plan breaks even near 1.03 million characters a month, Scale near 6.29 million. Math inside.
Threads video aspect ratio runs 0.01:1 to 10:1: Sume Timeline sizes
Meta's Threads page allows ratios from 0.01:1 to 10:1 with 9:16 recommended and 1920 px max width. Sizes that Sume Timeline can output inside those bounds.