Blog
Sume blog: agent-native video generation, Avatar 1.0 orchestration, and production APIs for builders and coding agents.
Sume webhooks: 10 attempts 30 seconds apart for a video receiver
A Sume job webhook is tried up to 10 times, 30 seconds apart by default, with a 10 s timeout each. What that means for a video receiver, plus a Python verifier.
A 50-page deck to AI video: five ten-slide Wan 3.0 batches on Sume
Alibaba caps document input at 50 pages. Sume wan-3.0 takes 10 reference images per job, so a 50-slide deck becomes five jobs. Cost at 6 s or 30 s per batch.
Product brochure PDF to a 30-second video: export pages for Wan 3.0
Alibaba says Wan 3.0 reads PDFs. On Sume, export the brochure pages as PNGs and send up to 10 as wan-3.0 references. $3.75 at 720p, $7.50 at 1080p for 30 s.
Add a dog or cat into an existing scene with Gemini Omni edit
Put a pet into footage you already shot: the edit prompt, why your own pet's photo cannot ride on an edit, and what to check for shadow and scale. 8 s is $1.00.
Two reference images, one clip: Omni's cat-and-yarn pattern on Sume
Google's docs show two images, a cat and yarn, producing one video. Here is the Sume request with IMAGE_REF tokens and what changes with a brand product.
Bash progress line for a Sume bulk queue from counts with jq
Poll a Sume bulk queue and print 'running: 60/100 done, 4 running, 2 failed' with curl and jq; back off, ride out 429 and 503, and exit 1 if any item failed.
Recolor sneakers to sand suede, keep the laces white: prompt test
Four prompt wordings for a colorway edit that changes one part and keeps another, run on Ideogram 4.5 at the low tier on Sume for $0.15 in total.
One approved backdrop for 200 product photos: reference edit cost
Use one approved scene as the first reference and each product shot as the second, in Ideogram 4.5 on Sume. 200 photos cost $7.50 at low and $15 at medium.
Inspect a 3-minute Short with a transcript: set duration_seconds 180
video-inspect reserves one minute of speech-to-text unless you send duration_seconds (max 600). Why a 180 hint matters, the $0.01 per minute rate, and the 400s.
Add Halloween decorations to a shop video with an Omni edit prompt
Turn a storefront or cafe clip into a Halloween version: send it as video_url to Gemini Omni Flash 1.1 with a keep-and-add prompt. Price per clip inside.
Test the hardest 5 seconds first: trim, then Recast before the full 30
Before paying $11.25 to recast 30 seconds at 768p, trim the worst 5 seconds and recast just that for $1.895. What the test can and cannot tell you.
MAI-Transcribe-2-Streaming 2.5% WER: what the test audio mix is
The 2.5% word error rate comes from a chunked-streaming index: 50% AA-AgentTalk, 25% VoxPopuli, 25% Earnings22. Here is how that maps to a recorded call.
sume login --no-browser on a remote server: approve the user_code URL
On an SSH box, sume login --no-browser prints the approval URL instead of opening a browser. After you approve the user_code, the CLI stores a CLI-scoped key.
Music job metadata: stored on the Sume job, not sent to Lyria
The metadata object on a Sume music request is stored on the job and never sent to the provider. Tag takes by scene and brief with it.
Show the product name the moment the voice says it: STT word times
Find when a voiceover says a word with Sume STT words[], then use that second as the next timeline video[].start. Offline Python script included.
Seedance 2.5 targeted editing: vendor claim vs what Sume accepts
Together says Seedance 2.5 edits specific moments after generation. Sume lists no video_url edit for it; only gemini-omni-flash-1.1 does. The reroll workaround.
GitHub Actions weekly Sume bulk queue that fails on counts.failed
A scheduled workflow that creates a Sume bulk queue, polls it through transient 429 and 503, and turns a completed queue with failed items into a red job.
Construction time-lapse from two photos with Omni first/last frame
Turn a site photo from week one and a photo of the finished building into an 8-second progress clip. Request shape, prompt, cost, and what to check.
Two-voice dialogue audio on Sume: TTS lines joined with audio concat
Build a role-play or interview track by generating one TTS line per turn in each speaker's voice, then joining up to 20 turns into one gapless file for $0.01.
Avatar clips of 60 seconds or less and the Shorts one-minute line
YouTube allows 3-minute Shorts, but claims behave differently above one minute. Why a Sume avatar clip capped at 60 seconds sits on the safe side, with sources.
Pandas DataFrame to a Sume bulk queue in 100-row chunks
Turn a product DataFrame into Sume Format bulk queues: one item per row, 100 rows per queue, a stable key per chunk, and SKU order saved beside each queue id.
video-filter invalid_filtergraph: every reason and the fix for each
A video-filter filtergraph is refused for eight reasons, from a [0:v] label to a quote character. What each one means and how to rewrite the graph so it passes.
Deno.cron weekly Sume video run with a per-day idempotency key
A Deno.cron job that starts one Sume Format run each Monday, keyed by UTC date so a double fire replays, skips overlap, and hands the result to a webhook.
Charter boat listing video from ten photos: Wan 3.0 vertical and wide
One set of ten boat photos, two Wan 3.0 jobs: 9:16 for stories and 16:9 for the listing page. $3.75 each at 720p for 30 s, $7.50 for both, with a shot order.
GET /v1/jobs/:id returns 404 for a job your teammate's key created
A Sume job id can answer 404 not_found to your key though it exists: an API key reads only jobs its own member created, and only that member can cancel.
Is image-to-video cheaper than text-to-video? Same rate
On Omni Flash 1.1, image-to-video bills the same per second as text-to-video. A 5-second 720p clip is $0.625 either way; the still adds $0.075 on Ideogram 4.5.
Speaking rate in words per minute from Sume STT word times (Python)
Compute words per minute for a recording from the words[] start and end times Sume STT returns, plus a per-minute pacing table. Offline Python, no API call.
Pan and scan a 1.85:1 trailer to vertical: one crop per shot
For a 1.85:1 trailer, trim each shot, then crop it at its own x offset: width 0.304054, x 0 to 0.695946. A three-shot plan with job count, cost and order.
What a scripted AI avatar clip cannot do: seven limits and fixes
A Sume avatar clip cannot take questions, run past 60 seconds or switch presenter mid-video. Seven documented limits, each with a workaround.
Re-upload an old vertical video: is it a Short or long-form?
YouTube's page says square or vertical uploads up to three minutes after October 15, 2024 are Shorts. What it means for a re-upload, and how to check the file.
A 60-second Seedance 2.5 piece: two 30 s jobs and one last frame
Together lists multi-round extension for Seedance 2.5; Sume documents none. Build 60 s as two 30 s jobs and a last-frame handoff: $34.68 at 720p, $7.50 on Wan.
Retry a failed Format create with the same Idempotency-Key
After a 402 or 503 on a Format create, Sume releases the Idempotency-Key. Fix the cause, then retry with the same key instead of minting a new one.
Spreadsheet to AI video: Wan 3.0 xls input vs chart images on Sume
Alibaba lists xls as a Wan 3.0 input, Sume does not. Export each chart as a PNG and send up to 10 reference images or a first and last frame.
Retouch budget: 40 photos, three revision rounds on Ideogram 4.5
A client retouch job of 40 photos with three revision rounds costs $9 at medium, $14 with a high final, and $33 at high throughout on Sume Ideogram 4.5.
Driver onboarding videos with an AI avatar: five 30-second clips
Five short avatar clips can cover a driver's first day. What each should say, how to handle a multilingual crew honestly, and the cost by tier.
Solar roof before and after clip: Wan 3.0 first and last frame, 7 s
Send the bare roof as first_frame and the finished array as last_frame to wan-3.0. A 7-second clip costs $0.88 at 720p on Sume. Rules for matching the photos.
One ad, five cities: swap the backdrop per market with Omni edit
Localize a street-scene ad by editing the backdrop per market with Gemini Omni Flash. Prompt pattern, five 8-second edits at 720p for $5.00, and what to verify.
Aerial establishing shot with Gemini Omni Flash: 1080p or 4K cost
A drone-style opening shot for a location video: prompt order, the two aspect ratios, and the 5, 8 and 10 second cost at 1080p against 4K on Sume.
Mute a clip or extract its audio? Trim audio drop vs audio detach
Video trim with audio drop returns a silent MP4; audio detach returns a separate wav or mp3 and leaves the video alone. Which to use, and the warnings.
Bedtime story video with a lullaby bed: Sume TTS, Lyria, timeline cost
Cost of a narrated bedtime story with a lullaby underneath: TTS for 3,000 characters, one Lyria 3.5 track, and a timeline render with soundtrack ducking.
video-inspect frames: at[] and fps together is a 400, pick one
video-inspect and video-frames take a list of times or a sample rate, never both. The codes, the 24-still cap, the fps 2 ceiling, and the empty object.
Stream a product feed NDJSON into 100-row Sume bulk bodies
Read an NDJSON product feed lazily with itertools.islice and write one bulk-run body per 100 rows, so a large file never has to fit in memory. Python stdlib.
video-filter dim amount 0 or 1.2 is refused: the (0, 1] range
Dim amount takes values above 0 up to 1. Zero, negatives and 1.2 return video_filter_amount_out_of_range. What it does, and how to lift a dark clip.
SUME_API_AUTH_MODE: make the Sume CLI send Bearer instead of x-api-key
The Sume CLI sends x-api-key by default. Set SUME_API_AUTH_MODE=bearer when your client or network layer expects Authorization: Bearer. Never send both.
Add my product to a store shelf photo: a two-reference edit on Sume
Place your packshot into an existing shelf photo by sending the shelf first and the packshot second to Ideogram 4.5 on Sume. Prompt, checks and costs.
Tavus PAL Maker, API, Enterprise vs the Sume avatar API: what matches
Tavus's site lists PAL Maker, a Developer API (CVI) and Enterprise. Which of them overlaps with Sume's avatar clips, and which does not. Pages read 2026-10-07.
Kling motion control silent clip with a Lyria music bed for a Short
Set keep_original_sound to false, generate a track with the Music Router, and join them on a timeline. Cost for a 12-second clip, and the YouTube rule to know.
Ruby Net::HTTP: create a Sume bulk queue and poll to an exit code
A 30-line Ruby script with only the standard library: create a Sume bulk queue from items.json, back off the poll, skip 429 and 503, and exit 1 on failed items.