Blog
Sume blog: agent-native video generation, Avatar 1.0 orchestration, and production APIs for builders and coding agents.
Wedding save-the-date: 12 seconds, 9:16, on Wan 3.0 references
A 12-second 9:16 Wan 3.0 save-the-date from up to 10 reference images costs $1.50 at 720p on Sume. Table of options.
ByteDance models on Sume: Seedance family and Seedream prices
Seedance 2.5, 2.0, Fast and Mini plus Seedream 5 Lite, 4.5 Edit and v4, with Sume prices, limits and the pick for each job. Read 2026-10-07.
Draft 20 Wan 3.0 clips at 480p, then finish the best six at 1080p
Twenty 5-second Wan 3.0 drafts at 480p cost $6.40; finishing six at 1080p adds $7.50. All twenty at 1080p cost $25.00.
Python ThreadPoolExecutor: submit 12 Wan 3.0 clips, four at a time
Stdlib-only Python: ThreadPoolExecutor with 4 workers posts Wan 3.0 jobs to /v1/videos, one Idempotency-Key per shot, results printed in completion order.
AI UGC creator package: cost basis for 3, 5 or 10 videos
Selling AI UGC videos? One $0.95 avatar and 20-second product clips cost $12.59 for three, $20.35 for five and $39.75 for ten at standard quality.
YouTube Shorts: Seedance 2.5 audio on, off or detached, same price
Shorts editors now separate audio and video tracks. On Sume a 15-second Seedance 2.5 clip is $8.67 with or without audio; detach the sound for $0.01 per job.
How much more does higher resolution cost? Wan 4x, Omni 10x, Kling 1x
Resolution multipliers on Sume's catalog: Wan 3.0 480p to 1080p is 4x, Omni Flash 360p to 4K is 10x, Kling 3 is flat across 720p and 1080p.
Charity appeal film: 3 versions of 30 seconds, 720p vs 480p
Three 30-second 16:9 versions of an appeal film cost $52.02 at 720p and $24.21 at 480p on Seedance 2.5 on Sume. How to pick which to test first.
curl -d @file strips newlines: a Sume webhook test that fails 401
Signing a fixture file and sending it with curl -d changes the bytes. Use --data-binary to test a Sume webhook receiver. Both signatures shown.
3-minute Reel from API clips: calls and cost by Sume video model
Instagram Reels now run to 3 minutes. Calls needed for 180 s at each Sume video model's longest clip, with the per-job price at 720p or 768p vertical.
Lip sync AI music video: split a song into 5 to 14.8 second windows
MiniMax H3 Max lip sync accepts 5 to 14.8 seconds of audio per clip. Split a 60-second song into five 12-second ranges: about $6.24 at 768p on Sume.
Seedance 2.5 cost per kept clip: 1 to 4 takes per keeper
What a kept 10-second Seedance 2.5 clip really costs on Sume when you reroll: $5.78 at one take per keeper, $17.34 at three, plus a 480p scouting variant.
Real estate listing: 6 room clips on Seedance 2.5 vs Wan 3.0, priced
Six 5-second room clips cost $42.66 at 1080p on Seedance 2.5, $17.34 at 720p or $7.50 on Wan 3.0 at 1080p on Sume. When the pricier model earns it.
AI music generator API key: a Python script that polls and downloads
Use your SUME_API_KEY to generate a track with the Music Router, poll the job and read the audio URL. A 28-line Python script with the $0.125 price.
One webhook handler for Sume agent, Format and schedule runs
A single receiver can handle all three Sume run types: verify the HMAC over timestamp.body, dedupe on request_id, and branch on outcome (ok, degraded, error).
Run a weekly short video series on autopilot with a Sume schedule
A weekly short-video series needs a schedule with a cron expression, an IANA timezone and a spend cap, plus an optional API trigger. How to set it up on Sume.
Cut a long talk into short vertical clips with the Sume API
Turn one long recording into short clips on Sume: video-inspect for a sentence transcript, video-trim per range, video-captions to burn the words. Rates inside.
Recreate a winning ad's structure for your product with Sume
sume-recreate keeps a reference video's scene order, beat timing and caption rhythm, casts a new presenter and generates every shot fresh. How to call it.
Stop an AI video agent overspending: the four Sume spend gates
An unattended Sume agent can only spend what four gates allow: the wallet, a per-run cap, a schedule ceiling, and a Format cap. Values and failure codes inside.
Polling a Sume agent run: next_action values and backoff
Poll Sume runs by branching on next_action (poll_status, retry_later, none), using status_url and result_url, backing off on 429 and 503.
Sume Formats that return a product still instead of a video
Nine Sume catalog Formats are declared with text input and image output, from water-splash-hero to magazine-cover-campaign. Use them for 9:16 hero stills.
A Sume bulk queue item failed: read the child receipt first
A Sume queue item only says format_run_failed. Open the child run by run_id, compare spend to the cap, read output_error, then retry with previous_run_id.
Nightly UGC ad job: Agent Completion, Format run or schedule?
Pick the Sume surface for a nightly UGC ad job: Format run when inputs change, schedule when only the clock starts it, Agent Completion for one-off tasks.
Run 300 ad variants as three bulk queues on Sume
One Sume bulk queue holds 1 to 100 Format runs, so 300 variants means three queues with their own idempotency keys, caps and failure checks. The limits to plan.
Get a structured ad pack back from one Sume Agent Completion
Bind an output_schema to a Sume Agent Completion to get hook, caption, CTA and a video URL as named fields, with the strict-subset rules and failure modes.
Which video model fits a 6, 15 or 30 second ad slot on Sume?
Video Router duration limits differ by model: Seedance 2.5 and Wan 3.0 reach 30 seconds, Gemini Omni Flash 1.1 is 3 to 10, and most others stop at 15.
Sume product commercial vs cinematic studio vs beauty studio
Three Sume catalog Formats make premium 9:16 product films: product-commercial, cinematic-studio-commercial and beauty-studio. How to call each.
Python: find the Sume Format that takes your input and returns video
Page through GET /v1/formats and filter on the io profile (input_kind, output_kind) to pick a Sume Format by shape. Includes the null-io case.
Localize one video ad into many languages with Sume batches
Two ways to localize one ad on Sume: a bulk Format batch with one item per language, or burn translated cue captions onto the existing video for $0.20 per job.
Python: turn a product CSV into Sume bulk Format queues
A short Python script splits a product manifest into batches of 100, builds one bulk body per batch with per-item spend caps, and keeps idempotency keys stable.
Restyle your own footage with the Sume restyle Format
sume-restyle keeps the motion, timing and cuts of a clip you attach, rewrites its look chunk by chunk, and reassembles it with your original audio.
Test several caption looks on one video without re-transcribing
Send source_caption_id with a new style to re-burn a video's captions in a different look. No second speech-to-text; each variant is still a $0.20 render.
Virtual try-on or virtual fitting: which Sume Format for clothing?
Both are Sume catalog Formats that take a garment still and return 9:16 video. Try-on is an outfit-change reveal; fitting is a fit preview with a small turn.
LinkedIn 'Seems like AI slop' button: what AI video posters change
LinkedIn lets members flag posts as AI slop. What that means for AI video on the feed, and a short Sume checklist to keep the work specific and yours.
TikTok's AI spam test: politics, finance and medical AI video niches
TikTok is testing account-level detection of AI spam on politics, financial advice and medical content. What an AI video team in those niches should check.
Read an AI video's transcript before posting: Sume video inspect
Spoken claims in an AI video are easy to miss. Use Sume video inspect with transcribe to read what was said before you post on TikTok or LinkedIn.
Approval gate before posting AI video: verify the Sume webhook
Hold every AI video for human review before it posts. A Node verifier for Sume's signed job webhooks that refuses an empty secret, plus the review steps.
Seven vertical AI videos a week: a stdlib Python batch with keys
A short Python script that submits seven 9:16 video jobs to Sume with Idempotency-Keys, so a rerun never bills a second job. Standard library only.
One AI video for TikTok, LinkedIn, Reels and Threads: limits table
Length, size and format limits from TikTok, LinkedIn, Meta and Threads pages in one table, and how to produce one master AI video that fits them with Sume.
LinkedIn sorts posts spam, low-quality or clear: batching AI video
LinkedIn says classifiers label each post spam, low-quality or clear in near real time. How to vary a batch of AI videos so each one stands alone.
AI presenter on LinkedIn: say who the avatar is and who backs it
LinkedIn says people come to connect with real people. How to use a Sume avatar video as a presenter without confusing viewers about who is speaking.
Cast pack for a series Format run: the 30-file attachment budget
One Format run carries 30 media files in total, at most 30 images, 10 videos and 10 audio, counting URLs in input. How to fit a cast and two locations in.
Weekly series drop: on_active_run reject stops a second render
Set on_active_run to reject and a second episode run on the same Format answers 409 format_run_in_progress instead of starting. skip records a skipped run.
Season outline as nested input: one key, 64-key limit, 2 MiB
Put a whole season outline under one key of a Format run's input. Only top-level keys count toward 64, and the body can be up to 2 MiB, so 40 episodes fit.
Does an image cost more at 16:9 than 1:1 on Sume? Flat rows vs GPT 2.5
On most Sume image rows the ratio does not change the price. ChatGPT Image 2.5 is the exception: cost follows pixels and quality. Billed figures for both.
Swap the lead in a finished AI episode: H3 Max Recast for 20 s
H3 Max Recast replaces the people in a 5-30 s source with 1-4 photos, keeping motion and sound. A 20 s source is $7.50 at 768p and $11.25 at 1080p on Sume.
Cameo in an AI episode: Avatar Face Swap (Beta), 4 to 15 s source
Sume's Avatar Face Swap (Beta) puts your avatar into a source video of 4 to 15 seconds and requires a quality setting. Use it for a cameo, not a whole episode.
Bulk queue says completed: it does not mean every episode rendered
A Sume Format queue is completed when every item is terminal, failed and canceled included. Check counts.failed and counts.canceled before publishing a season.