Blog
Sume blog: agent-native video generation, Avatar 1.0 orchestration, and production APIs for builders and coding agents.
Which tool exports 4K: Descript, Kapwing, HeyGen, Runway or Sume
4K is a plan tier at Descript, Kapwing and HeyGen. Sume gates it by request: 4K video via sume/auto, timeline edges capped at 2160, avatar video at 720p.
Korean burned-in captions in a CapCut style: Sume Hangul styles
Sume burns Korean captions with six Hangul styles and 29 open-license fonts at $0.20 per clip up to 60 seconds. Latin styles reject Korean with a 400.
YouTube AI disclosure: which avatar pipeline steps are exempt?
YouTube exempts scripts, captions and upscaling from its AI label but not realistic synthetic people. Map each step of a Sume avatar pipeline to the rule.
Does a scripted AI presenter video need YouTube's synthetic label?
YouTube asks for a label when content makes a real person appear to say something they did not. Here is how to read that for a made-up Sume Avatar presenter.
Non-skippable YouTube ads: cut a 75-second spot to 60 with Sume
Google Ads lists non-skippable in-stream at 60 seconds or shorter. Drop 15 seconds from the middle of a 75 second spot with Sume detach and Timeline: $0.11.
A 13-scene, 117-second storyboard at 720p: Wan, Seedance, Omni
Thirteen 9-second scenes at 720p cost $14.63 on Wan 3.0, $67.60 on Seedance 2.5 and $14.63 on Gemini Omni Flash 1.1; Omni needs no stitching under 10 s.
A 30-minute voiceover: the 1,800 s Timeline cap and two TTS jobs
A 1,800-second Timeline render costs $3.00 and a 30-minute read exceeds the 20,000-character TTS cap, so it takes two jobs. Per-vendor math, read 2026-10-08.
90-second explainer from fifteen 6-second scenes, plus the render
Fifteen 6 s scenes at 720p cost $11.25 on Wan 3.0 or Omni and $52.00 on Seedance 2.5; the 90 s timeline render adds $0.20 (2 billed minutes at $0.10).
A 12-second webhook handler times out at 10 s: dedupe Sume deliveries
Sume gives each webhook attempt 10 seconds and retries up to 10 times. A 12 s handler gets repeats. Verify, store job_id once, and return 2xx early.
A/B a Format run on Haiku 5.5 and Astra: use different keys
Same Idempotency-Key with a different body is a 409 on Sume. When the model field changes, derive the key from the body. A Python sketch that does.
Cheapest 1080p AI video per second on Sume, ranked, 21 s example
At 1080p Omni Flash is $0.1875 a second, H3 Max $0.20, Wan 3.0 $0.25, Seedance 2.0 $0.8505 and Seedance 2.5 $1.42155. A 21 s clip ranges from $3.94 to $29.85.
Cut out 25 product photos from an agent: rmbg_create and jobs_wait
An MCP agent submits 25 rmbg_create calls, waits with jobs_wait in two slices (20 ids, then 5), and pays 25 x $0.0225 = $0.5625. Idempotency keys inside.
Crop 1920x1080 to 9:16 around a left-third subject: x = 0.1751
Video filter crop fractions for an off-center 9:16 strip from a 1920x1080 clip: x 0.1751, width 0.3164, height 1. About $0.02 plus a $0.10 Timeline render.
Eight shots, 2 to 11 s: how each Sume video model bends the plan
A storyboard with shots of 2, 3, 4, 5, 6, 7, 9 and 11 s is 47 s on Wan 3.0 ($5.88 at 720p) but 50 s on Seedance 2.5 and 11 s is out for Omni.
Which new text models can a Sume Format run name as model?
Format runs take a catalog id as model. Haiku 5.5, GLM 5.3 Flash and GPT-6 Astra are in Sume's registry; Mistral Large 4 and GLM 5.3 Fast are not.
Gemini Flash-Lite TTS lacks 29 languages that Flash covers
Google's table lists 132 languages for Gemini 3.8 Flash TTS and 103 for Flash-Lite, so 29 are Flash-only, including Thai, Swedish and Finnish. Read 2026-10-08.
Find the six removed Sora identifiers in your repo, then move to Sume
OpenAI's removal list names the Videos API and five model ids, all dated 2026-09-24. A runnable repo scan finds them, and what changes on Sume's /v1/videos.
Before a Recast swap with hired talent: a 4-person release checklist
H3 Max Recast uses one photo per person, up to 4. A process checklist for approvals and YouTube disclosure before you spend $0.375 per second.
How long can a Sume agent run take? Timeouts for slow models
Format runs are finalized at 90 minutes, or after 10 silent minutes once past 25. Agent Completions publish no deadline. Webhooks time out at 10 s per attempt.
Kling motion control: send image_url or avatar_handle, not both
Sume's Kling 3.0 Motion Control takes exactly one visual source: image_url or an avatar_id/avatar_handle. A short Python check before the $0.1575/s submit.
Kling motion control on Sume: keep_original_sound and orientation
Two optional Kling motion control fields on Sume: keep_original_sound defaults to true and character_orientation defaults to video, or image.
MAI-Voice-2.1 ~550 ms vs Flash ~45 ms: does it matter for voiceover?
Microsoft lists ~550 ms and ~45 ms model-inference latency for MAI-Voice-2.1 and Flash. In series, 1,000 lines take 9.2 minutes vs 45 s. Read 2026-10-08.
Photo dance via hosted MCP: kling-motion-control_create needs a key
The hosted MCP exposes kling-motion-control_create as a paid tool that requires an idempotency_key. 8 s of motion control costs 8 x $0.1575 = $1.26.
Which BPM makes a 3 to 10 second Omni clip a whole number of bars?
Gemini Omni Flash 1.1 clips run 3 to 10 seconds. At 120 BPM a 10 s clip is 5 bars; at 128 BPM no whole-second length fits. A table of the BPMs that line up.
Omni Flash references: 10 images, 3 clips of 3 s, 9 s of footage
Gemini Omni Flash 1.1 on Sume takes up to 10 reference images and 3 reference clips of 3 s each (9 s of footage), with no audio reference. Cost: $1.00 for 8 s.
A Preview video model in production: pin a Sume id or use sume/auto?
Vidu calls Q4 a Preview. On Sume, pin a model id for a repeatable look, or use sume/auto, which hides the family and defaults to 720p and 8 seconds.
Python: chain generate, cutout and upscale with one polling helper
One Python helper submits any Sume job with mode async, polls status until terminal, and returns the artifact URL. Three calls chain image, RMBG and upscale.
Reconcile a mixed project with GET /v1/usage: debited, held, refunded
How to read what one job or run really cost: debited, held and refunded fields, with the arithmetic for a project where one of six shots fails and is refunded.
Replace line 7 of 12 voiceover lines: one TTS job and a concat
Fixing one 180-character line of a 12-line voiceover costs $0.02 (one TTS job at the 1-cent floor plus a $0.01 concat) vs $0.12 to redo all 12. Read 2026-10-08.
RMBG costs the same for a 24 MP photo and a 0.5 MP thumbnail
Sume remove-background is $0.0225 per image regardless of size, so resizing before a cutout saves upload time, not money. 600-image arithmetic inside.
480p to 1080p price multiplier: Seedance 5.29x, Wan 4x, Omni 5x
Going from the lowest to 1080p costs 5.29 times more on Seedance 2.5, 4 times on Wan 3.0, 5 times on Gemini Omni Flash (360p) and 3.2 times on MiniMax H3 Max.
Silent 11-second b-roll: Omni rejects generate_audio false
Gemini Omni Flash 1.1 on Sume always has audio and rejects generate_audio false, and stops at 10 s anyway. For 11 s of b-roll, Wan 3.0 at 720p is $1.375.
RMBG, image upscale and video upscale request fields side by side
The three Sume media tools take different bodies: image_url, upscale_factor 1-4, video_url, duration_seconds 1-30, enhancement_tier. One table, prices included.
Sume run webhook: 10 s timeout, so queue before you call a slow model
Sume gives your webhook 10 seconds per attempt, up to 10 attempts. Verify the signature, store the event, return 2xx, then call any slow model. Python verifier.
Does the Sume spend cap cover the agent's own model tokens?
No: Sume's cap excludes the agent's LLM turn; debited_usd_micros includes it. At list rates that turn is about $0.004 on Haiku 5.5 and $0.40 on Astra.
Sume TTS is 2.16x MAI-Voice-2.1 per character: where it still fits
At $47.50 per million characters Sume TTS costs 2.16x MAI-Voice-2.1 and 3.17x Flash, with no break-even at any length. What the extra buys, read 2026-10-08.
Upscale a 1280x720 thumbnail to 3840x2160: factor 3 or generate at 4K?
Factor 3 turns 1280x720 into 3840x2160 for $0.20; GPT Image 2.5 at 4K is $0.2225. For 36 thumbnails: $7.20 to upscale, $8.01 to regenerate.
MAI-Voice-2.1 audio in a Sume Timeline: 9 clips, 261 s
Sume does not list MAI-Voice. To use its audio in a Timeline, import the files, concat at $0.01, render 261 s as 5 minutes ($0.50). Read 2026-10-08.
Which 10 of 15 reference images to keep for Sume Omni reference video
Vidu Q4 Preview takes 15 reference images; Sume's Omni takes 10 plus 3 short clips. A role-by-role way to cut a 15-image brief down to 10 slots.
Write the Lyria brief from your shot list: a Python timestamp builder
Lyria responds to [m:ss-m:ss] section markers. A short Python script turns a 6-shot, 59-second shot list into a brief under the 5000-character limit.
YouTube AI use label on a face-swap clip: does it hurt monetization?
YouTube says its altered-content label has no monetization impact, while undisclosed realistic fakes can be penalised. What that means for a Sume face swap.
120 ms first partial vs a 600-second file job: pick the STT shape
MAI-Transcribe-2-Streaming lists ~120 ms to first partial; Sume STT is a file job capped at 600 s. Workloads, shapes and prices side by side, read 2026-10-08.
A 19-second single shot: which Sume video models take it in one job
Wan 3.0 and Seedance 2.5 take 19 s in one job ($4.75 and $27.01 at 1080p). Seedance 2.0, H3 and Omni top out at 15 s, 15 s and 10 s, so they need two jobs.
8:1 strips and 21:9 banners: which image models take them
Nano Banana 2.1 lists 8:1, 4:1, 1:4 and 1:8 on Sume; Pro stops at 21:9 and 9:16. Prices per image for each ratio choice, read 2026-10-08.
Agent Completion or Format run: where can you pick Haiku 5.5?
Only a Format run takes a catalog model id. An Agent Completion accepts model sume-agent and nothing else, so Haiku 5.5 is a Format-run choice.
Add captions to a Vidu Q4 clip with Sume: what the caption API needs
Vidu Q4 Preview can voice characters. Sume's video captions API burns captions onto any public HTTPS clip; it fails on silent clips unless you send cues.
Cheapest possible job per Sume video model: shortest clip, lowest tier
Smallest valid job: Omni $0.1125 (3 s, 360p), Wan 3.0 $0.125 (2 s), H3 and H3 Max $0.3125 (5 s), Seedance 2.0 $0.7031, Seedance 2.5 $1.0747.
Count TTS characters first: Sume counts spaces and punctuation
Sume TTS bills every character including spaces and punctuation, up to 20,000 per request. A runnable Python counter prices a script on Sume and MAI.