Blog
Sume blog: agent-native video generation, Avatar 1.0 orchestration, and production APIs for builders and coding agents.
Wan 3.0 reference clips: four 4-second videos break the 15 s cap
Wan 3.0 accepts up to 5 reference videos totalling 15 seconds at 16 fps or more. Four 4-second clips total 16 seconds and fail; three 5-second clips pass.
Which Sume wait mode fits which job: image, video, avatar, swap
Sume's async, sync, subscribe and webhook modes differ only in how you learn the outcome. A table by job length, and why 30 seconds is a request budget.
YouTube can add an AI label you cannot remove: C2PA 2.1 and Sume jobs
YouTube auto-labels files with Content Credentials 2.1+ and can add a label creators cannot remove. What to record per Sume job, and what the docs do not claim.
YouTube's in-Shorts AI clips are 10 s and auto-labelled; yours are not
YouTube's Shorts AI tools make clips up to 10 seconds, auto-disclosed. A clip made on Sume and uploaded is not auto-labelled; you set the answer yourself.
Library song capped at 90 s in a 3-minute Short: 90 seconds left bare
YouTube caps most Shorts Audio Library songs at 90 seconds in a 3-minute Short, some at 60 or 30. Uncovered time per cap, and how to score the rest on Sume.
Agency Batch of 10 Clips: Draft at 480p, Final at 1080p, Priced
Budget a 10-clip client batch on Seedance 2.5: 480p drafts, two revision rounds on 4 clips, then 1080p finals, versus rendering everything at 1080p.
App Promo Half-Banner: A Headline Still Over a Screen Recording
Stack a headline image above an app screen recording in one 9:16 shot with timeline compose: layout ratio, overlay mode, output size and a $0.02 price.
Avatar Presenter Training Video Over 60 Seconds: Split Into Jobs
A single avatar video job covers 4 to 60 seconds. Plan a 3-minute presenter video as three jobs and stitch them: costs at standard, plus and max quality.
Face Swap a 15-Second Clip With a Consented Avatar: What It Costs
Beta Avatar Face Swap on Sume applies a ready avatar to a 4 to 15 second source video. Price at standard, plus and max, the avatar fee and the consent rules.
Narrated Lesson Video From 10 Slides: A Seven-Minute Cost Breakdown
A 7-minute course lesson from a script and 10 slide images: TTS, gpt-image-2.5 slides and a Timeline 1.0 render, with each line of the bill worked out.
Localize One Ad to Korean: Burned-In Captions With Hangul Styles
Add Korean captions to an ad: why Latin styles return a 400, which Hangul style to pick, how a restyle skips a second transcription, and what a job costs.
Shadowing Audio for 30 Sentences With script_run: Cost and Setup
Make 30 sentence-length audio clips for a language lesson with one script_run program and tts_create: the cost for 80-character lines and the limits.
YouTube Shorts Over 60 Seconds: Why a Music Claim Blocks the Video
A YouTube Short longer than one minute with an active copyright claim is blocked worldwide. What that means for music choices, with Sume's Music Router noted.
Swap a Product in an Existing Clip With Omni Flash Video Edit: Cost
Edit a finished clip instead of regenerating it with Gemini Omni Flash 1.1 video-to-video on Sume: what you send, what you cannot, and prices for 3, 5 and 10 s.
Talking Still Testimonial: VEED Fabric 1.0 or H3 Max Lip Sync, Priced
Make a 14.8-second talking-head clip from a still and a TTS line on Sume: Fabric at 720p versus MiniMax H3 Max Lip Sync at 768p, with each price multiplied out.
TikTok Ad Teardown Before You Remake It: Trending Search and Ingest
Research what works on TikTok, then read one reference clip into shots, on-screen text and audio facts before a remake brief. Sume costs and dest-first limits.
TikTok Non-Spark Ad Upload Specs: Check Your Render Before You Upload
TikTok non-Spark ads need 9:16 at 540x960 or more, up to 500 MB, 516 kbps or higher. Use Timeline 1.0, video trim and video inspect to meet and check them.
Virtual Staging 12 Empty Rooms With an Image API: Cost by Model
What it costs to stage 12 empty-room photos with image edits on Sume: gpt-image-2.5 medium and high, and Nano Banana 2.1 at 2K, with the multiplication shown.
Format API: webhook or polling for a run that takes 15-30 minutes
A Sume Format run returns 202 and finishes in minutes, not seconds. Take the result by signed format.run.terminal webhook or poll the run; no SSE stream.
Queue 100 Format runs at concurrency 16: 7 waves, one idempotency key
Sume bulk runs accept 1 to 100 items and a concurrency window of 1 to 16. The queue has no webhook, and completed does not mean all succeeded.
90-minute webinar live captions: AssemblyAI bills open socket time
AssemblyAI bills streaming by open WebSocket time, not audio sent. A 90-minute webinar is $0.675 on U3.6 Pro Realtime; Sume STT has no live mode.
Crop an avatar video to 1:1: video-filter fractions for every ratio
Exact x, y, width and height fractions that turn a 9:16, 3:4, 4:3 or 16:9 Sume avatar clip into a centered square with video-filter, $0.02 a job.
Dim an avatar clip to 0.45 for overlay text: run the free check first
Darken a Sume avatar video with video-filter dim 0.45, run the unbilled /check call first, then pay $0.02 for the encode. What dim does and what it refuses.
Lip sync an 11-second mascot line: H3 Max 480p, 768p, 1080p vs Fabric
An 11-second line on MiniMax H3 Max Lip Sync costs $0.6875 at 480p, $1.10 at 768p, $2.20 at 1080p; Fabric 720p is $2.0625.
A 1,300-second narration fails tts_duration_exceeded: split the script
Sume TTS fails a request whose audio runs past 1,200 seconds with tts_duration_exceeded and captures no credit. Split the script at sentence ends.
150-second YouTube Short from three avatar jobs (60+60+30): cost
A 150-second avatar Short built from 60, 60 and 30 second jobs costs $27.60 at standard video, plus $0.125 music and $0.30 Timeline render.
3840x2160 YouTube thumbnail on GPT Image 2.5: valid size, 2 MB trap
3840x2160 is a legal GPT Image 2.5 custom size on Sume at exactly 8,294,400 pixels. YouTube recommends it, but mobile uploads stop at 2 MB, so check the file.
90-second founder update: two 45 s avatar jobs plus a Timeline join
A 90-second avatar update split into two 45-second jobs costs $16.76 at standard: $16.56 for video plus $0.20 for two Timeline minutes.
Before a 20-clip MCP burst: dry_run, admission preview, max_spend_usd
A single Sume MCP create needs none of these. A 20-job burst should use dry_run or generation_admission_preview, set max_spend_usd, and wait in one jobs_wait.
Can a 4:5 video be a YouTube Short? The page says square or vertical
YouTube's Shorts page says square or vertical and up to 3 minutes, and lists no ratio chart. A 4:5 feed render is taller than wide; how to make one on Sume.
Run an Agent Completion nightly from CI with curl and a spend cap
A 25-line bash job that starts a Sume Agent Completion with a cap and an idempotency key, polls the receipt, and fails the build unless the run completed.
ElevenLabs v4 vs v4 Turbo promo prices, and where Sume TTS sits
ElevenLabs v4 is $0.022 and v4 Turbo $0.011 per 1,000 characters until Oct 12; the struck-through prices are $0.08 and $0.04. Sume TTS is $0.0475 in between.
Facebook Feed's 241-minute video ad limit vs Sume's 30-minute render
Facebook Feed video ads can run 241 minutes; Sume Timeline renders 30 minutes per job. Covering 241 minutes is nine jobs and $24.10 at the public rate.
Four status vocabularies in the Sume API: job, resource, run, delivery
Jobs say queued and processing, resources say ready, deliveries say exhausted, queue-shaped fields say IN_QUEUE. A table so a client branches on the right one.
gemini-3.1-flash-image deprecated: Nano Banana 2.1 on Sume
Google deprecated gemini-3.1-flash-image on Oct 6 when Nano Banana 2.1 went GA. Which ids Sume accepts, what it bills per image, and what Google lists.
Gemini Omni edit mode on Sume: video_url only, 720p, no aspect_ratio
Omni edit mode takes a prompt and one video_url. Resolution defaults to 720p, aspect_ratio is rejected, and no image or reference field may be added.
Idempotency key from the order id, not a fresh UUID per attempt
A random UUID generated inside the retry loop gives every attempt a new key and a new paid job. Derive the Sume Idempotency-Key from the order instead.
Lint a video against TikTok, Reels, Feed and Shorts limits in Python
A 24-line Python check for duration, file size and average bitrate against the limits TikTok, Meta and YouTube publish, plus where Sume jobs fit before it.
MCP read-only OAuth session: which Sume tools you do and do not see
With OAuth mcp:read, hosted Sume MCP hides write and paid tools and returns insufficient_scope if you call them. What tools_list shows with Write off and on.
Meta lists Reels ads from 0 seconds; Sume's shortest trim is 0.2
Meta's Reels ad page lists lengths from 0 seconds, Feed from 1. Sume's shortest trim is 0.2 s and shortest Timeline render 1 s; shortest clip per model.
MiniMax H3 Max vs Wan 3.0 at 1080p: 5, 10 and 15 seconds on Sume
On Sume, MiniMax H3 Max at 1080p costs $0.20 a second and Wan 3.0 costs $0.25. For 5, 10 and 15 seconds: $1.00/$2.00/$3.00 vs $1.25/$2.50/$3.75.
Nano Banana Pro 4:5 at 1K is 928x1152, not 1080x1350: the 1.16x gap
Sume's docs say Nano Banana Pro sends 4:5 at about 928x1152 at 1K, and exact 1080x1350 comes from a post-step. That is 1.16x wide and costs $0.1875 per image.
Non-skippable YouTube ads run 15 to 60 s: how many Sume clips make 60?
Google lists non-skippable in-stream ads at 15 to 60 seconds. No Sume video model makes 60 s in one clip: clips needed per model, with prices.
One 180-second vertical master: which ad placements take it
A 3-minute 9:16 file fits a YouTube Short, TikTok, Reels and Feed ads, but misses Shorts ads and bumpers, and is not under 3 minutes for skippable in-stream.
OpenAI's three tiers vs Sume plan concurrency of 1, 4, 8 and 20
OpenAI cut API usage tiers from five to three on Oct 6. Sume sets concurrency by plan, not spend. The plan numbers and the queue math, side by side.
Porting chat history to Agent Completions: assistant turns fail
Sume Agent Completions take OpenAI-style messages but return 400 on any assistant turn and start each call in a new thread. What to send instead.
Python asyncio loop for a Sume job: next_poll_after_seconds
A runnable httpx and asyncio loop for GET /v1/jobs/{id}/status that honors next_poll_after_seconds, backs off otherwise, and leaves the job running on timeout.
Recast 3 old 10-second clips with the face swap beta: cost by tier
Face swap beta reserves 15 seconds of the tier price per run, so 3 runs cost $8.28 standard, $11.025 plus, $24.75 max.