Guides
Sume guides: the API, Formats, agents, models, media tools and pricing, each answer dated and sourced.
Claude Code MCP behind a proxy: 403 now fails fast, allow Sume's host
Claude Code 2.1.290 and 2.1.292 stop retrying an MCP server your proxy blocks with 403. What to ask your proxy admin to allow so Sume's hosted MCP connects.
Thicken or thin bright shapes in a video: dilation and erosion by API
dilation grows bright areas and erosion shrinks them, two allowlisted filters in Sume's video filter. When they help and how to check cheaply.
HelixWorld needs ~70 GB of VRAM and has no hosted API: what to use
HelixWorld is a real-time audio-visual world model with open code but no hosted API. What it needs to run, what HelixBench measures, and the Sume job route.
HelixWorld spatial audio vs native stereo clips on Sume
HelixWorld renders camera-aligned stereo sound in real time. Sume's video rows give native stereo or synced audio per clip, not live spatial audio. Compare.
Swap the colours of a video by API: the swapuv chroma-swap look
swapuv exchanges a clip's U and V chroma channels and leaves brightness alone, a quick surreal colour look. How to run it in Sume's video filter.
Teams custom background: JPG, PNG or BMP, and what Sume outputs
Microsoft Teams accepts a custom meeting background as .JPG, .PNG or .BMP. Sume image models output png, jpeg or webp, so pick png or jpeg.
Which ffmpeg filters does Sume's video filter allow? All 66
Sume's video filter accepts a fixed allowlist of 66 pixel filters in five groups: tone, blur, geometry, time, compositing. The list, the caps, and a free check.
Zoom video background: MP4 or MOV, 360p to 1080p, from Sume
Zoom accepts an MP4 or MOV video background between 480x360 and 1920x1080. Generate a 16:9 clip on Sume at 720p or 1080p and skip the 4K trap.
AI video resolution price steps: where Sume cost jumps 2x or more
Step by step ratios between resolution tiers for 11 Sume video rows. The sharp steps are 360p-720p on Omni, 1080p-2K on Wand Vega Pro, 768p-2K on Heygen.
Community theater show promo video: poster, 8-second clip, dates
A 12-second promo for a local play: poster as first frame, a music bed, and ticket dates burned in as text. About $1.43 per cut on Sume.
Instagram Reel boost rules: 90 s, 9:16, no licensed music
Instagram's help page limits boosted Reels to 90 seconds, 9:16, and no licensed music. Three 30 s Wan 3.0 clips with generated audio fit, for $11.25 at 720p.
Poetry reading video from your own recording, captions and music
Turn a 60-second poem recording into a captioned vertical video: six image or clip slots, a ducked music bed, script-aligned captions. About $0.43 plus visuals.
Sound surcharge per second: Kling 3 and three Veo 3.1 rows on Sume
Sound adds $0.025/s on Veo 3.1 Fast and Lite, $0.07 on Kling 3 and $0.25 on Veo 3.1. Priced for 8 and 15 seconds, from the Sume catalog rates.
Gift-wrap pop-up promo clip: three cues and a pickup cutoff
A gift-wrap or alteration pop-up sells time, so the clip must show the cutoff. Build it from one photo, burn three caption cues, and check the finished file.
Lovart credits reset monthly, top-ups last 366 days: Sume's wallet?
Lovart's FAQ: plan credits reset each billing cycle, no rollover; top-ups last 366 days. Sume top-ups last 365 days; its plans carry no credits.
Nutcracker ticket promo for a dance school: one poster, one short clip
The Nutcracker premiered Dec 18, 1892 and runs about 90 minutes uncut. A dance school can animate its poster into a ticket promo and burn showtimes as cues.
App preview video that loops and works muted: first and last frame
Apple says previews autoplay, repeat and are usually muted. Use one image as first and last frame so the loop has no jump, with generate_audio off.
App Store creative assets must be 4+, even for a 17+ app
Apple's asset rule: every App Store asset must meet a 4+ rating whatever the app's own rating. A review table for generated art, and a $1.20 candidate-set cost.
App Store creative assets: no prices, discounts or URLs in the art
Can a Black Friday app asset say 40% off? Apple's asset rules bar pricing, discounts and URLs. What to leave out, and how to generate art with room for type.
Bannerbear refunds an unrendered part; Sume cancels only before start
Bannerbear says cancelling an animation refunds the part not yet rendered. Sume cancels only before generation starts, then returns 409. How the money differs.
Creatify Starter runs 1 generation at a time; Sume queues up to 6
Creatify Starter allows 1 generation at a time, Pro up to 5. Sume Free runs 1 and queues 5; Pro runs 4 and queues 20. The same 12 jobs counted on each.
Gemini 3.8 Live Avatar is enterprise-only; what Sume offers for clips
Google's Live Avatar in Gemini 3.8 Live is a real-time persona for Gemini Enterprise in 97 languages. Sume's Avatar 1.0 renders English clips. How they differ.
Google Ads promotion occasions: Q4 dates and a creative set per window
Google lists Black Friday as Oct 15 to Dec 15 and Christmas from Nov 1. See the overlaps, the 6-month rule, and how to batch one creative set per window.
Merchant Center Product Studio video vs the Sume video API
Product Studio saves 720p, caps images at 32 MB and works in six countries. When a Sume video API request fits better, read from Google's page and Sume docs.
HeyGen enterprise writes rise to 30/s; how Sume limits submits
HeyGen raised enterprise limits: writes 30 per second, polling 1,500 per minute. Sume publishes no per-second number; it uses plan queues and rate headers.
HeyGen instant voice clone audio in Sume? Only Sume-hosted audio
HeyGen's new instant voice clone can output TTS audio, but Sume's talking-still routes take only Sume-hosted audio up to 10 MB. What it blocks and what works.
Lambda Powertools idempotency and Sume's Idempotency-Key: use both
Powertools idempotency guards your Lambda handler; Sume's Idempotency-Key guards the paid submit. How they differ, and how to derive one key from an order id.
LangGraph replay re-fires API calls: key Sume submits per fork
LangGraph time travel re-runs every node after the checkpoint, API calls too. Derive the Sume Idempotency-Key from the body so replays dedupe and forks run.
Logo in the first 2 seconds plus an end card: Pinterest and Amazon
Pinterest wants the logo in the first 1-2 seconds; Amazon says beginning or end. Build both cuts from one clip with Timeline 1.0 for $0.10 each.
Mastra Connect 1.0: no MCP approval by default, Sume not listed
Mastra Connect 1.0 stopped asking approval for discovered MCP tools, and its docs list seven MCP providers, not Sume. Use MCPClient and gate paid tools.
Avatar micro twitches: what Sume's 2 fps frame check can see
Tavus says Phoenix-4.5 cut micro twitches. To check your own clip, Sume's frame sampler tops out at 2 fps; use explicit times for 24 neighbouring frames.
OpusClip API alternative: what Sume can cut and what it cannot
Sume has no highlight-finding API like OpusClip's. It trims, captions and assembles clips you pick; the source must be on media.sume.com first.
Runway fails a redirected media URL; Sume follows up to five hops
Runway's API fails a media URL that answers 3XX. Sume follows up to 5 redirects if every hop is public HTTPS. Compare both URL rule sets before you port.
Same video model, different limits: OpenRouter vs Sume rows
Ten video models are on both OpenRouter and Sume; eight differ in duration or resolution. Veo, Kling, Seedance, Hailuo, Grok side by side, read 2026-10-11.
Stability's Stable Image edit tools, matched to Sume image routes
Which of Stability's Stable Image edit, upscale and control tools have a Sume route: mask edits, background removal and upscale yes, outpaint and sketch no.
Submagic caps video at 2, 5 or 30 min; Sume prices up to 60 seconds
Submagic allows 2, 5 or 30 minute videos by plan. Sume's $0.20 caption job is priced for videos up to 60 seconds. Which one fits which video length.
Synthesia Syren writes video as code: what Sume has instead
Syren is Synthesia's prompt-to-video agent in early access. Sume has no equivalent; it has a script-to-avatar API and a timeline. The gap, mapped.
Syren mixes the first 24 audio tracks; Sume's timeline takes 20 parts
Syren mixes only the first 24 audio tracks in an export. Sume's timeline takes one spine in up to 20 gapless parts plus one bed. What fits where.
Syren can't use your own avatar or cloned voice; Sume has a handle
Synthesia says own avatars and cloned voices need its regular editor, not Syren. Sume builds an avatar from a photo; voice cloning is app-only.
Syren's 3-minute cap and 30-second avatar scenes vs Sume's 60 s
Syren runs up to 3 minutes but avatar scenes over 30 s can time out; Sume takes 4-60 s per avatar job. How to plan a 3-minute presenter video.
Tavus adds Agora support; Sume has no real-time transport to plug in
Tavus announced official Agora support on Oct 8. Sume has no WebRTC, Agora or live-stream path for avatars; it returns finished clips as job results.
Three app previews show in search: queue three hook variants at once
Apple shows up to three app previews in search results. Queue three hook variants of one app video as one Sume bulk run.
Usage hold_counts: open, browser_session, billing_pending
A scoped Sume usage summary can show final false with money held. hold_counts says whether a turn runs, a Browser session is open, or work is being priced.
One agent turn's total cost: pass its job id to /v1/usage
Pass an agent turn's job id as job_id on GET /v1/usage and Sume sums the turn's own model row plus every job it commissioned, with turn_job_id on each row.
Ten sub-cent usage rows: why cents fields never match debited_usd
Sume prints three roundings for the same money: credit_amount rounds up, billable_amount_usd_cents rounds half-up, debited_usd rounds once. Sum micros instead.
Voice actor or AI voice for a 30-second ad? What an API covers
Hire a voice actor or use AI text to speech for a 30-second ad: what Sume's TTS API does, what it cannot do, and the $0.315 audio bill for three takes.
WaveSpeed levels come from one top-up; Sume concurrency is plan-only
WaveSpeed sets Bronze to Ultra by a single top-up amount. On Sume a top-up never raises concurrency; the plan does (1, 4, 8, 20). What that means for budgets.
What one script_run call cost: the usage script_runs array
Sume's usage summary lists each script_run call with its rows, debited, held and refunded micros, so a fan-out of TTS or image jobs has one price.