Guides
Sume guides: the API, Formats, agents, models, media tools and pricing, each answer dated and sourced.
TTS latency vs quality: Sume has async jobs, no streaming tier
Sume TTS is an async file job on Sonic models at one price. There is no streaming or latency tier, so you trade by script size and wait mode, not model.
Sume TTS transcript_receipt: it proves the text, not the pronunciation
A source-bound Sume TTS job returns a transcript_receipt with a SHA-256 of the submitted text. It proves what was sent, not how it was spoken. Listen too.
TTS sentence_ids: max 1,000 per request vs the 20,000-character cap
Sume TTS transcript_source takes 1 to 1,000 sentence ids, but the 20,000-character limit usually binds first: only sentences under 20 characters hit 1,000.
Sume TTS voice.id 400: why 'alloy' fails and a UUID works
Sume TTS voice.id must be a UUID or a voi_ library id, not a voice name. Names borrowed from other vendors return a 400 before any work starts.
Sume TTS: webhook_url, sync wait or polling? The four modes
Sume TTS jobs take mode async, sync, subscribe or webhook. A webhook needs a public HTTPS webhook_url, and sync waits at most 30 seconds.
Upgrading Sume Pro to Scale mid-month: what happens to unused usage
An upgrade starts a new billing cycle now and grants the new plan's full included usage. Unused usage from the old plan is superseded, not added.
Veo 3.1: how many Lite takes cost one Standard take (5 at 1080p)
At 1080p one Standard take costs 5 Lite takes or 3.3 Fast takes; at 720p, 8 Lite or 4 Fast. Break-even math from Google prices, read 2026-10-06.
video_analyze or video_segment not in Sume tools_list? Dev host only
Sume's docs list video_analyze and video_segment as dev-only, shown only when a flag is on. On mcp.sume.com use video_inspect for what is in a clip.
video_url on Sume video: only Omni edit, Genjutsu and Recast take it
On the Sume video API, video_url is a source clip, taken by gemini-omni-flash-1.1 (edit), higgsfield-genjutsu and h3-max-recast. Others use reference_video_u...
Wan 3.0 file_url, web_url, enable_thinking: not in Sume v1
Sume's wan-3.0 id does not expose file_url, web_url or enable_thinking, and provider.options must stay empty. Use references and prompt fields instead.
Wan 3.0 reference videos: 5 clips, 15 s total, 16 fps floor
Wan 3.0 on Sume takes up to 5 reference videos, 15 seconds combined, each at 16 fps or higher. Check clip count, length and frame rate before you submit.
Sume webhook blocked target is never retried; a DNS miss is
A blocked Sume webhook target fails once and is never retried, while a hostname that does not resolve is retried. What each looks like and how to recover.
Sume webhook_url with :8443 is rejected: HTTPS default port only
Sume rejects any webhook_url with an explicit port such as :8443. Use the default HTTPS port, or poll instead. What the check accepts, and where it runs.
Sume webhook hostname with a private DNS answer is blocked entirely
If any A or AAAA answer for your Sume webhook_url hostname is private, the whole target is blocked. How the resolve-once check works and how to debug it.
What to log from a Sume MCP agent: ids and status, never signed URLs
Log request ids, job ids when needed, high-level status and sanitized media metadata. Keep API keys, signed URLs, raw private media URLs and transcripts out.
What YouTube's Shorts and TikTok's ad pages list, and what they omit
The two help pages I could read list limits unevenly: no codec on either, no file size on YouTube's. Meta and Instagram pages would not load.
Which Sume credits are spent first: included usage, top-ups, grants
Sume spends plan-included usage first, then top-up credits, then grants and signup credit, each by expiry date. Why the order matters for your balance.
Which wait to use for a 30 second clip: sync, webhook or a poll
A table of the waits Sume documents for a 30 s clip: a bounded sync of up to 30 s, webhooks with retries, and a roughly 20 minute client deadline. Which fits.
Which workspace does a Sume MCP call spend from? Key or sign-in
The API key or the signed-in account selects the workspace; Sume tools never take a workspace id from the prompt. Confirm it with account_me before you spend.
x-sume-webhook-timestamp is Unix seconds: the Date.now() mistake
Sume's x-sume-webhook-timestamp is Unix seconds. Comparing it to Date.now() in Node is off by 1000 and rejects every delivery. A short, tested fix.
Yearly budget for 10 ten-second AI videos a week, ten models
Ten 10-second videos a week cost $1.30 a week on Grok Imagine and $57.80 on Seedance 2.5 at 720p on Sume. Per-clip, weekly and 52-week totals by model.
YouTube Shorts max out at 1080p: what resolution to request on Sume
YouTube's Shorts page lists a 1080p maximum. Ask Sume for 1080p, not 4K: only Gemini Omni Flash 1.1 offers 4K, and the extra pixels have no Shorts use.
YouTube Shorts up to 3 min: how many Sume clips each model needs
YouTube's Shorts page says up to 3 minutes. One Sume request tops out at 30 s, so a full-length Short is 6 to 18 clips plus a Timeline render.
Zed tool permissions: confirm by default, rules per Sume tool
Zed confirms every tool action by default and can allow or deny one MCP tool by name. Keep Sume's paid generators on confirm while reads run freely.
GPT Image 2.5: image_size wins over aspect_ratio, plus the pixel rules
Send aspect_ratio 1:1 and image_size 1280x720 to GPT Image 2.5 on Sume and the pixels win. The multiples-of-16, 3:1 and pixel-count rules, with a validator.
Ideogram 4.5 vs GPT Image 2.5 for 9:16 covers with title text
For a 9:16 cover with a title, Ideogram 4.5 runs $0.0375 to $0.275 and GPT Image 2.5 $0.004 to $0.142 on Sume by quality tier. Compare before choosing.
One prompt, three ratios: a 1:1, 16:9 and 9:16 set priced by model
All 19 Sume image models list 1:1, 16:9 and 9:16. A three-image set costs $0.075 on Qwen Image, $0.137 on GPT Image 2.5 high, $0.56 on Nano Banana Pro.
Check a reference URL before an image edit: HTTPS, HEAD, content type
Sume rejects non-public or non-HTTPS reference URLs with a 400. A 15-line Python preflight checks scheme, HEAD status and image content type before you spend.
Which Sume image models reject input_references: five text-only rows
Five of 19 Sume image models are text-to-image only and return 400 unsupported_parameter if you send input_references. The other 14 take 5, 10 or 16 references.
Live sale replay: three highlight trims, keyframe then exact
Cut three highlights from a live-sale replay under 30 minutes with Sume video-trim: keyframe for a fast first look, exact for the ad, $0.02 per cut.
Video filter: a live replay over 300 seconds needs a trim first
Sume video-filter takes a source of 300 seconds or less and returns output_duration_exceeded above it. Trim the live replay highlight first, then dim or crop.
Batch of 100 image covers on Sume: waves by plan, from Free to Scale
A 100-image batch fits one wave on Scale (120 accepted) but needs 5 waves on Pro (24 accepted). Capacity by plan, wave_size_hint and the 429 queue_full rule.
Scheduling: TikTok 10 days, Meta One stories 30. Clips to render
TikTok schedules posts 15 minutes to 10 days out and locks them; Meta One Advanced schedules stories 30 days out. How many Sume clips to have ready for each.
Meta One Advanced schedules stories 30 days out: render a month first
Meta One's Advanced plan ($49.99 a month) can schedule stories up to 30 days ahead. Render 30 vertical clips in one pass with Sume Timeline for about $3.
Do you need Meta One to post Reels? No: the core stays free
Meta says the core experience stays free. Meta One adds plans from $2.99; creator tiers start at $14.99. What changes for the video files you render with Sume.
A 5xx on a Sume paid submit never proves no job: retry with the key
On Sume, only a validation, authorization or balance error proves a paid create was refused. A 5xx does not, so retry with the same Idempotency-Key.
Why a voiceover on an AI video clip never lip-syncs
Video models do not lip-sync to TTS or a later voiceover. Per Sume's docs a talking face comes from Fabric, H3 Max Lip Sync or an avatar talking-video job.
One 45-second voiceover into 15 and 30 second cuts with audio split
One Sume timeline-audio split job can slice a 45-second voiceover into a 15-second and a 30-second file for $0.01. Cut on a sentence end.
Korean caption fonts on Sume: why weight-shift needs Pretendard
Sume's weight-shift and korean-ad caption styles animate a font weight axis that only Pretendard has. Other faces keep color and scale emphasis only.
Why Sume's scheduled-run API does not return instructions text
The public scheduled-run shape omits the instructions text. Read and edit it in the dashboard; schedules have no MCP tool or write endpoint.
VoltAgent MCP 30000 ms default timeout vs Sume jobs_wait
VoltAgent's MCP timeout defaults to 30000 ms; Sume's jobs_wait holds 50 to 55 seconds. Raise it past 55000 and re-wait on wait_slice_expired.
Voiceover longer than your clips: Timeline pads, loops, 0.5 s rule
If the voiceover outruns your clips, Sume Timeline still renders to audio.duration_seconds: short sources pad or loop, and coverage may stop 0.5 s early.
Crop to 1:1 and dim a bright studio clip: read ops_applied
White studio clips wash out captions. Sume video-filter crops 16:9 to 1:1 and dims 15 percent in one $0.02 job. Check it free first, then read ops_applied.
Plan a Black Friday spot free: billable minutes and cost in micros
Sume timeline-1.0 plan compiles a Black Friday spot without a job or a charge and returns segment_count, billable_minutes, and estimated_cost_usd_micros.
Video filter crops 16:9 to 9:16 only up to 300 seconds of source
Sume video filter takes a source of at most 300 seconds. To crop a longer 16:9 recording to 9:16 for Shorts, trim first, then crop with x 0.3418 width 0.3164.
Cutdown batch for Cyber Monday: check trim rows before posting
Cut a Cyber Monday ad into 6, 15, and 30 second versions: a short Python check catches Sume video-trim range errors in each row before you pay $0.02 a job.
15-second AI video via API: price on nine Sume model settings
A 15-second AI video via API runs from $0.19 on Grok Imagine to $21.32 on Seedance 2.5 at 1080p. One job each, priced from Sume's per-second rates.
Bulk queue idempotency: header or body key, and the header wins
A Sume bulk queue takes an Idempotency-Key as a header or as a body idempotency_key. If you send both, the header wins. Replays return 202 with the old queue.