Blog
Sume blog: agent-native video generation, Avatar 1.0 orchestration, and production APIs for builders and coding agents.
Is Nano Banana 2.1 free in the Gemini API? The price table says no
Google's pricing page shows no free-tier price for Nano Banana 2.1 image output: only paid rates, $0.0336 per 1K image. Sume bills per image from a wallet.
LinkedIn landscape video ad maxes at 1920x1080; Sume allows up to 2160
LinkedIn's landscape video range is 640x360 to 1920x1080, but Sume timeline accepts sizes to 2160. Pick 1920x1080 on purpose and verify with video inspect.
Nano Banana 2.1 for 37 images: Google's price vs Sume at 1K, 2K, 4K
A 37-image set on Nano Banana 2.1 costs $1.24 to $4.18 at Google's list and $3.70 to $7.40 on Sume. Per-tier arithmetic and what the gap buys you.
Fix one wrong sentence in a voice-over: 130 characters, not 3,000
Redoing a 130-character sentence costs $0.006175 in Sume TTS plus a $0.01 audio join. Re-synthesizing the 3,000-character script costs $0.1425.
MCP tool error or JSON-RPC error: where Sume's failures land
Sume returns tool failures as results with isError true and keeps JSON-RPC errors for protocol faults, per the 2025-11-25 MCP spec. Here is the split.
Claude Code maxResultSizeChars 500,000 vs Sume's 256 KiB output cap
Raising Claude Code's per-tool result limit to 500,000 characters does not change what Sume returns: the server caps one MCP result at 262,144 bytes.
A full queue on each Sume plan: reserved dollars for 10-second jobs
If every accepted job holds its estimate, a full Scale queue of 120 ten-second Seedance 2.5 jobs at 1080p reserves $1,705.86. Free, Pro and Startup too.
Can polling hit the Sume read limit? 24 Pro jobs at 2 s use 6%
Polling every accepted job every 2 seconds uses 3.75% to 7.5% of a Sume plan's read budget. The arithmetic for Free, Pro, Startup and Scale, with the caveats.
Sume API key with actions:read only: list schedules, not start runs
A Sume key with actions:read can list and read schedules and runs. Starting or canceling a run needs actions:write; keys made before the trigger lack both.
Sume TTS: 20,000 characters or 1,200 seconds, which fails first?
At 15 characters a second, 20,000 characters is 1,333 seconds, so the 1,200-second audio cap trips first. Split near 15,000 characters per job.
100-clip batch on a Pro workspace: 24 accepted jobs, 18-job waves
On Pro a workspace holds 4 processing plus 20 queued jobs, so a 100-clip batch needs waves. Wave arithmetic for 100 six-second Wan clips costing $75.00.
X video ad thumbnail: PNG or JPEG under 5 MB, pulled from your own MP4
X accepts a PNG or JPEG thumbnail up to 5 MB that matches the video. Sume video frames returns a still at a time you pick, at source size.
Photo avatar of a real person: when YouTube's AI label applies
A photo avatar scripted to say new words falls under YouTube's 'real person says something they did not' test. How Sume's three avatar inputs map to it.
X video ad: 2:20 at 6,000 kbps is 105 MB, so the 1 GB cap never binds
X allows 1 GB and 2 minutes 20 seconds. At its recommended bitrate the length limit hits first. The math, plus the 9:55 case for approved advertisers.
Format run expires_at: a 90 minute ceiling for your Python poll loop
A Sume Format run receipt has an expires_at deadline: 90 minutes from creation, sooner if the run goes quiet. Use it as your loop ceiling, with Python backoff.
Check a GPT Image 2.5 size before you call it: a Python validator
Sume rejects GPT Image 2.5 sizes that break four rules: multiples of 16, edge 3840, ratio 3:1, 655,360 to 8,294,400 pixels. A Python check, run on 8 sizes.
MCP spec: tool annotations are untrusted. What that means for Sume
The MCP spec says clients must treat tool annotations as untrusted unless they come from a trusted server. Treat Sume as trusted only for your own workspace.
One ad for X and Snapchat: no shared aspect ratio, render two masters
X lists 16:9 or 1:1 for video ads and Snapchat lists 9:16, so they share no ratio. A 140-second length fits both. Two Sume renders cost $0.60.
Should an agent tool take a workspace_id argument for Sume calls?
No. Sume's safe-automation guidance says the API key or app session selects the workspace and tools must not take a workspace id from the user.
Fabric accepts image_url or avatar_handle, never both: which to send
On Sume's VEED Fabric 1.0 route, send one visual source: the image_url of an inspected still, or an avatar_handle the user named. Both together is not allowed.
Sume TTS is a job, not a stream: cost and fit for a voice agent reply
A 45-second spoken reply is about 675 characters: $0.0320625 on Sume TTS, $0.010125 on MAI-Voice-2.1-Flash. Sume returns a finished file, not a live stream.
Real footage trimmed and captioned on Sume: TikTok AI disclaimer?
TikTok requires an AI disclaimer for fully AI-generated or significantly AI-modified media. Which Sume jobs use no generative model, and which ones do not.
MCP spec: token in the header on every request, never the URL
The 2025-11-25 MCP spec forbids access tokens in the URL query string and wants a Bearer header on every request. Keep the Sume key in headers, not the URL.
Does Copilot's MCP policy apply to Pro+ or Max? Sume hosted MCP
GitHub's MCP servers in Copilot policy covers Business and Enterprise, off by default. Free, Pro, Pro+ and Max users can add Sume's hosted MCP server directly.
Keeper rate math: 24 eight-second Seedance 2.5 clips at 25% to 80%
At a 40% keeper rate, 24 Seedance 2.5 clips at 720p cost $277.34, not $110.94. Direct takes vs 480p drafts plus one 720p final, and the break-even.
LinkedIn allows 25 video uploads in 24 hours: plan a 60-variant test
LinkedIn Campaign Manager caps video uploads at 25 per 24 hours, desktop only. A 60-clip test needs three days. Cut and price the variants on Sume first.
Sume MCP read_busy 503: four concurrent reads per owner
read_busy is a retryable 503 from Sume's MCP host. Reads inside tools, including jobs_wait polls, are capped at 4 at once per owner, 512 across the host.
Sume STT: omit duration_seconds and it reserves 1 minute, not 9
A 9-minute file is $0.09 of Sume STT, but without duration_seconds the reserve is 1 minute, $0.01. Send 540. The field takes 1 to 600 seconds.
Dub a 3-minute video into Spanish: a 47.8-cent Sume ledger
Detach, transcribe, translate, speak and re-render a 3-minute video: $0.47825 in Sume jobs at 900 characters a minute. Translation is billed elsewhere.
MAI-Voice-2.1 lists 23 languages; how Sume TTS sets its language
Microsoft lists 23 languages and 26 locales for MAI-Voice-2.1. Sume TTS takes one language field, defaults to English and asks you to confirm a voice mismatch.
Storyboard: 12 rough and 3 hero frames on three Sume image models
A 15-frame storyboard with 12 rough frames and 3 hero 4K frames costs $0.9645 on GPT Image 2.5, $1.50 on Nano Banana 2.1 and $3.375 on Nano Banana Pro.
Nano Banana 2.1 vs GPT Image 2.5: 4.04x at 1K, 0.9x at 4K
On Sume, Nano Banana 2.1 costs $0.10 at 1K and $0.20 at 4K; GPT Image 2.5 costs $0.02475 low and $0.2225 high. The ratio flips between 2K and 4K.
Which Sume API routes work without a key? Six public routes
Six Sume routes need no API key, including GET /v1/catalog and GET /v1/health. What they return, how they are limited, and a Python preflight for CI.
Seedream 5 Lite 4:5 portrait: render 1920x2400, resize to 1080x1350
Seedream 5 Lite lists 4:5 on Sume at $0.04375 per image. A 1920x2400 render shrinks to Instagram's 1080x1350 at an exact 0.5625 scale. 100 images cost $4.38.
Pinterest max-width video from a 9:16 master: the 1:1 height limit
Pinterest max-width video cannot be taller than 1:1. A 1080x1920 master becomes 1080x1080 with cover (56.25% of rows) or blur (608 px wide picture).
Sora API removed, no replacement listed: pick a Sume video id
OpenAI removed the Videos API on 2026-09-24 and lists no replacement. Pick a Sume video model by duration cap and the price of an 8-second 720p clip.
YouTube AI label: three auto-detection signals and Sume outputs
YouTube may label video automatically for its GenAI tools, C2PA metadata or internal detection. Sume docs do not mention C2PA, so here is what to check.
Gemini CLI authProviderType: connect Sume with dynamic_discovery
Gemini CLI defaults authProviderType to dynamic_discovery. For Sume that means OAuth discovery, not Google credentials. What each of the three values does.
Devin Desktop ${file:} interpolation for the Sume API key
Devin Desktop expands ${file:/path} and ${env:NAME} in mcp_config.json. Put the Sume API key in a file outside the config; the 100-tool cap still applies.
Trim a Reel clip's dead start and end from first and last word
Take the first word's start and the last word's end from STT, then add 0.15 s lead and 0.30 s tail: a 29.9 s clip becomes 1.69-27.92 s for $0.03.
Validate a cut list against video-trim's rules before you submit
Video trim has no unbilled check endpoint, so mirror its rules locally: 1800 s source, 0.2 s to 900 s output, one of end or duration, exact with output.
Stable idempotency keys for pause-cut trims: hash source, start, end
Derive each trim's Idempotency-Key from a hash of source, start, end and precision. A retried batch then returns the same jobs and bills 8 trims once.
Pause-cut with Timeline alone: 14 slots from one source, no trims
One Timeline render can cut 14 pauses from a single clip with source_in slots and audio.parts: $0.11 total, versus $0.39 with 14 trim jobs first.
Re-cut a Reel at a new pause threshold: the source never changes
First Draft is reversible; your API cut can be too. Sume media jobs never edit the source, so a new threshold is a new cut list and a $0.10 render.
Keep 0.24 s of breath: shorten pauses instead of deleting them
Pad each kept range by 0.12 s per side and 9 pauses totalling 6.35 s lose 4.19 s, not 6.35 s. The speech keeps a 0.24 s breath at every join.
Check pause-cut seams with video frames: 24 stills cover 12 seams
Video frames returns up to 24 stills per call. Two stills per seam covers 12 seams per call, so a 14-cut Reel needs 2 calls to see every join.
Timeline body from a cut list in TypeScript: starts and coverage
Turn a cut list of source ranges into Timeline slots: starts are running sums rounded to 3 decimals, and 3 ranges of 3.55, 4.68 and 3.36 s make 11.59 s.
Cut away to B-roll without breaking the voice: one detached spine
Detach the A-roll audio once ($0.01) and render three slots in one Timeline job ($0.10): 32 s of picture, B-roll over the middle, the voice unbroken.