Blog
Sume blog: agent-native video generation, Avatar 1.0 orchestration, and production APIs for builders and coding agents.
Does MiniMax H3 have an API? Three endpoints, or one route on Sume
Yes. MiniMax lists three Open Platform endpoints for H3: create, h3-context-ir and regeneration. Sume wraps H3 in one async video route, minimax-h3.
SynthID on every Gemini TTS clip: keep the original file for ad review
Google marks every Gemini 3.8 TTS clip with SynthID and C2PA credentials. What that means for ad voiceovers, and what Sume docs do and do not say.
Crop 21:9 to a 1:1 square: width 0.4286, centered x 0.2857
A 2520x1080 ultrawide clip becomes 1080x1080 with a crop of width 0.4286. TikTok's in-feed page lists 1:1 at a 640x640 minimum. Request and numbers inside.
/v1/videos size returns 400: send resolution and aspect_ratio
Every v1 Sume video model reports supported_sizes null, so a size field gets 400 unsupported_parameter. Send resolution and aspect_ratio; where to read values.
Facebook feed ad text 50-150 characters, headline 27: what Sume burns
Meta's Facebook feed video page lists 50-150 characters of primary text and a 27-character headline. Sume's burned captions never touch those ad fields.
SQLite: add a sume_job_id column to rows that used sora-2 snapshots
OpenAI removed the Videos API and every sora-2 snapshot on Sept 24, 2026. A migration that tags old rows, then re-renders each once through Sume.
Runway Gen-4 Turbo at 5 credits a second: Sume's closest rates
Runway lists gen4_turbo at 5 credits a second, $0.05 at $0.01 a credit. Gen-4 Turbo is not on Sume; its cheapest video seconds start at $0.0375.
Wan 2.7 vs Wan 3.0: clip length and resolution limits by model id
Wan 2.7 tops out at 15 s and starts at 720P; Wan 3.0 reaches 30 s and adds 480P. Alibaba's ids side by side, and which one Sume lists.
Wan 3.0 API rate limit: 300 RPM on Model Studio vs Sume jobs
Alibaba lists 300 requests per minute for Wan 3.0 on Model Studio. Here is what that means for a batch, and how Sume submits Wan 3.0 as async jobs.
GPT Image 2.5 banner: 3840x1280 passes the 3:1 rule
3840x1280 is exactly 3:1, both edges are multiples of 16 and it is 4,915,200 pixels, so Sume accepts it on GPT Image 2.5. The four checks shown.
Crop a 720p 16:9 clip to 1:1: 720x720 clears TikTok's 640x640
A 1280x720 clip cropped with width 0.5625 gives 720x720, which clears TikTok's 640x640 square minimum with 80 px to spare. Request and numbers inside.
Nano Banana 2.1 on Sume in Python: handle the 200 and 202 replies
Call google/nano-banana-2.1 on Sume's /v1/images in Python and handle both the 200 image reply and the 202 job envelope, with polling to the result URL.
How many languages do AI avatar tools list? Six pricing pages
Creatify lists 75+, Captions 100+, Colossyan 120+, Synthesia 140+, HeyGen 175+ dialects. Sume's avatar docs give no language count. What to test instead.
YouTube lists 1440p at 16 Mbps; Sume Timeline cannot output 2560x1440
YouTube recommends 16 Mbps for 1440p SDR. Sume Timeline caps each edge at 2160, so 2560x1440 is not possible; the closest 16:9 render is 2160x1214.
curl and jq: count Sume jobs by status in one command
GET /v1/jobs with limit=100 piped into jq group_by prints how many jobs are queued, processing, failed and so on, and warns when next_cursor says more.
Instagram feed video ad 1% ratio tolerance in pixels at 1080 wide
Meta lists a 1% aspect ratio tolerance for Instagram feed video ads. Read as relative, at 1080 wide that is about 1902-1938 px tall; Sume's 1080x1920 is exact.
Rate avatar clips like Griffin-Lite: naturalness, trust, flow
Tavus reports Griffin-Lite averages of 5.4 naturalness, 5.6 trust and 4.9 conversation flow. Score your own avatar clips on the first two; a test costs $14.69.
"Keep the product identical": the Sume docs background-swap prompt
Sume docs use one prompt for a product edit: keep the product identical, swap the background. The request, the fields it needs, and what to check afterward.
Put your 16 strongest hooks first: Sume bulk queues start in order
A bulk queue returns 202 with the first `concurrency` items already running, up to 16. Order your 100 weekly ads so the best hooks are the first ones you see.
Bulk queue item error: format_run_canceled vs format_run_failed
In a Sume bulk queue, a failed child reads format_run_failed and a canceled child reads format_run_canceled. The real reason is in the child run receipt.
MCP wants a human in the loop: Sume gates for unattended agents
MCP says a human SHOULD be able to deny tool calls. For unattended agents, Sume adds scopes, idempotency keys and a required Agent Completions cap.
OpenRouter video client on Sume: six documented differences
Sume says /v1/videos matches the OpenRouter Video Generation API except six items: base path, auth, model ids, auto routing, size and provider.options.
Sume Avatar max costs 2.24x plus and 3x standard: when to pay it
Sume Avatar 1.0 is $0.184 a second on standard, $0.245 plus, $0.55 max: ratios 1 : 1.33 : 2.99. Iterate on standard, ship plus, pay max for hero ads.
Go httptest: a four-case table test for a Sume webhook handler
Sign a fixture body with sume-v1 HMAC, then assert 204 for a valid call and 401 for a stale timestamp, an empty server secret and a wrong secret.
TikTok horizontal ad minimum is 960x540: which 16:9 sizes clear it
TikTok's in-feed page lists 960 x 540 as the minimum for 16:9. Sume Timeline accepts 960x540, 1280x720 and 1920x1080, all at $0.10 per output minute.
Claude Code allowedMcpServers: the serverUrl rule for mcp.sume.com
Allow only Sume's hosted MCP endpoint in Claude Code managed settings with an allowedMcpServers serverUrl entry, and what the rule does not control.
Image API n: 10 per call in the schema, lower for each model
The Sume Image API schema allows n from 1 to 10, but each model lists its own n range. How to read the ceiling and what you pay for n images.
Which Sume video models take an audio_url reference, which do not
Sume video rows list supported_input_references per model. The docs show seedance-2 with audio_url, H3 Max with audio, and Gemini Omni Flash 1.1 without it.
Python asyncio.Semaphore sized to a Sume plan's job capacity
Free holds 6 paid jobs at once, Pro 24, Startup 48, Scale 120. A Semaphore of that size keeps a 60-job batch from hitting 429 queue_full.
Node: flip a video backend to Sume a day before the Veo 3.1 shutdown
Google ends three Veo 3.1 preview ids on Oct 22 and its page gives a date, not an hour. A 19-line Node switch moves traffic to Sume early, with an override.
Image API 200 or 202: branch on the status code, not the body
POST /v1/images returns images with 200 or a job envelope with 202. A small Python handler that branches on the code and prints the URLs to poll.
Image 1.0 is retiring and its URLs point at Auto: what to change
Sume says Image 1.0 retires soon and its public URLs are compatibility aliases for the Auto pipe. The three edits for a client that still calls /v1/image-1.0.
Griffin-Lite study: 81% confident about the AI, 79% about people
Tavus reports participants were 81% confident about the AI and 79% about real people. Confidence did not track the truth, so label AI avatar clips yourself.
TypeScript SDK: createVideoGeneration, then poll for 'cancelled'
The @sume-com/sdk video functions return {data, error, response} and never throw. An 18-line loop passes an Idempotency-Key and handles the British spelling.
Can Gemini Omni edit change a video's aspect ratio? No
On Sume, Omni's edit mode takes a video_url and a prompt, defaults to 720p and accepts no aspect_ratio or duration. A 16:9 source stays 16:9, so plan the crop.
Gemini 3.5 Transcribe lists 85+ languages; Sume STT takes a hint
Google's changelog lists Gemini 3.5 Transcribe with 85+ languages and word timestamps. Sume STT returns words always and takes an optional language hint.
Gemini 3.8 Flash TTS takes 8,192 input tokens: splitting a long script
Gemini 3.8 Flash TTS lists 8,192 input tokens and 16,384 output tokens. Sume TTS takes 20,000 characters per job. Here is how to split a long script.
Business Profile video takes 24-48 hours to show: plan the clip early
Google says Business Profile photos and videos can take 24 to 48 hours to appear; clips run up to 30 seconds and 75 MB at 720p or higher. A restaurant plan.
Auditing agent tool calls: MCP log advice, Sume script_run journal
The MCP spec tells clients to log tool usage for audit. For Sume's script_run, the response includes a calls[] journal and child jobs[] to use with jobs_wait.
MCP outputSchema vs Sume output_schema: who sets the contract
In MCP the server declares a tool's outputSchema. In a Sume Agent Completion you send output_schema per run, and the result can still come back degraded.
MAI-Voice-2.1 neutral-only voices: Hungarian, Romanian, en-IN
Several MAI-Voice-2.1 voices list only a neutral style: five Hungarian, four Romanian and both en-IN voices. Check the style list before you cast an ad.
ElevenLabs v4 voice cost vs a 30-second avatar clip: under 1%
ElevenLabs v4 lists $0.08 per 1,000 characters. A 30-second script is about 450 characters, or 3.6 cents, against $7.35 for a plus-tier Sume avatar clip.
Akool concurrent video generations: 2, 6, 8, 20, vs Sume's queue
Akool allows 2, 6, 8 and 20 concurrent video generations by plan. Sume accepts extra avatar jobs as queued and returns 429 queue_full only when both are full.
claude mcp add --header Bearer: a Sume API key for CI runs
For unattended Claude Code runs, pass a Sume API key as a Bearer header. That session sees every tool, so cap spend and send idempotency keys.
Akool streaming avatar sessions: 15 to 60 minutes vs Sume clips
Akool's streaming avatar session runs 15 minutes on Starter and Pro, 30 on Pro Max, 60 on Business. Sume renders clips of up to 60 s. Which one you need.
Crop 3:2 to 4:5 for the feeds: width 0.5333, x 0.2333 centered
A 1620x1080 (3:2) clip becomes 864x1080 (4:5) with a crop of width 0.5333. Meta lists 4:5 for the Facebook feed; here is the request and the catch.
Codex 0.160.1 remote stdio MCP fix: does it affect Sume?
Codex 0.160.1 preserves SYSTEMROOT, TEMP and TMP for remote stdio MCP launches on Windows hosts. Sume's hosted MCP is HTTP, so no process is launched.
Nova Reel 2-minute videos in 6 s steps vs Sume's 30 s video jobs
Nova Reel makes up to two minutes through async invoke in 6-second increments. Sume's catalog tops out at 30 seconds per job, so a long video needs several.