Blog
Sume blog: agent-native video generation, Avatar 1.0 orchestration, and production APIs for builders and coding agents.
Should my backend call Sume over hosted MCP or the REST API?
REST from a backend, hosted MCP from an agent client. Where they differ: auth, wait limits, REST-only Image 1.0 and Video 1.0, and write budgets.
Speech to text API in Node.js: transcribe audio with fetch
Transcribe audio in Node.js with built-in fetch: submit to Sume STT, poll the job and print sentence segments with timestamps. 30 lines, no dependencies.
Speech to text API in Ruby: transcribe audio with Net::HTTP
Transcribe audio in Ruby with only the standard library: submit to Sume STT, poll the job, print text and word times. A 30-line script at one cent a minute.
Speech to text API in Swift: transcribe audio with URLSession
Transcribe audio in Swift with async URLSession: submit to Sume STT, poll the job, print text. No packages, 28 lines, one cent per audio minute of audio.
A Sume job looks stuck: wait, cancel or poll the events?
Read status, then events. Queued and processing mean wait, cancel only works before generation starts, and a client timeout never cancels the job.
Swap the presenter per market: one ad, two Recast jobs on Sume
Recast keeps the motion, camera, cuts and sound of one source ad and replaces the people. How to make one version per market, the limits and the price.
Text-to-video or image-to-video: when the prompt alone is enough
Use text-to-video when the look is open, and image-to-video when a frame, a face or a product must match. The Sume models that take each, and a decision rule.
MiniMax H3 Max on Sume: video, lip-sync and recast, which id to call
Three Sume ids carry the MiniMax H3 Max name: minimax-h3-max for video, a lip-sync route and h3-max-recast. What each takes, its window and its rate.
How do I put three products in one AI video from reference photos?
Send three product photos as input_references to Gemini Omni Flash 1.1: one 8-second 720p clip is $1.00 on Sume, against $1.89 for three single clips.
Transcribe an interview with timestamps via API (no speaker labels)
Send interview audio to Sume STT for text, word start and end times and sentence segments. It returns no speaker names. 25 minutes costs 28 cents.
Transcribe audio with curl and jq: a Sume STT shell script
A 16-line bash script that submits audio to Sume STT, polls the job with curl, and prints every word with start and end times through jq. One cent per minute.
Translate a coffee bag label to Japanese with Ideogram 4.5
Translate a packaging label into Japanese with ideogram/ideogram-v4.5: give the exact string, one language per call, check characters. $0.075 per bag at medium.
Turn summer footage into autumn with an edit prompt (Omni Flash)
Change the season of a clip you already shot with a video_url edit on Gemini Omni Flash 1.1 at Sume: prompt wording, what to verify, and a 6-second price table.
unsupported_capability names sume/auto: fix a Sume ad clip request
A 400 unsupported_capability on sume/auto hides the resolved model but lists accepted values in supported. How to fix duration, resolution or audio.
One lip-sync model per video: Fabric runs at 25 fps, measure H3 Max
Mixing Fabric and MiniMax H3 Max lip-sync clips in one Sume video risks a frame-rate mismatch. Pick one per run and check fps with ffprobe on the first clip.
verifyWebhook in a fetch handler: four rules, 204 for unknown events
Use @sume-com/sdk verifyWebhook on the raw body, await it, treat false as 401 and answer unknown events with 204. A runnable handler for Workers, Deno and Node.
Video edit prompt: say what stays, then what changes (Omni Flash)
A prompt pattern for Gemini Omni Flash 1.1 video edit on Sume: one change per pass, an explicit keep clause, and what the video_url route fixes for you.
Virtual staging an empty room photo with AI: GPT Image 2.5 via API
Stage an empty room photo with openai/gpt-image-2.5: the room plus up to 15 furniture references, an optional mask, and about $0.08 per staged image at high.
Wan 3.0 API request cheat sheet: three modes, 2 to 30 seconds
Wan 3.0 on Sume: the request body for text, first/last frame and reference modes, the 480p/720p/1080p rates and the 2 to 30 second window, on one page.
What is a partial transcript in streaming speech to text?
A partial is a provisional transcript a streaming model revises as audio arrives. Why subtitles for a finished clip only need final text and word times.
What to show a viewer while an avatar video job is queued
Avatar jobs on Sume are async: queued, processing, then completed, failed or canceled. A status-to-UI map for waiting screens, with polling rules.
Which Sume catalog Formats make ad creative: the slugs by ad type
Sume ships 27 ready-made Formats at the sume handle. Which slugs fit which ad type, and how to read one with GET /v1/formats/sume/{slug} before paying.
Which Sume job and run endings send a webhook, and which stay silent?
Jobs send job.completed, job.failed and job.canceled. Canceled or skipped runs send nothing. A matrix of terminal states and what your receiver can expect.
Which Sume image models accept image_size for custom pixel dimensions?
Eleven of Sume's 19 image models accept image_size with a width and height: ChatGPT Image 2.5 and 2, Seedream, FLUX.2, Qwen and Recraft. Eight do not.
Which Sume image models are text-to-image only, with no photo edits
Five Sume image models reject reference images: Higgsfield Soul, Imagen 4 Fast, Imagen 4 Ultra, Recraft V4 and Qwen Image Max. Prices, ratios and edit options.
Write a lip-sync script in 5 to 14.8 second beats (H3 Max voice-over)
MiniMax H3 Max lip-sync on Sume takes audio of 5 to 14.8 seconds. Write the script in beats that fit the window, then voice each beat; the price per beat.
YouTube bumper ad 5-6 seconds: generate at 6 or trim on Sume
YouTube's help page puts bumper ads at 5 to 6 seconds. Which Sume video models can generate a 6 second clip directly, and when a $0.02 trim is the better route.
Zoom in on a small product in frame: AI recompose or crop and upscale
Product too small in the photo? A Pillow crop plus Sume Image Upscale ($0.20) keeps real pixels; an Ideogram 4.5 recompose edit costs $0.075 and redraws them.
30, 15 and 6-second cutdowns from one master with video trim
Make three cutdowns from one master clip with video trim: $0.02 each, $0.06 in all. Use exact precision, conform to 1080x1920, and keep or drop the audio.
A 429 from the Sume API: read budget, write budget or queue_full?
Three different 429s share a status. Read error.code and error.details.scope to tell read, write and queue_full apart, then retry each the right way.
How do I animate my podcast cover art into a 5-second video intro?
Use the cover as a first frame for a 5-second clip ($0.63 at 720p), add a $0.125 sting and a $0.10 render: $0.855 once, reused every episode.
API key scopes for Sume: which key can call which endpoint family?
Sume API keys carry fixed scopes: formats:write, actions:read, agent_completions:write, account:read. Which scope each route needs, and why old keys get a 403.
Avatar product image adds about 5% per second on every Sume tier
Adding a product_image to a Sume avatar video raises the per-second rate by 1 cent on standard, 1.3 cents on plus, 3 cents on max. The table and a 30 s example.
Avatar video webhook mode: what arrives and what to poll anyway
Use mode webhook for a Sume avatar video and Sume posts one terminal event: completed, failed or canceled. Payload, signature headers and the polling backup.
Balance check before an ad variant burst: Sume API 402 guard
Read GET /v1/balance before sending a burst of video variants. Sume reserves 1.25 times list price on submit and returns 402 insufficient_credits below it.
Cap Sume spend from an agent loop: dry_run, max_spend_usd and run caps
Four optional guards cap what an automated Sume caller can spend: dry_run, max_spend_usd, generation_spend_cap_usd on Formats, and a balance check.
Caption a talking-head video: style, placement and phrasing
Caption a talking-head clip with the Sume API: pick a style, move the line off the face, set words per card. One request, $0.20 for up to 60 seconds.
Captions for a video with loud background music: script text or cues
Music can bury speech and trip speech-to-text. Sume documents three ways to supply your own wording: script_text, a words array or cues. Which to pick.
Captions out of sync with the audio: check STT word times and offsets
Captions running early or late usually trace to an unapplied offset. How Sume STT word times work, which offset to add, and a Python merge that applies it.
Change prices on a chalkboard menu photo with Ideogram 4.5
Update dish prices on a chalkboard menu photo with ideogram/ideogram-v4.5: list old and new text, one change set per call, check each digit. $0.075 at medium.
Cost per ad variant: build a ledger from usage.cost on Sume
Every completed /v1/videos poll carries usage.cost. Sum it by hook and ending to get the cost per ad variant before media spend. Node script and the caveats.
Create an AI avatar and its first talking video in one bash script
Two Sume jobs in order: create the avatar, wait, then render a talking video with its handle. A bash script with curl and jq, plus the cost of both steps.
Crop 16:9 to 9:16 with video filter: the fractions to send
For a 16:9 source, a centred 9:16 crop is x 0.3418, y 0, width 0.3164, height 1. One $0.02 video-filter job, checked free first. Examples for 1920x1080.
CTA end card for an AI video ad: use last_frame on Sume
End an ad clip on your CTA card by sending it as a last_frame image on /v1/videos. Which models accept it, which do not, and the Timeline alternative.
How do I automate a daily morning briefing audio with a TTS API?
A daily briefing of 2,200 characters is one TTS job: 11 cents a day, $3.30 for 30 days on Sume, plus a 1-cent concat if you add a music sting.
Day to night exterior photo: GPT Image 2.5 or Ideogram 4.5?
Turn a daytime building photo into twilight: Ideogram 4.5 keeps the source shape at $0.075; GPT Image 2.5 needs aspect_ratio auto, $0.0835 at high.
Do low, medium and high mean the same on every Sume image model?
No. On Sume only ChatGPT Image 2, 2.5 and Ideogram 4.5 accept low, medium and high, and the tiers change price in different ways. Defaults and the tier ladders.
Does PNG, JPEG or WebP change the price of an AI image on Sume?
No. On Sume's Image API, output_format picks the file type, not the price: per-image cards and GPT Image 2.5 token math ignore it. Which models list which.