Blog
Sume blog: agent-native video generation, Avatar 1.0 orchestration, and production APIs for builders and coding agents.
Music router 400 negative_prompt_unsupported: the fix and its cost
A non-empty negative_prompt on Sume's music router returns 400 negative_prompt_unsupported. Move the exclusions into prompt; a rejected call is not billed.
Voiceover plus music bed: TTS and Lyria cost, duck_db in Timeline
A 700-character voiceover ($0.03325) and one Lyria bed ($0.125) cost $0.15825 before the render. Timeline duck_db (0 to 20) lowers the bed under speech.
Lyria 3.5 has no duration field: how to ask for a 45-second track
Sume's music router rejects duration and duration_seconds. Write the length into the prompt, with section timestamps, and pay a flat $0.125 per generation.
Gate announcements: 8 messages in four languages, 32 TTS jobs
8 gate messages of 220 characters in 4 languages is 32 Sume TTS jobs at $0.01045 each, $0.3344 total. Set language and a matching voice per job.
Korean to Japanese voiceover: translate, then TTS with language ja
Sume TTS does not translate: send Japanese text with language ja. A 400-character Japanese line costs $0.01900. Transcribe Korean with STT first.
One slogan in English, Korean and Japanese: three TTS jobs, cost
Voicing one 120-character slogan in three languages is three Sume TTS jobs at $0.00570 each, $0.01710 total. Set language per job: en, ko, ja.
TTS sentence slices: segmentation, wav output and boundary_lead_ms 70
Sume TTS can return gapless sentence segments. Needs timestamps.words and wav or raw for per-sentence audio_url. A 900-character script costs $0.04275.
TTS word timestamps for captions: timestamps.words and what you get
Set timestamps.words true on Sume TTS and the result carries words[] with start and end seconds. A 600-character caption script costs $0.02850.
Audio job failed: read /events and /status before you retry
A failed TTS, STT or music job is terminal. Read GET /v1/jobs/{id}/events and status, fix the cause, then resubmit under a new key. A new TTS job bills again.
Silent captioned Reel: drop the audio first, burn with words
To burn captions on a Reel with no sound, run STT first, trim with audio drop, then send your own words: $0.02 + $0.01 + $0.20 = $0.23.
Re-base caption words after a trim: subtract actual_start_seconds
Caption words must use the trimmed file's clock. A word at 12.3-12.7 s becomes 1.9-2.3 s when the trim really started at 10.4, not 0.3-0.7 s.
Soundtrack for a 3-minute Short: one prompted track or a looped bed
Music Router is $0.125 per generation whatever the length, so a 3-minute Short pays the same for one prompted track or a 30 s bed looped by Timeline soundtrack.
Digital signage 1:4 portrait totem: cost of a 6-creative loop on Sume
A tall 1:4 totem screen takes Nano Banana 2.1's 1:4 ratio directly. Six creatives cost $0.45 at 0.5K, $0.60 at 1K, $0.90 at 2K, $1.20 at 4K on Sume.
Square product photo to a 4:1 banner: reference plus aspect_ratio
Turn a 1:1 photo into a 4:1 banner with Nano Banana 2.1 on Sume: input_references plus an explicit aspect_ratio of 4:1, not auto. $0.10 at 1K. Body and gotchas.
Sume video price = provider list x 1.25: Omni, H3, H3 Max, Recast
Provider list per second times 1.25 gives the Sume price: Omni 720p $0.10 becomes $0.125, H3 Max 768p $0.08 becomes $0.10. All 13 rows checked against the tool.
Hook, body and CTA from three video models in one 30-second Short
Seedance 2.5 for a 4 s hook, Wan 3.0 for a 20 s body and Omni Flash for a 6 s call to action cost $9.31 in generation, $9.41 with the Timeline join.
Read your Sume plan from ratelimit-limit on a GET: 4,800 to 48,000
A GET /v1/balance returns ratelimit-limit. 4800, 12000, 24000 or 48000 reads a minute maps to Free, Pro, Startup or Scale, and a full Wan 720p queue reserve.
Avatar handle with a hyphen is rejected: the valid pattern
Sume avatar handles allow letters, digits, periods and underscores, 2 to 30 characters. A hyphen fails the pattern. Check yours in Python first.
A/B test 6 hooks, 2 openings each: 12 vertical 5-second clips priced
Twelve 5-second vertical test clips cost $7.50 on Wan 3.0 at 720p and $34.668 on Seedance 2.5. Per-model totals, plus two 1080p winners.
Omni reference tokens follow list order: swap list, swap meaning
<IMAGE_REF_0> is the first URL in reference_image_urls, so reordering the list rewires the prompt. A check function and an 8 s body, $1.00 at 720p on Sume.
Retro TV bumper in 4:3: 11 s of Seedance 2.5 costs less at 480p
An 11-second 4:3 Seedance 2.5 clip costs $2.90 at 480p and $6.40 at 720p on Sume. At 480p, 4:3 is cheaper than 16:9.
Find callers still sending nano-banana-2 or gemini-3.1-flash-image
Sume runs the retired nano-banana-2 id as Nano Banana 2.1 and echoes your id back. A repo scan finds stale ids and gemini-3.1-flash-image before you migrate.
Daily news brief voiceover: 1,400 characters a day for 30 days
A 1,400-character daily brief costs $0.0665 on Sume TTS, so 30 days is $1.995. Table for 1,000, 1,400 and 2,000 characters, plus a retry-safe daily job.
Twelve 45-second Shorts a month: what each Sume video model costs
Twelve vertical 45 s Shorts a month cost $41.70 on MiniMax H3 at 768p and $68.70 on Wan 3.0 at 720p, up to $313.21 on Seedance 2.5 at 720p, Timeline included.
A failed Sume video poll has error as a string, not an error object
On /v1/videos, HTTP errors use {error:{code,message}} but a failed poll carries error as a string. Read both without a TypeError in Python and TypeScript.
Voice greeting on a still photo: TTS first, then H3 Max lip sync
To make a still photo speak, make 5 to 14.8 seconds of TTS audio, then call MiniMax H3 Max lip sync. TTS costs $0.0475 per 1,000 characters.
Cost per approved banner: 0.5K drafts, 2K final, break-even at 1 in 2
With Nano Banana 2.1 on Sume, drafting at 0.5K ($0.075) then rendering one 2K final ($0.15) beats rendering all at 2K once fewer than 1 in 2 drafts is approved.
Voice 50 listings at 300 characters each: Sume TTS cost
50 listing scripts of 300 characters each cost $0.7125 on Sume TTS ($0.01425 each). Key each by listing id so a rerun skips finished jobs.
Does a bigger top-up raise Sume video concurrency? No, the plan does
A top-up adds balance, not processing slots: Free runs 1 job and accepts 6, Pro runs 4 and accepts 24. What each plan holds and the reserve to fill it.
sume/auto and aspect_ratio 8:1: name Nano Banana 2.1 instead
sume/auto does not tell you which family ran, and the repo gives it the GPT Image ratio list, which has no 4:1 or 8:1. Name google/nano-banana-2.1 for strips.
Divide by 1.25: find the provider list price behind a Sume price
Sume bills provider list x 1.25, so a $1.50 Omni Flash 1080p clip of 8 seconds has a list of $1.20. Worked table for Wan 3.0, Omni, H3 and Nano Banana.
300x1200 half-page ad image: it is exactly 1:4, so request 1:4
A 300x1200 slot is 1:4. Nano Banana 2.1 lists 1:4 on Sume, from $0.075 at 0.5K to $0.20 at 4K per image. Request body, crop math for 160x600, and gotchas.
Animatic on Wan 3.0: six 5 s shots, approve, redo, upgrade
Six 5-second Wan 3.0 shots at 480p cost $1.875 in total, the same as one 30-second job. Split by shot so each frame can be approved alone.
Transcribe the first 60 seconds of 500 support recordings
500 screen-recorded support sessions: detach seconds 0 to 60 ($5.00), then STT 60 s each ($5.00). Total $10.00 at $0.01 per job and per minute.
4:1 banner into a 1:4 skyscraper: regenerate instead of cropping
A crop from 4:1 to 1:4 keeps 6.25% of the width. Send the banner as an input_reference with aspect_ratio 1:4 on Nano Banana 2.1 instead: $0.10 at 1K on Sume.
Face swap API has no prompt or aspect_ratio: the three-field body
Sume's face swap beta takes three fields: avatar_handle, video_url and quality. No prompt or aspect_ratio exists. It reserves 15 seconds at the tier rate.
Cost per kept second: Omni 720p vs H3 Max 768p at 1 take in 3
Omni Flash 720p is $0.125 a second and H3 Max 768p is $0.10. Keep one take in three and a kept second costs $0.375 or $0.30 on Sume. Break-even math inside.
Printable 2x8 inch bookmark is 1:4: cost of 6 to 24 designs
A 2x8 in bookmark is 1:4, or 600x2400 px at 300 dpi. Generate it directly with Nano Banana 2.1 on Sume: $0.15 at 2K, $0.20 at 4K per design. Costs for 1, 6, 24.
Python clip planner for a Short: split any length into model jobs
A short Python function splits a Short's length into equal whole-second jobs within each Sume video model's min and max, then prices them with Timeline.
Stream a Sume video download in Python: .part file, then rename
Download /v1/videos/{id}/content with requests stream=True, write a .part file and rename on success. A lost 30 s clip costs $1.875 to $17.334 to re-buy.
Where Sume documents plans, wallet and rates: which page answers what
Eight billing questions mapped to the docs page that answers each: the wallet, top-ups, list x 1.25, plan concurrency, 402 vs 429 and refunds, read 2026-10-09.
Poster motion: Seedance 2.5 at 3:4 costs 0.6% more than 9:16 at 720p
At 14 seconds, a 3:4 Seedance 2.5 clip is $0.05 dearer than 9:16 at 720p. Full 480p/720p/1080p table from the Sume price tool.
1080p per-second vs per-clip at 7, 13 and 21 seconds: who accepts each
At 1080p a 7-second clip fits five Sume models, a 13-second clip four, a 21-second clip only Wan 3.0 and Seedance 2.5. Per-second and per-clip prices for each.
Seven pre-flight checks before an Omni, H3 or Recast job
Duration, resolution, ratio, edit conflicts, reference counts, refused fields and URLs: seven per-model checks that catch a refused Sume video request.
Cancel 16 of 20 queued Seedance 2.5 clips: what you keep
Twenty 10-second Seedance 2.5 clips reserve $115.56. Cancel the 16 still queued and $92.448 is released; the 4 running clips complete and cost $23.112.
curl -w http_code: branch a Sume video submit in bash on 202, 402, 429
A bash submit that saves the body, reads the HTTP code with curl -w and branches on 202, 402, 429 and 5xx. Three Wan 3.0 payloads cost $1.875, $3.75 and $7.50.
Vertical recipe short on Wan 3.0: 16 s with a plated-dish last frame
A 16-second 9:16 Wan 3.0 recipe clip costs $1.00 at 480p, $2.00 at 720p and $4.00 at 1080p on Sume, and the vertical ratio does not change the price.
32:9 super ultrawide wallpaper from a 4:1 image: crop 11% of width
32:9 is 3.56:1, between GPT Image 2.5's 3:1 cap and Nano Banana 2.1's 4:1. Generate 4:1 on Sume at $0.20 for 4K and keep 88.9% of the width. Request and math.