ACE-Step on Sume: 240 s max, billed per second not per minute
ACE-Step is the Music Router id ace-step: 5–240 s via duration_seconds, billed per second of audio, so a 45 s clip costs less than a 60 s one.
Models
The models behind Sume's video, image, speech, and music endpoints: ids, clip lengths, resolutions, aspect ratios, audio, references, and migration.
Start with: What AI models does Sume support? Every model id, by API surface
CassetteAI Music on Sume: a 90 s track bills 2 minutes
CassetteAI Music runs on Sume's Music Router as cassetteai-music: 10–180 s via duration_seconds, billed per started minute, so 90 s costs two minutes.
ElevenLabs Music v2.5 on Sume: 300 s max, started-minute billing
ElevenLabs Music v2.5 is the Music Router id elevenlabs-music-2.5: 10–300 s via duration_seconds, billed per started minute, so 61 s costs two minutes.
Music Router duration_seconds: 4 models take it, 5 do not
Since Oct 14, 2026 four Sume Music Router models accept duration_seconds (5–300 s ranges); five still set length from the prompt and reject the field.
Lyria 2 on Sume: one 30 s instrumental clip, no length field
Lyria 2 on Sume's Music Router writes one 30-second instrumental clip per call and refuses duration_seconds; stitch clips or pick another engine.
MiniMax Music 2.6 on Sume: vocals from a 10-2,000 character prompt
MiniMax Music 2.6 on Sume's Music Router writes lyrics from your 10 to 2,000 character prompt, takes no duration_seconds and bills a flat price per track.
Stable Audio 2.5 on Sume: a 190 s track costs the same as a 5 s one
Stable Audio 2.5 on Sume's Music Router takes duration_seconds 5 to 190 and bills one flat price per track, so ask for the full length you need.
HelixWorld makes 121 frames at 24 FPS: Sume rows that run longer
HelixWorld's preview caps a run at 121 frames, about 5 seconds at 24 FPS. Sume's video rows with sound run 5 to 15 seconds; here is the arithmetic and the rows.
Is HelixWorld on Sume? No: a preview checkpoint, no hosted API
HelixWorld, Noiz AI's interactive video-and-stereo-audio model, ships as weights and a browser demo. Sume lists no id for it; these rows return sound.
Split an image into 2-4 editable layers by API: Qwen Image Layered
Send one picture to qwen-image-layered on Sume and get 2 to 4 transparent PNG layers, billed at $0.05 each. num_images is required; the call runs as a job.
Qwen Image 2.1 Pro API price: $0.04 in Singapore, $0.0353 elsewhere
Alibaba lists qwen-image-2.1-pro at $0.04 an image in Singapore and $0.035333 elsewhere. Sume has no 2.1 Pro row; its Qwen rows bill $0.025 and $0.094.
Eleven Multilingual v2 or Turbo v2.5 on Sume: 28 vs 31 languages
On Sume's TTS Router, Eleven Turbo v2.5 lists 31 languages, 20,000 characters and half the list price of Multilingual v2, which lists 28 languages.
HelixWorld license: Apache 2.0 code, weights and paper differ
HelixWorld's code is Apache 2.0, its weights have a separate license on Hugging Face, and the arXiv page lists CC BY-NC-SA 4.0. Read all three.
Is Qwen Image 2.1 Turbo on Sume? No: use these three Qwen rows
Sume's image catalog has no Qwen-Image-2.1 Turbo row. It lists Qwen Image, Qwen Image Max and Qwen Image Layered, and one call shows which.
Can Meta Muse Wonder generate images? It returns text; use /v1/images
Muse Wonder is listed on OpenRouter as a chat-completions model with text responses. It is not in Sume's image catalog; pick an image row instead.
Mistral Large 4 function calling: validate arguments against Sume
Mistral Large 4 (Public Preview) lists function calling and structured outputs. Validate its arguments with Sume's tools_schema before any paid call.
AI video API news, Oct 13 2026: Grok Imagine, Omni Flash, HelixWorld
What the vendor pages say today about Grok Imagine Video 1.5 and Lite, Gemini Omni Flash as Google's default, and HelixWorld, and which have a Sume id.
AI video model news, Oct 13, 2026: what Sume lists today
Sume lists Seedance 2.5, Kling 3, Wan 3.0, MiniMax H3, H3 Max and Omni. It lists no Kling 4.0, no Vidu Q4. Catalog snapshot, read 2026-10-13.
Amazon Nova 2.5 Sonic: four regions, seven languages, 256K context
AWS announced Nova 2.5 Sonic on October 5, 2026: four regions, seven languages and a 256K context window for live voice agents, not rendered voiceovers.
Decagon Chord's 45.7% human test: what the number measures
In Decagon's test, 45.7% of listeners thought the real human voice was the AI. That is a customer-call result, not proof for ad voiceover.
Is Decagon Chord a public API? What Decagon has published
Decagon introduced Chord on October 1, 2026 as the speech model inside Voice 3. Its post does not state a public API, so a voiceover job needs another model.
Every Sume image model compared: edit, text-only, n and formats
Sume's repo catalog lists 20 image rows: 14 take edit references, 5 are text-only, 1 makes layers. Grok is capped at n 1; Recraft is webp only.
H3 Max lip sync and audio over 14.8 seconds: split or Fabric
Audio over 14.8 seconds is refused on Sume's H3 Max lip sync because the provider silently clips the rest. Fabric 1.0 takes the long line.
H3 Max lip sync rejects audio under 5 seconds: what to do
Sume's MiniMax H3 Max lip sync takes 5 to 14.8 seconds of audio and returns invalid_request outside that window. Fabric is the fallback for shorter lines.
Porting a Sume agent loop to Haiku 5.5: drop temperature, top_p, top_k
Anthropic says to omit temperature, top_p and top_k on Haiku 5.5, and some values return a 400. Strip them from your loop before the first Sume tool call.
HelixBench: what spatial-acoustic consistency tests in AI video
HelixBench scores whether synthesized sound fields track camera movement. What the arXiv page says it measures, and a manual way to spot-check any AI clip.
Image-to-video API with sound: five-second cost by Sume row
Omni Flash 1.1 (always on), Wan 3.0 (audio), MiniMax H3 and H3 Max (native stereo) take a first frame and render sound. A 5 s clip is $0.19 to $0.94 on Sume.
Long AI video clips on Sume: which rows pass 15 seconds
Only two Sume rows pass 15 seconds in one job: Seedance 2.5 (4-30 s) and Wan 3.0 (2-30 s). All other rows stop at 15 s or less. Read 2026-10-13.
MAI-Voice-2.1 emotion control vs Sonic's six emotions on Sume
Microsoft lists granular emotion control for MAI-Voice-2.1. On Sume, Sonic 3 rows take six named emotions in generation_config.emotion; Eleven rows have none.
Sume video model ids in one table (October 2026)
All Sume video ids with resolution and length limits: Seedance, Kling 3, Wan 3.0, H3, H3 Max, Omni, Grok, plus Google Cloud and Tencent rows. Read 2026-10-13.
Decision-1 GlobalStandard or DataZoneStandard for a Sume gate?
Microsoft lists both deployment types for Decision-1. DataZoneStandard keeps processing in a data zone; GlobalStandard may vary more in latency. Benchmark.
Nano Banana 2.1 thinking levels and search grounding: not on Sume
fal documents thinking levels, web-search grounding and a system prompt for Nano Banana 2.1. Sume's /v1/images row lists none and returns 400 for them.
Nova 2.5 Sonic turn-taking: does a voiceover need it?
No. Controllable turn-taking and mixed voice and text sessions in Nova 2.5 Sonic serve live agents. A rendered voiceover is one script in, one file out.
Qwen Image 2.1 Turbo API price: $0.016 per image, and Sume's Qwen row
Alibaba lists qwen-image-2.1-turbo at $0.016 per image in Singapore. Sume does not list 2.1 yet; its Qwen Image row bills $0.025.
Seedance 2.5 pro tools vs the Sume API: which fields exist
Sume exposes Seedance 2.5 by prompt, frames, references, duration and resolution. No white-model, green-screen or blocking field is documented. Read 2026-10-13
Sonic and Eleven v4 share 44 language codes on Sume
Sume's TTS Router lists the same 44 language codes for Sonic 3 and for Eleven v3 and v4, so the language field alone does not pick the engine.
AI video models with 4:3 and 3:4 on Sume: which rows list them
For 4:3 or 3:4 video on Sume, use a Seedance, Wan 3.0 or H3 row. Kling 3, Omni and Veo do not list those ratios. Read 2026-10-13.
Veo 3.1 on Sume: Google Cloud rows, 720p, text-to-video only
Veo 3.1, Fast and Lite appear on Sume only where Google Cloud serves them: text-to-video, 720p, 4, 6 or 8 s, 16:9 or 9:16. Check your own list. Read 2026-10-13.
WAND-Vega for e-commerce video: what Tencent says, what Sume lists
Tencent describes WAND-Vega as trained for vertical industries such as e-commerce. On Sume it is three text-to-video rows, 4-15 s, up to 4K. Read 2026-10-13.
Which video_url row to call: Omni edit, Recast or Motion Transfer
Three Sume rows accept video_url: gemini-omni-flash-1.1 edits by prompt, h3-max-recast swaps people, higgsfield-genjutsu moves motion. Limits and prices.
Whistle's 86,174-utterance benchmark: test your own clips first
Cactus Whistle's card reports word error rates over 86,174 utterances. Score 20 of your own clips against a human transcript before choosing it over hosted STT.
Whistle speech model: Apache 2.0, 16.9 MB, offline or Sume STT?
Cactus Whistle is a 16.9 MB Apache 2.0 speech-to-text model for seven languages. Run it on-device, or send the clip to Sume STT for other languages.
Can you fine-tune Qwen-Image-2.1-Turbo? Alibaba's pages disagree
Alibaba's Turbo page says it supports fine-tuning, its fine-tuning overview does not list it. Test with reference images on Sume before you pay to train.
Muse Wonder reviews a video: send Sume inspect stills and transcript
Muse Wonder takes video and audio in but replies only in text. To have it review a Sume render, use video_inspect: 8 stills by default, 24 at most.
Nano Banana 2.1, Pro, GPT Image 2.5, Seedream or FLUX.2: 5 checks
Pick between five Sume image rows by elimination: 1:8 ratio, 11+ references, transparent PNG, native 4K, then price from 3.75 to 18.75 cents.
Midjourney Rerun (Thinking) is alpha-only: the Sume workaround
Midjourney's Rerun (Thinking) button retries a job that missed the prompt, but only on the alpha site. Here is the nearest retry loop with Sume's image API.
HelixWorld needs ~70 GB of VRAM and has no hosted API: what to use
HelixWorld is a real-time audio-visual world model with open code but no hosted API. What it needs to run, what HelixBench measures, and the Sume job route.
HelixWorld spatial audio vs native stereo clips on Sume
HelixWorld renders camera-aligned stereo sound in real time. Sume's video rows give native stereo or synced audio per clip, not live spatial audio. Compare.