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
Gemini Omni Flash GA: extension and interpolation vs Sume's inputs
Gemini Omni Flash is generally available with extension and interpolation between images. What Sume's gemini-omni-flash-1.1 catalog entry documents instead.
Seedance 2.5 on BytePlus excludes the US: check access first
BytePlus lists Seedance 2.5 on ModelArk in supported markets excluding the United States. What that means for a US team, and how to read the Sume catalog.
PixVerse V6 native audio and camera work: what to check in an API
PixVerse's blog lists V6 with camera work and native audio, plus R2 and a $439M Series C total. How to test those claims against any video API's catalog.
MiniMax H3 limits: 9 images, 3 videos, 3 audio, file caps
MiniMax's H3 guide caps prompts at 7,000 characters and references at 9 images, 3 videos and 3 audio files. Cheat sheet with the Sume limits beside it.
Scribe v2, Realtime or Medical: which model for interview clips
ElevenLabs lists Scribe v2 with up to 32 speakers and 90+ languages, Realtime at about 150 ms, and Medical with 35% fewer errors. Which fits interview clips.
PixVerse V6 adds native audio: which Sume video ids make sound
PixVerse V6 ships camera work and native audio. The Sume video docs name no PixVerse id, so here is how to find models that generate audio and read the flag.
HunyuanVideo 1.5 on 14 GB of VRAM: run locally or call an API
The HunyuanVideo 1.5 repo lists 8.3B parameters, 480p to 1080p and a 14 GB VRAM minimum with offloading. A local-run versus API checklist.
Self-host MiniMax H3 or use an API: a decision table
MiniMax released H3 as an open-weight omni-modal model on Jul 31, 2026. A decision table for running it yourself versus calling a hosted video API.
Voxtral Mini Transcribe 2 and Realtime v26.02: what Mistral lists
Mistral lists Voxtral Mini Transcribe 2, Voxtral Realtime v26.02 and Voxtral TTS v26.03. How to prepare video audio for any transcription model.
What is FLUX 3? The family map: image, video, audio, action
BFL's docs describe FLUX 3 as one family covering image, video with synchronized audio, audio and action. Which pieces have open weights and which do not.
Veo 3.1 extend adds 7 s up to 20 times, 720p only: plan a long clip
Google's Veo docs let you extend a clip by 7 seconds up to 20 times, at 720p only. Read the arithmetic, then compare it with one-call 30 second models on Sume.
Wan 3.0 doubles clip length to 30 seconds: Alibaba ids vs Sume
Alibaba Model Studio lists wan3.0-video and wan3.0-video-prime at 2 to 30 seconds, up from 15 on Wan 2.7. What Sume's wan-3.0 accepts and a 30 s request.
Gemini 3.8 Flash TTS tops Hume's VoiceEQ board: run your own test
Hume's blog lists Gemini 3.8 Flash TTS atop its Real-World VoiceEQ board. Why a vendor-run board is only a lead, and how to run a blind A/B on your script.
Lyria 3.5 blocks artist-voice prompts: how to write briefs that pass
Google's Lyria 3.5 docs note that prompts asking for specific artist voices are blocked. Describe the sound instead, then run it through the Sume Music Router.
GPT Image aspect ratios: Firefly's 10 vs Sume's 17
Firefly's September 2026 update gives GPT Image 2 ten aspect ratios instead of three. Sume's image API documents 17 ratios plus auto, and custom pixels.
Eleven v4 stacked tags: direct emotion without them
ElevenLabs v4 adds stackable expression tags and 10-second cloning. Sume's docs list no clone route; here is how to direct a voice via tts_create.
Wan 3.0 two-second clips for Reddit video ads
wan-3.0 on Sume accepts 2 to 30 seconds, so a 2 s Reddit video ad is a single request. Platform minimums vary; check them against the catalog first.
SF3D makes a GLB in 0.5 s: what it skips
Stability's SF3D is listed at $0.07 on fal.ai, GLB only, about 0.5 seconds. Sume does not carry 3D, so here is what to use for product visuals instead.
Wan 3.0 takes 10 images, 5 videos, 5 audio clips: Sume check
Wan 3.0 accepts ten images, five videos and five audio clips per prompt. Sume honors audio and video references on Wan 3.0, but caps vary; read the catalog.
Wan 3.0 has no open weights: call it by API on Sume
llm-stats reports no open-weight Wan 3.0 and WAN 2.2 as the last open Wan release. Call the hosted wan-3.0 id on Sume through POST /v1/videos.
Qwen-Image-2.1 is non-commercial: hosted options
Qwen-Image-2.1 has open weights but a Qwen Research License that is non-commercial unless licensed. Check the license first, then the hosted catalog.
Inworld TTS 2 style steering and the Sume voiceover path
What is reported about Inworld TTS 2 style steering and 100+ languages, and how to produce a voiceover with Sume's tts_create tool and join takes.
Hume Octave 2 covers 11 languages: is yours one?
Hume Octave 2 is reported to cover 11 languages. Before you pick a voice engine, check your language, then verify what Sume's tts_create recorded for the job.
Hy Image 3.5 reference limit: 5 or 20, and Sume's caps
Hy Image 3.5's sources disagree on references: 5 in the announcement, 20 in the API guide. Sume's caps differ by model; read input_references from the catalog.
Stable Audio 3 sizes: which one for your job
Stable Audio 3.0 ships Small SFX, Small, Medium and an API-only Large. Match size to job, and see how Sume's Music Router handles length with a prompt.
HunyuanVideo 1.5 SSTA and FP8: or just call an API
HunyuanVideo-1.5 has 8.3B parameters, SSTA for a 1.87x speedup at 720p, FP8 GEMM and a 14GB VRAM floor. Decide whether to run it or call a hosted video API.
AI sound effects: Suno Sounds vs a music prompt
Suno Sounds V5.5 generates sound effects for 2 credits flat. Sume has no SFX endpoint; a Music Router prompt with no vocals gives ambient audio you must check.
MiniMax H3 Max: the prompt-adherence variant on Sume
fal describes MiniMax H3 Max as tuned for prompt adherence. On Sume, minimax-h3-max runs 480p to 1080p for 5 to 15 s with frames and references.
Genjutsu 30 reference images vs Sume's 1 to 8
Higgsfield's Genjutsu takes up to 30 reference images. Sume's higgsfield-genjutsu takes one source video plus 1 to 8 reference images, at 480p or 720p.
Trending TikTok metadata to a Seedance 2.5 clip
Sume's trending-videos search returns TikTok metadata for $0.10 a call, not video files. Use it to write a brief, then render an original Seedance 2.5 clip.
Nano Banana 3: does it exist? Current ids
Google autocomplete suggests a Nano Banana 3, but suggestions are not releases. What autocomplete lists and which Nano Banana id Sume accepts.
Lip sync with a hand over the mouth: sync-3 vs Fabric on Sume
Sync Labs says sync-3 handles obstructions on faces. Sume's lip-sync routes start from a still plus audio. What each takes as input and what is promised.
Pixal3D multi-view: prepare the input images
Pixal3D added multi-view inference in September 2026 under an MIT license. Sume has no 3D, but reference-image edit can prepare input views.
11-second AI video API: what works past Omni's 10 s cap
Gemini Omni Flash 1.1 stops at 10 s, so an 11 s clip needs H3, H3 Max, Wan, Kling or Seedance. Which Sume ids accept 11 s and what each bills.
13-second AI video API: MiniMax H3 at 768p and other Sume ids
Thirteen seconds fits MiniMax H3 and H3 Max (5-15 s), Wan 3.0, Kling 3 and Seedance. Billed price per resolution for a 13 s clip on Sume.
14-second AI video API: one Sume request or stitched extensions
Fourteen seconds is one request on Sume with H3, Wan, Kling or Seedance. Veo reaches that length only by extension. Prices for a 14 s clip inside.
17-second AI video API: only Wan and Seedance 2.5 clear 15 s
Seventeen seconds is past the 15 s wall of H3, Kling and Seedance 2. Wan 3.0 and Seedance 2.5 accept it; Wan bills 17 s at 480p to 1080p below.
18-second AI video API: draft at Wan 480p, then render final
An 18 s Wan 3.0 clip bills three different amounts on Sume, one per resolution. Draft at 480p, check the cut, then pay for 720p or 1080p once.
21-second AI video API: Seedance 2.5 or Wan 3.0 by aspect ratio
A 21 s clip on Sume means Seedance 2.5 or Wan 3.0. Seedance supports 21:9 and Wan does not. Eight Seedance ratios, seven Wan ratios, and Wan prices.
28-second AI video API: near the 30 s ceiling on Sume
A 28 s clip is one request on Wan 3.0 or Seedance 2.5, jobs are async, and status moves pending to completed. Price for 28 s and how to poll.
7-second AI video API: which Sume ids take 7 s and the price
Veo documents 4, 6 and 8 second clips, so 7 is not an option there. All twelve Sume ids accept 7 s; Omni Flash 1.1 at 360p is cheapest.
9-second AI video API: Omni takes 9 s, Veo does not
A 9 second clip fits Gemini Omni Flash 1.1 (3-10 s) and most Sume ids, but not Veo 3.1, which tops out at 8 s. Billed prices for 9 s inside.
Music from a thumbnail: image_url on Sume's music router
Generate a music bed that matches a still: pass one public HTTPS image_url with the prompt to Sume's Music Router, and clear it with null when reusing objects.
MAI-Voice-2.1 emotion control vs Sume's emotion field
Microsoft lists emotion control on MAI-Voice-2.1. Sume's TTS takes a free-text emotion string, speed and volume. What each gives you, and how to test it.
Tidal stopped paying wholly AI tracks: what a video score is
Tidal stopped paying royalties on wholly AI tracks on July 15, 2026. What that means for a video soundtrack, and how Sume's Music Router fits.
Where can I use MAI-Voice-2.1 today, and what Sume offers instead
Microsoft lists its Playground, Copilot Audio Expressions and Foundry; the launch post adds OpenRouter and Vercel. What Sume's text to speech gives you instead.
Kling 4.0 Flash is 720p only: Sume models that reach 1080p and 4K
Kling 4.0 Flash early access outputs 720p, 3 to 20 seconds. If you need 1080p or 4K today, these Sume models take those resolutions, with the exact prices.
Seedance 2 mini, Seedance 2.5 or Wan 3.0 for holiday ad drafts?
At 480p a 5-second draft costs $0.3125 on Wan 3.0, $0.4394 on Seedance 2 mini and $1.3434 on Seedance 2.5. Here is when each is worth it.