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
Does Kling 3.0 do 9:16? Its guide is silent; Sume lists 3 ratios
Kling's Kling 3.0 guide gives 720p, 1080p and 3 to 15 s but no aspect ratios. Sume's kling-3 row lists 16:9, 9:16 and 1:1. Read the catalog to confirm.
Kling 3.0 multi-shot: Sume's kling-3 has no field, so cut on Timeline
Kling's guide describes automatic and custom multi-shot for Kling 3.0. Sume's kling-3 request has no multi-shot field; join separate clips with Timeline.
LTX-2.3 makes 9:16 clips up to 20 s; what Sume offers instead
fal lists LTX-2.3 with portrait 9:16 and up to 20 seconds. Sume's catalog has no LTX row; Wan 3.0 and Seedance 2.5 reach 30 s, Kling 15 s.
mask_url returns 400 on Nano Banana 2 and Ideogram 4.5: your options
Only ChatGPT Image 2.5 accepts mask_url on Sume; other models answer 400 unsupported_parameter. Three ways to still change one region of a photo.
Nano Banana 2 image search grounding: what Sume lets you send instead
Google lists Web Search and Image Search grounding for Nano Banana 2. Sume's image catalog has no grounding field, so fetch the references and pass up to 10.
One voice in 23 languages: Microsoft's claim vs Sume's language guard
Microsoft says MAI-Voice-2.1 uses one voice across 23 languages. Sume sets language per job and warns on a voice mismatch. What that means when you buy.
Three image edits at once or one by one? Cost and drift on Sume
FLUX 3 Image edits several boxes in one pass. On Sume, one prompt with three numbered changes costs one edit; three chained edits cost three and can drift.
Veo 3.1 Lite 1,024-token prompt limit vs Omni's 20,000 characters
Google's Veo 3.1 Lite page caps text input at 1,024 tokens. Sume's gemini-omni-flash-1.1 row takes 20,000 characters. What that means for long shot lists.
Which Sume image models cannot edit a photo? Five reject references
Imagen 4 Fast and Ultra, Recraft V4, Qwen Image Max and Soul are text-to-image only on Sume. List them from the catalog in 10 lines before you wire up an edit.
4:1 and 8:1 panorama images: only Nano Banana 2 lists them on Sume
Of Sume's 19 image rows, only Nano Banana 2 lists 4:1, 1:4, 8:1 and 1:8. Price is $0.10 at 1K, $0.15 at 2K and $0.20 at 4K. A 5-line request.
Is Z-Image Turbo on Sume? Not listed: the cheapest rows instead
Sume's image catalog on 2026-10-04 has no Z-Image row. Check with one request and see the cheapest listed models: Soul $0.005, Grok and Qwen $0.025.
9:8 aspect ratio image: which Sume rows list it
Of Sume's 19 image rows, only the three ChatGPT Image rows list 9:8, along with 5:4 and 4:5. What the other rows do with 9:8 and how to crop.
Grok Imagine on Sume: n is 1, so four takes is four calls
Grok Imagine's image row on Sume caps n at 1 and costs $0.025. Four takes means four calls and $0.10. Other models accept n up to 4.
Is Ideogram 4.0 open source? Apache code, non-commercial weights
Ideogram 4.0's inference code is Apache 2.0, but the weights on Hugging Face ship under a non-commercial agreement. Selling the images needs a paid licence.
Which Sume video model fits your inputs: text, photo, clip, audio
Match the input you hold to a Sume video model: prompt, first frame, end frame, references, audio sample, or a clip to edit. With the 400s each mix causes.
Shortest AI video clip via API: 2 seconds on Wan 3.0 only
Sume's video catalog by minimum duration: Wan 3.0 takes 2 seconds, Omni 3, Seedance and Kling 4, MiniMax H3 5. Plus how to trim below a floor.
AI voice agent latency budget: who owns which 100 milliseconds
Vendor numbers for speech-to-text, the language model and text-to-speech side by side, with a note on where an async file API like Sume belongs and where not.
Which AI video models take 1080p on Sume, and which do not
Seedance, Kling, Wan and Omni accept 1080p on Sume; H3 Max refines to it from native 768p; H3, Grok and Genjutsu stop lower. Full matrix.
FLUX 3 Image 768sq and 1.5k tiers vs Sume's 512, 1K, 2K, 4K
FLUX 3 Image sells 768sq, 1k, 1.5k, 2k and 4k; Sume's Nano Banana tiers are 512, 1K, 2K, 4K. A Python tier translator that rounds up and checks the catalog.
Silent AI video: generate_audio false or drop the audio after
Seedance, Kling and Wan take generate_audio false; Omni and MiniMax H3 always make sound. How to get a silent clip on Sume, and what it costs.
Omni Flash vs H3 Max for reference-to-video: limits side by side
Gemini Omni Flash 1.1 takes 10 images and 3 short videos; MiniMax H3 Max adds audio refs, 12 files total. Limits, tags and a sample request on Sume.
FLUX 3 Image open weights: release date and what to use now
FLUX 3 Image's open weights are reported as coming within weeks, with no date. What is known, what is not, and the hosted FLUX.2 route on Sume today.
Which Sume image models make 2K or 4K output, by model
FLUX 3 Image added 4K; Sume's catalog has two ways to ask for big images, a resolution tier or custom pixels. Which models take which, and the 3840 edge cap.
Is MiniMax H3 Max open source? What the H3 card says and Sume lists
MiniMax H3 has a Hugging Face card under a community licence. The card lists no H3 Max weights. Compare the two Sume rows with one script.
ChatGPT Try On from a screenshot: the same edit through an API
ChatGPT Try On starts from a selfie plus a product screenshot. Do the same edit with openai/gpt-image-2.5 on Sume: two references, one prompt, one Python call.
Is FLUX 3 Image open source? Commercial weights or the BFL API
BFL offers FLUX 3 Image weights under a commercial licence for companies running it at scale, or through its API. What that means, and what Sume lists today.
Kyutai Pocket TTS languages: what it speaks vs hosted Sume TTS
Pocket TTS is a 100M-parameter open model you run yourself. Its README lists seven languages; Sume's TTS is hosted, per character, with a voice id.
Pocket TTS languages: six or seven, and Sume's language field
Kyutai lists six Pocket TTS languages on its model card and blog, seven in the GitHub README. Here is how to read that, and how Sume TTS sets a language.
Cheap AI video drafts: which models start at 480p or lower on Sume
Seedance, Wan 3.0, MiniMax H3 and Grok Imagine start at 480p, Gemini Omni Flash at 360p, Kling 3 at 720p. Where a draft is possible before a 1080p final.
AI video audio by model: toggle, always on, or none on Sume
Seedance, Wan 3.0 and Kling 3 let you switch audio; MiniMax H3 and Gemini Omni Flash always make it; Grok Imagine has none. What each row does.
How to choose an AI video model: five questions before you pin one
Length, resolution, audio, start image and references decide the model, not the leaderboard. Five questions mapped to Sume's video catalog.
Which AI video models can't do text-to-video? Rows that need a source
Grok Imagine Video 1.5, Genjutsu Motion Transfer and H3 Max Recast refuse a prompt-only request. What each needs, and which rows accept text alone.
AI video reference limits: how many images and clips per model
Reference image and reference video caps for Wan 3.0, MiniMax H3, Gemini Omni Flash, Genjutsu and H3 Max Recast on Sume, in one table with the odd limits.
Clone a voice from an MP4 or WebM file: Sume Voices takes video, 20 MB
Sume's Voices clone accepts wav, mp3, m4a, mp4, webm and ogg up to 20 MB. How to prepare a clean clip from a video, and what the upload check rejects.
End-frame AI video: which Sume models take a last frame
Seedance, Wan 3.0, Kling 3, MiniMax and Gemini Omni Flash accept a first and last frame; Grok Imagine and the swap rows do not. How to send both frames.
Gemini Omni Flash 4K: Google says it is upscaled, so what do you get?
Google's Omni page calls 1080p and 4K upscaled. What that means for a 4K order, and when to upscale a 720p clip yourself on Sume instead.
Higgsfield Soul on Sume: text to image, 1 or 4 images, 720p or 1080p
Soul is a text-to-image row in Sume's image catalog: seven aspect ratios, no references, 1 or 4 images per call, 720p or 1080p. Limits and price.
Longest AI video clip in one request: 30, 15 or 10 seconds by model
Seedance 2.5 and Wan 3.0 reach 30 seconds on Sume; most other rows stop at 15 and Gemini Omni Flash at 10. Ceilings per model, and when to stitch instead.
Polish, Dutch, Swedish, Turkish text to speech API: Sume pl nl sv tr
Sume's Voices library tags voices pl, nl, sv and tr alongside 12 other languages. What to send for each, and how Eleven v4's list compares.
Chinese, Hindi and Russian text to speech API: Sume zh, hi and ru
Sume's Voices library has zh, hi and ru tags. How to request each, what Eleven v4 lists, and the one field that stops an English-sounding read.
Reference audio for AI video: which models accept a voice clip
Seedance 2.5, Wan 3.0 and MiniMax H3 take reference audio on Sume; Kling 3, Gemini Omni Flash and the swap rows do not. Limits and the one-reference rule.
Seedance 2.5 timestamp-level editing: what Sume exposes today
ByteDance describes timestamp-level editing and 30-second clips for Seedance 2.5. See which of those fields Sume's Video Router accepts for seedance-2.5.
Shortest AI video clip by model: 2, 3, 4 or 5 second floors on Sume
Wan 3.0 starts at 2 seconds; Gemini Omni Flash at 3; Seedance and Kling at 4; MiniMax H3 at 5. The minimum duration for each Sume video model.
Sume auto or a pinned video model: what changes past 10 seconds
Auto video defaults to 720p and 8 seconds with 3 to 10 second clips. For 12, 15 or 30 seconds, 1080p or 21:9, pin a model id. How to decide on Sume.
9:16 vertical AI video: which Sume models take an aspect ratio
Seedance, Wan 3.0, Kling 3, MiniMax H3 and Gemini Omni Flash take 9:16; Grok Imagine, Genjutsu and H3 Max Recast take no aspect ratio. A per-model table.
AuK: MIT speech model that edits audio, and what Sume covers
AuK is Tencent's 1.5B MIT-licensed speech model for generation and editing. What its card lists, what it omits, and which parts Sume's audio tools cover.
DMAD 4-step MiniMax H3: 50-step baseline vs Turbo LoRA
DMAD distills MiniMax H3 from 50 steps to 4. Why that 4 is not the Turbo LoRA's 4, what the card says, what it omits, and where hosted jobs fit.
Fastest Seedance or Kling model on Sume: latency labels
Sume labels seedance-2-fast and seedance-2-mini fast, seedance-2.5 and seedance-2 medium, and kling-3 medium-slow. What the labels mean and how to test.