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
Which Nano Banana models can you call on Sume? Google lists five
Google's Gemini API lists five Nano Banana models; Sume's image catalog lists two, Nano Banana 2.1 and Pro. Nano Banana 2 ids still run, as 2.1.
Should a Shorts season use sume/auto or a pinned video model?
Pin one model per season so episodes look alike. Use sume/auto for pilots and one-offs where a consistent look is not required.
Is there a seed to keep Shorts characters the same across episodes?
No. No video model in Sume's catalog accepts a seed, and the docs say a seed is not deterministic on all providers. Use references and first frames instead.
Keep a Shorts series consistent with Omni Flash video references
Gemini Omni Flash 1.1 on Sume accepts image and video references for 3 to 10 second clips. Feed the last episode's clip in as the next one's reference.
Does Wan 3.0 allow 2-second clips for a Shorts cold open?
Yes. Wan 3.0 on Sume accepts 2 to 30 seconds per call, so a 2-second cold open at 720p is $0.25, and it is the cheapest hook test in a season.
Fix one detail in a Shorts episode: Omni Flash edit or re-roll?
Gemini Omni Flash 1.1 can edit an existing clip through video_url instead of regenerating it. A fresh 10-second re-roll at 720p costs $1.25 on Sume.
Recast the actor in a short video with AI: H3 Max Recast, TikTok rules
H3 Max Recast swaps the people in a 5-30 s source video for 1-4 photos you supply. TikTok's Shop pages ban impersonating public figures and need AI disclosure.
Fix a wrong product color in an AI clip with Gemini Omni Flash edit
TikTok Shop bans AI that changes a product's color, size or material. Gemini Omni Flash 1.1 on Sume edits one clip from a prompt instead of re-rolling it.
MAI-Voice-2.1 or Flash for narration? Microsoft's own guidance
Microsoft lists MAI-Voice-2.1 for long-form narration and Flash for agents. The prices, the latency claims and which one to test first for voiceover.
Which Sume video models take 30 seconds: four ids in the catalog
Four ids in Sume's video catalog reach 30 seconds: seedance-2.5, wan-3.0, higgsfield-genjutsu and h3-max-recast. Which make a clip from a prompt.
Shortest AI video on Sume: Wan 3.0 at 2 s, Seedance 2.5 at 4 s
Wan 3.0 takes 2-second clips on Sume; Seedance 2.5 starts at 4 s. A 2 s Wan clip at 480p is $0.125 on Sume. Duration floors across the catalog.
Which AI video models take an audio file as a reference on Sume?
MiniMax H3 and H3 Max, Wan 3.0 and the Seedance 2.x rows accept audio references on Sume; Omni and Kling do not. H3 limits: 3 clips, 2-15 s each, 15 s total.
Omni replaces Veo in the Gemini app: does your API code change?
Google's page says Omni replaces Veo in the Gemini app. What that means for API code, and the one Omni row Sume lists, with its limits and per-second price.
MiniMax says H3 edits across modalities. Which Sume row edits a clip?
MiniMax lists reference and editing across modalities for H3. On Sume, H3 does text, image and reference video; clip edit by prompt is Omni's video_url mode.
MiniMax H3 open weights: what is released and what stays API-only
MiniMax opened H3-Base on 2026-08-03 with two checkpoints, but 2K regeneration and Context-IR stay API-only. How that maps to Sume's minimax-h3 rows.
First Draft only cuts clips you have: generate the b-roll first
A starting edit needs footage. Generate 9:16 cutaways with Seedance 2.5 or Gemini Omni Flash 1.1 first: durations, aspect ratios and one Video Router call.
Wan 3.0 references on Sume: 10 images, 5 videos, 5 audios
Wan 3.0 takes more reference files per Sume job than Seedance 2.5: up to 10 images, 5 videos and 5 audios. How to send them and what the API rejects.
Seedance 2.5 references: ByteDance says 30 images, Sume allows 9
ByteDance describes up to 30 images, 10 video clips and 10 audio clips for Seedance 2.5. Sume's API request check caps one job at 9, 3 and 3, 12 files total.
Gemini Omni Flash 1.1 from a still: 16:9 or 9:16 only, cost 360p to 4K
A still-to-video handoff to Gemini Omni Flash 1.1 on Sume must be 16:9 or 9:16. A 5-second clip costs $0.1875 at 360p up to $1.88 at 4K.
Does Seedance 2.5 generate audio through the Sume API?
Yes: Seedance 2.5 on Sume has an optional generate_audio flag. ByteDance describes joint audio-video generation. What to send and how it is priced.
Wan 3.0 audio on Sume: generate_audio and audio references
Wan 3.0 on Sume can make its own audio with generate_audio or take up to 5 audio references. What to send, and how the 30-second price changes (it does not).
Cheapest model for a 5-second vertical hook with sound on Sume
For a 5-second 9:16 hook, Omni Flash at 360p is $0.19 and Wan 3.0 at 480p $0.31; at 720p both are $0.625, and Seedance 2.5 at 480p is $1.34.
Gemini 3.8 TTS: speech_metadata for a whole turn, tags for one moment
Gemini 3.8 Flash TTS styles delivery two ways: speech_metadata per turn and angle-bracket tags for a cough or pause. When to use which, and the Sume side.
Hy Image 3.5 Preview dated id 20260922: should you pin the snapshot?
The Hy Image 3.5 Preview listing carries a dated id ending 20260922. Pin it for repeatable edits; float only if you want improvements. How to decide.
Image aspect ratios per model: GPT 2.5 lists 8, Banana 2.1 lists 14
On Sume, ChatGPT Image 2.5 lists 8 aspect ratios, Nano Banana 2.1 lists 14, Ideogram 4.5 lists 15. A table of which ratios each model accepts, from the catalog.
Omni Flash 10-second clip price at 360p, 720p, 1080p and 4K
A 10-second Gemini Omni Flash 1.1 clip costs $0.375 at 360p, $1.25 at 720p, $1.875 at 1080p and $3.75 at 4K on Sume, with synced audio included.
Where can I use Gemini 3.8 TTS? API and Studio now, Enterprise later
Gemini 3.8 Flash TTS and Flash-Lite TTS availability from Google's launch page: Gemini API, AI Studio, Notebook, Vids, and what is still coming soon.
Ideogram 4.5 low, medium or high: a text grid to pick the tier
Run one headline through Ideogram 4.5 at low, medium and high on Sume, read the billed cost of each, and keep the cheapest tier whose text you can read.
Accented letters in AI image text: a proof list for Spanish and French
Headlines with é, ñ, ü or ¿ can come back with a wrong accent. Test two Sume models on your real copy, then check every mark against a proof list.
Bumper sticker art: a 3:1 strip at 1536x512 on GPT Image 2.5
Make a wide bumper sticker on Sume with GPT Image 2.5 at a custom 1536x512 size, why 3:1 is its widest shape, and what to do for longer stickers.
ChatGPT Images 2.5 Sketch: the API way is a drawing as a reference
ChatGPT Images 2.5 added a Sketch feature. On Sume there is no sketch canvas, but a drawing sent as an input reference to GPT Image 2.5 does the same job.
LTX-2.5 audio-to-video: three ways to drive a clip with audio on Sume
LTX-2.5 is not on Sume. To make video that follows a voice or track, use audio references on Seedance 2.5, Wan 3.0 or H3, or H3 Max lip sync. Limits and prices.
MiniMax H3 reference limits: 9 images, 3 videos, 3 audios on Sume
MiniMax H3 reference-to-video on Sume takes up to 9 images, 3 videos and 3 audio clips, 12 in total. The duration rules, the extra-image fee and a request.
Native 4K AI video API: LTX-2.5 claims 4K, what Sume offers instead
LTX-2.5 is listed up to 4K but is not in Sume's catalog. Sume's 4K options are Gemini Omni Flash 1.1 and MiniMax H3 4K upscales. Limits and per-second prices.
Kling 3 as your Sora replacement on Sume: four limits to check
Kling VIDEO 3.0 is 15 s with native audio per a video tracker. On Sume kling-3 takes 4-15 s at 720p or 1080p and no reference URLs. Four checks before a port.
AI video models with sound: which toggle it, which are always on
Seedance 2.5, Wan 3.0, Kling 3 and MiniMax H3 all return audio on Sume. Kling prices audio separately; H3 and Omni cannot turn it off. What each does.
Which new video model for which job: Seedance, Wan, Kling, H3, Omni
A decision table for the five new video models on Sume, by clip length, resolution, sound and price per second. Veo 3.1 and LTX-2.5 are not on Sume.
gpt-image-2 or gpt-image-2.5 to replace gpt-image-1 on Sume?
OpenAI names gpt-image-2.5 as the gpt-image-1 replacement. On Sume, gpt-image-2 and 2.5 differ on references, mask_url, background and quality tiers.
seedream-v4 or bytedance-seed/seedream-4: which Sume id to store
Sume accepts legacy bare ids like seedream-v4 and seedream-4-5-edit, but the catalog lists org/slug ids. Store the canonical id; Python normalizer inside.
Ideogram 4.5 vs Nano Banana 2 for multi-turn edits: cost per turn
Ideogram 4.5 claims clean multi-turn edits. Compare it with Nano Banana 2 on Sume: price per turn, six-turn totals and a drift test you can run.
Retired AI model id to Sume id: GPT Image, Gemini and Sora lookup
One lookup table that maps retired gpt-image-1, gemini-2.5-flash-image and sora-2 ids to Sume ids, with dates, so a failing call tells you what to switch to.
Nano Banana reference limits: Gemini slots vs Sume input_references
Google counts Nano Banana references by tier and by object versus character. Sume exposes one input_references range per row. How to read it and what to send.
Seedance 2.5 reference-to-video: what Sume accepts per pass
ByteDance lists up to 30 images, 10 videos and 10 audio clips per pass for Seedance 2.5. Here is how that maps to Sume's seedance-2.5 reference types.
Seedance 2.5 API: Seed blog says soon, BytePlus says live
Two ByteDance pages disagree on Seedance 2.5 API access. Here is what each says, and the seedance-2.5 id that runs on Sume's video router today.
MiniMax H3 Max 1080p: docs list 768P, Sume lists 1080p
MiniMax's own guide gives H3 Max 480P and 768P. Sume's minimax-h3-max row also lists 1080p. Here is the gap, and what the extra tier means for you.
MiniMax H3 open weights: 768p locally, 2K only hosted
MiniMax's open H3 Base checkpoints top out at 768p. The 2K tier needs a proprietary module. What that means if you want to self-host or call H3 through Sume.
AI video generator for long videos: 30 seconds per pass, then assemble
ByteDance says Seedance 2.5 makes 30 seconds a pass and extends to minutes. On Sume, a video job tops out at 30 s; Timeline 1.0 joins clips to 1800 s.
AI video ad text in 12 languages: Wan 3.0 or burned-in captions?
Alibaba says Wan 3.0 renders text in 12 languages. Sume lists wan-3.0 at 2 to 30 s. When to let the model draw words and when to add captions yourself.