Summer 2026 video model features mapped to Sume routes, and the gaps
Seedance 2.5, Wan 3.0, Omni 1.1 Flash and Ray3.2 launched features this summer. Which map to a Sume route, which are vendor-only. Read 2026-10-05.

Of the new video features announced this summer, the ones with a documented Sume route are: 30-second generation (seedance-2.5, wan-3.0), reference-guided generation, first and last frames, Omni video edit, and assembly with Timeline 1.0. Vendor-only, with no matching Sume page, are: Ray3.2 (not in the Sume catalog), Omni's scene extension to 40 seconds, Wan 3.0's document references, and Seedance 2.5's green-screen replacement and timestamp-level editing as model features. This table separates them, with each row dated to when the vendor page was read.
Feature map
"No dedicated route" and "not shipped" describe the Sume docs and catalog I read, not the vendors' models.
| Vendor feature | Vendor | On Sume |
|---|---|---|
| 30 s per generation | Seed (Seedance 2.5), Alibaba (Wan 3.0) | Yes: seedance-2.5 4 to 30 s, wan-3.0 2 to 30 s |
| Many references per pass | Seed: 30 images, 10 video, 10 audio; Alibaba: 20 assets | Reference fields exist; counts are per model in the catalog |
| Scene extension to 40 s | Google (Omni 1.1 Flash) | No extension parameter documented; Omni jobs are 3 to 10 s |
| 360p draft mode | Yes: Omni 1.1 on Sume lists 360p | |
| Video edit by prompt | Google (Omni), Seed, Alibaba | Omni edit via video_url plus prompt |
| Green-screen replacement | Seed | No dedicated route in the pages I read; video filter is a filters-only ffmpeg graph |
| Timestamp-level audio/video editing | Seed | Build it from video trim, audio detach, timeline audio and Timeline render |
| Auto scene splitting | Alibaba | Explicit scenes via Timeline video[] slots |
| 16 keyframes, EXR export | Luma (Ray3.2) | Not shipped; first and last frames only on supported models |
Where each gap leaves you
- Extension past a model's cap: two jobs plus a Timeline join, with a late frame from video frames as the next first frame.
- Background replacement: regenerate with a new background prompt and references, or use Omni edit on the clip.
- Document inputs: summarize into a prompt and attach images.
- HDR delivery: not available through Sume's video catalog.
Check live before you build
Catalogs change. List the models and read the capability fields on the day you build:
curl https://api.sume.com/v1/video-router/models \
-H "Authorization: Bearer $SUME_API_KEY"How to use this table
Use the table as a first filter, not a verdict. If a feature is a gap, check whether the job needs it or just resembles it. A background swap, for example, might be solved by a new generation with a different background prompt, and a 40-second shot might be better as two shots anyway.
Re-read the vendor pages before you commit to a feature in a client brief, because vendors update. The Sume docs also change, so check the model list and the page for each route on the day you build.
- Needs the feature, or something like it?
- Re-read vendor pages before promising.
- Check live catalog fields on build day.
Dates matter
The vendor pages were read on 2026-10-05: Seedance 2.5 launched on 2026-07-31, Wan 3.0 about 2026-08-24, Omni 1.1 Flash on 2026-08-27 and Ray3.2 on 2026-06-09. Each of these vendors updates its pages, and Sume's catalog changes as well, so a feature in the gap column today may have a route next month. Check before you rule something out.
Vendor sources: Seed, Wan README, Google, Luma.
Sources
Related posts
More in Models
- Talking video, face swap, Recast or motion control: pick by your input
Four Sume routes make a person on video. Which one fits depends on whether you have a script, a source video, a photo of each person or a motion clip.
- Tavus Human Interaction Model: what Griffin is, and what ships today
Tavus calls Griffin a Human Interaction Model. Only select testers have Griffin-Lite. Here is what it is, and what ships today on Sume for a talking presenter.
- Thai, Vietnamese, Indonesian voiceover: on MAI's list, not Sume's tags
MAI-Voice-2.1 lists th-TH, vi-VN and id-ID; Sume's 16 voice tags do not. What that means, a 1-cent audition, and the Unicode trap in Vietnamese.
- Translate text in an image but keep brand names: the edit prompt
A prompt pattern for translating the words in an image on Sume while leaving brand names, prices and codes alone, with an Ideogram 4.5 request.
Written by Sume