Video-to-video style transfer: three ways to restyle a clip
Higgsfield's Genjutsu Restyle redraws a video in 20+ styles. On Sume you restyle with the Restyle Format, a prompt edit, or Motion Transfer. Compared.

Higgsfield's changelog entry dated 2026-09-30 describes Genjutsu Restyle as redrawing a whole video in a new style while keeping motion and camera, with 20+ animation styles. Sume does not ship a Restyle mode of that name. It offers three video-to-video routes that overlap with it: the sume-restyle Format, the video_url edit mode of Gemini Omni Flash 1.1, and Genjutsu Motion Transfer.
Higgsfield facts are from its changelog, read 2026-10-02. Sume facts are from the Format catalog and Video Router docs, plus the first-party Format descriptions in Sume's code.
What are the three Sume routes?
sume-restyle is a catalog Format; its description in Sume's code says it restyles a clip the user owns, keeping the motion and cuts and rewriting the look with Seedance 2.5 reference-to-video. The Gemini Omni Flash 1.1 video_url mode takes a source clip plus a prompt that describes the edit. Genjutsu Motion Transfer takes one source video and 1 to 8 reference images, and is listed only when its provider is configured.
| Route | Input | Length note |
|---|---|---|
sume-restyle Format | Your clip plus instruction | See the Format's own page |
gemini-omni-flash-1.1 edit | video_url plus prompt | Resolution optional; no duration or ratio field |
higgsfield-genjutsu | video_url plus 1-8 reference_image_urls | 4 to 30 seconds, 480p or 720p |
Which one is closest to Genjutsu Restyle?
sume-restyle is described the closest way: keep motion and cuts, change the look. But the Sume docs do not claim a number of built-in styles; you describe the look in the instruction. Treat any claim of parity with Higgsfield's 20+ styles as untested until you run your own clip.
How do I choose?
Use the Format when you want the agent to handle the steps. Use the Gemini Omni edit when one targeted change fits in a prompt, such as the docs' example of replacing an object. Use Motion Transfer when the new look comes from reference images. Genjutsu Restyle's 20+ styles vs Sume motion transfer inputs goes deeper on that last route.
How do I check the result?
Pull matching stills from the source and the output and compare them; check a restyled clip kept the camera move shows how.
Whichever route you pick, start with a short clip of a few seconds. Style transfer problems, such as flicker or drifting faces, show up fast, and a short test costs little. Keep the source and the output in the same folder with matching names so frame comparisons are easy, and write down the instruction or prompt you used next to each pair.
Sources
Related posts
More in Comparisons
- Vimeo AI dubbing in 29 languages vs Sume: one file per language
Vimeo serves dubbed audio in 29 languages from a single link. Sume returns one finished file per language, so the language switch lives in your player or page.
- Voice cloning consent statement example: Azure, Google, OpenAI
Azure, Google and OpenAI each script the consent a speaker records before a clone. Compare the wording, then write your own release for a Sume voice.
- Voxtral TTS 2-minute native limit vs Sume TTS 1200-second cap
Mistral says Voxtral TTS natively generates up to two minutes and the API handles longer. Sume TTS fails audio over 1200 seconds with tts_duration_exceeded.
- Walmart bans seller logos, Coupang wants yours: one prompt per site
Walmart's image guide bars seller logos; Coupang's main-image page says to show yours clearly. Keep a separate prompt per marketplace in a Sume batch.
Written by Sume