Decart Lucy Restyle 2: 30 minutes, 200 MB, vs Sume Genjutsu

Decart Lucy Restyle 2 takes up to 30 minutes and 200 MB at 720p. Sume's restyle path, Genjutsu motion transfer, takes 4-30 seconds and 1-8 reference images.

5 min readSume
All posts

Decart's Lucy Restyle 2 restyles a whole video from a text description, with a stated limit of up to 30 minutes and 200 MB at 720p. Sume's nearest route is Genjutsu motion transfer, which is a different mechanism: you pass one source video plus 1 to 8 reference images, and the clip must be 4 to 30 seconds. So the limit that matters most is clip length: 30 minutes versus 30 seconds.

Decart's figures come from its Video Restyling page and the Lucy Restyle 2 API reference, read 2026-10-02. Sume's come from Video Router and Video generation.

What are Decart's Lucy Restyle 2 limits?

The restyling page lists input up to 30 minutes, a maximum of 200 MB, and MP4, WebM, or QuickTime. Output defaults to 720p (720x1280 portrait or 1280x720 landscape). The required inputs are the video and a text description of the style. The API reference adds an optional reference_image, an optional seed from 0 to 4,294,967,295, a resolution that only offers 720p, and enhance_prompt, which defaults to true. The endpoint path is /v1/jobs/lucy-restyle-2.

Decart Lucy Restyle 2 versus Sume Genjutsu (read 2026-10-02)
QuestionDecart Lucy Restyle 2Sume Genjutsu
Source lengthUp to 30 minutes4 to 30 seconds
Style inputText description (reference image optional)1 to 8 reference images
Resolution720p480p or 720p
Output lengthNot stated on the pageSet to the inspected source length

What does Sume's Genjutsu take instead?

Genjutsu is listed in the Video Router as Motion Transfer. The catalog says one source video and 1 to 8 reference images are required, there is no text-only generation, the resolution is 480p by default, and the duration you send must be the inspected source video's length (4 to 30 seconds). It does not accept aspect_ratio, frame_images, generate_audio, bitrate_mode, or audio references.

That last point matters for the query. Genjutsu does not take a free-text style like "make it a watercolor". It takes images that show the target look. If you want a pure text restyle on Sume, the Omni edit mode on the same router is the route, with the prompt describing the change.

curl -X POST https://api.sume.com/v1/video-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: genjutsu-001" \
  -d '{
    "model": "higgsfield-genjutsu",
    "prompt": "Apply the look of the reference images",
    "video_url": "https://example.com/source.mp4",
    "reference_image_urls": ["https://example.com/style-1.png"],
    "duration": 12,
    "mode": "async"
  }'

How do I restyle something longer than 30 seconds on Sume?

Probe it with video inspect, cut it into ranges of 30 seconds or less with video trim, run each range, and join the clips afterwards. Cut at natural shot boundaries so the look does not shift mid-shot. This is more work than Decart's single call, and the docs do not promise that separate ranges will match each other, so spot-check the seams.

Genjutsu is also only listed in the catalog when its provider is configured, so call GET /v1/video-router/models first and confirm it is there.

Can the two be compared on price?

Not from the pages read for this post. Decart's restyling page and API reference do not state a price for Lucy Restyle 2, and its editing page lists $0.04 per second only for Lucy 2.5, Lucy 2.1, and the legacy Lucy Clip. Sume prices Genjutsu per second by resolution at list times 1.25, and the live rate is in the catalog response. Without a Decart number for Restyle 2 there is nothing honest to divide.

So the decision today is about limits and control, not cost: long source and text prompt on one side, short source and reference images on the other. Re-read both pages when you are ready to buy, because neither vendor promises these figures are stable.

Which one should you pick for a long restyle?

For a single long source, Decart's published limits are the better fit, and its queue statuses are pending, processing, completed, and failed per its Python queue page. For short clips where you want reference-image control and a Sume job, use Genjutsu. This post makes no quality claim about either; neither vendor page publishes one.

Sources

Related posts

More in Comparisons

All Comparisons posts

Written by Sume