Higgsfield face swap and Genjutsu, explained by a team that uses it
What Higgsfield’s Genjutsu motion transfer does, what it needs from you, what it costs to run, and when to use Higgsfield directly instead of an app built on it.
vtrends runs on Higgsfield’s Genjutsu Light model, so we have used it a lot. This is what we learned building on it. We have no partnership with Higgsfield beyond paying for the API like anyone else.
What Higgsfield is
Higgsfield is an AI video platform. It hosts its own models and many third-party ones under one account, with tools for generating clips, images, camera moves and swaps. “Higgsfield face swap” usually refers to its swap and motion-transfer tools, of which Genjutsu is the one that changes the person in a real video.
How Genjutsu works
Genjutsu is motion transfer. You give it a source video and between one and eight reference images, plus an optional text prompt. It keeps the motion, timing and camera of the video and regenerates the people from the references. That is why it changes skin tone, hair and clothes, not just the face. We explain why that matters in AI body swap video.
- Source length. 4 to 15 seconds. Shorter is rejected; longer must be trimmed first.
- References. Up to eight images: faces, outfits, places.
- Resolution. We render at 720p; higher resolutions cost more.
- Sound. Plan to handle audio yourself. Our pipeline adds a watermark after the render, so we put the original sound back on as the last step.
- Time. Renders take minutes, not seconds. Ours usually finish in 10 to 15 minutes including our own steps.
What we learned using it
- Crop the face photo. Given a full-length photo, the model also copies the outfit you are wearing in it. We crop face photos to head and shoulders so the outfit comes from the outfit photo, or from the original video.
- Order the references. With several faces, it maps by prominence: the first face to the most prominent person. We tell it so in the prompt and order uploads to match. See multiple face swap.
- The prompt helps, a bit. A short prompt explaining which image is which improves mapping. It doesn’t override what the references show.
- Short clips win. The 15-second ceiling suits TikTok trends; it doesn’t suit long videos.
What it costs
At the time of writing, Genjutsu Light on Higgsfield’s API is priced per second at each resolution, and the billable seconds include both the source and the output. Higgsfield’s consumer plans use credits instead. Prices move, so check Higgsfield’s pricing page rather than trusting any blog, this one included.
Use Higgsfield directly, or an app built on it?
| Use Higgsfield directly if… | Use an app like vtrends if… |
|---|---|
| You want many models and settings | You want one result without learning a tool |
| You will trim, crop and re-add sound yourself | You want the trim, face crop and sound handled |
| You make videos often enough for a plan | You want to pay per video, no account |
| You want higher resolutions | 720p is enough for a phone |
If you just want to be in a trend, pick a clip and add your photo. If you want to compare it with the alternatives, see the best face swap video apps in 2026.
Questions
Is Higgsfield AI free?
Higgsfield has had free credits and trials on its consumer plans, but its video models are mainly paid. Check its pricing page for the current offer.
Does the finished video keep the sound?
On vtrends, yes: we put the original sound back on as the last step, after the watermark. If you use the API yourself, check the audio of every output before you post it.
How many faces can Genjutsu swap?
It takes up to eight reference images in total. vtrends uses up to four of them for faces and the rest for outfits and locations.