Higgsfield Genjutsu

Higgsfield Genjutsu is a video tool that keeps a clip you already filmed and changes the person, the place, or one object in the shot.

The performance stays. The room can go.

Genjutsu is Higgsfield's name for this. Before the name, the job is ordinary: you already have the take, and you want a different person in it, or a different room, without shooting again. It is not a box where you type a scene and hope a clip appears. The product page puts it in one line: the movement and camera stay as filmed, and everything performing that movement can be yours.

There are two buttons, and they do opposite things.

• Motion Transfer keeps the motion, the camera, and the timing, then rebuilds who is on screen and where they are from pictures you supply. Higgsfield's own demo of "keep only the motion" is labeled Original, New character, and New world. A second demo, "keep your character in every video," takes a phone clip and changes wardrobe, set, and grade. The three results on that demo are named Rooftop, Dreamland, and Park.
• Object Swap changes one thing you name, such as a person, a product, or the text in frame, and leaves the rest of the shot untouched. The page calls the comparison "same take, two realities": you drag between the original and the new version, the motion and the cuts stay in sync, and only the scene changes.

The picture on Higgsfield's Genjutsu post is that idea in one frame. The same running stride is shown twice, once through a cold tiled room and once down a gold hall. The jacket and the step match. The building does not.

What one generation accepts

The product blog, updated 2 September 2026, is where the limits are written down. The marketing page and the API catalog do not all agree, so the numbers below stay attached to the page they came from.

• On the product: one clip, 4 to 30 seconds. The blog says this works on footage you shot and on video that was itself generated.
• Up to 30 reference photos in that same generation, for a character, an outfit, a product, a location, or a style. The blog's example is cycling several outfits on one clip.
• A full prompt is optional. The blog says one of 30 or more presets can drive the change alone, and a short description is there when you want to be more specific. The product FAQ says you do not need video-editing experience.
• Output is up to 1080p. The product FAQ says you can publish the result on organic and paid channels, subject to Higgsfield's Terms of Use.

The API is a separate surface. Higgsfield's API catalog lists Genjutsu Motion Transfer and Genjutsu Object Swap as released on 17 September 2026, with input duration of 1 to 30 seconds, not the 4-second minimum on the product page. The Motion Transfer sample calls the model higgsfield/genjutsu/motion-transfer/v1.0 and sends a video URL, image URLs, and a resolution. The same catalog lists a third name, Genjutsu Restyle, with no price and no release date. The product page does not explain Restyle.

What a 15-second clip costs

Inside the app, both buttons use Higgsfield credits, and the generator shows the credit cost before you run it. Motion Transfer and Object Swap are priced the same. The blog's table, which it says was checked in September 2026 and may change with settings, prices a 15-second clip like this:

• 480p: 40 credits, about $2.00.
• 720p: 104 credits, about $5.20.
• 1080p: 144 credits, about $7.20.

Those dollar figures are five cents times the credit count. That is arithmetic on their table, not a separate rate card. The public pages do not list a free Genjutsu allowance.

The API price is different, and it is worth not blending the two. On the API playground, each second of input costs $0.318 at 480p, $0.681 at 720p, and $1.632 at 1080p. Duration is rounded up to the next whole second, and the page says those rates are before any customer discount. A 15-second clip at those rates is about $4.77, $10.22, and $24.48. The same 15 seconds on the blog's credit table is about $2.00, $5.20, and $7.20. Both numbers are Higgsfield's. They are not the same price.

The jobs they wrote down

The blog's use-case table is more specific than "make content." Each row names what changes and what the source clip keeps.

• Style swap: an anime-style clip becomes live action, or the reverse. Motion, timing, and camera work stay.
• Wardrobe: several outfit references on the same clip. Identity, pose, and movement stay.
• Ad variation: a different character, product, or setting for another market. The pacing and the structure of the original ad stay.
• Repurposing: a raw phone clip turned into a polished commercial. The original action and framing stay.
• A repeating character: who is on screen changes, and the motion underneath does not.

Who is behind it

Genjutsu is a feature of Higgsfield, not a separate company. Higgsfield's about page says the studio launched in 2025 and is based in San Francisco. Alex Mashrabov's own profile dates his CEO role from October 2023, so those two public dates do not match. His speaker bio, which the Milken Institute notes was supplied by him, says he is co-founder and CEO, that he previously co-founded AI Factory, the company behind Snapchat's Cameos and face filters, that Snap bought AI Factory in 2020, and that he then led generative AI at Snap.

The about page's own totals, which are Higgsfield's figures, are 25 million users, 850 million generations, about 6 million generations a day, 300 million videos, and about 2 million videos a day. It says the company is backed by Accel, Menlo Ventures, GFT Ventures, and AI Capital Partners, at a valuation of over $1.3 billion, and that it has shipped more than 200 releases since launch. Those numbers describe the whole studio. They are not a count of Genjutsu renders.

The Genjutsu post itself is dated 31 August 2026 and was modified on 2 September 2026. Higgsfield's LinkedIn introduction of the feature is also 2 September 2026. The API models follow on 17 September. The Genjutsu page shows a block of customer quotes, but they praise Higgsfield in general, and they are not signed with a name, so they are not quoted here.

What you still have to bring

Higgsfield's other video tools, including DOP and Cinema Studio, start from a description, a still, or a link. Genjutsu starts from a clip. If you do not have the performance, this tool has nothing to keep. That is the line that would be false if you described it as another text-to-video model.

In September 2026 the name was also searched because people were restaging a COLORS performance called Hotel Lobby: same dance, a different face. Higgsfield's own Genjutsu pages do not use that clip as the example. The example they do publish is the split run above, one stride in two buildings.