Higgsfield Genjutsu: changing wardrobe in post with no roto, no tracking, no reshoot

I shot Pau walking with flowers, Open Gate on the FX5, wearing a brown jacket. Now the jacket is black — fabric texture and the hair falling over it both respected — and there was no roto, no tracking, no green screen: I asked for it in writing. That’s Higgsfield Genjutsu, and it’s the first generative video model that fits what I actually need as a cinematographer: it doesn’t generate a new shot for me, it improves the one I already filmed.

Pau’s original shot and the version with the jacket swapped, side by side on my timeline.

The facts

• Genjutsu is video-to-video: upload your take, describe the change in plain language (wardrobe, weather, a vehicle, the background) and the model keeps camera movement, light, grain, lens character and cut points. It touches only what you asked for.
• Outputs at 480, 720 and 1080p. A wardrobe change at 1080 took me around 10 minutes with the model under heavy demand.
• It does motion transfer: swap the whole character or location and the action stays identical, frame for frame.
• Mistakes get fixed with the same prompt. One comment said “it looks like it only rains in front of the car” — you ask for “rain along the whole road” and you’re done.
• I showed it live at Semana AMC at Estudios Churubusco. There was backlash. Keep reading.

Four tests on my own footage

Pau’s jacket. Open Gate shot on the FX5, sunlight, yellow flowers. I swapped the jacket for a black one I generated as a reference, then clouded the sky, added rain and changed the color of the flowers. A few minutes in total, and the fabric texture and the hair over the jacket hold up in a medium shot. Ask a post house for this and they’ll struggle to deliver it at this quality for that money.

Formula 4 to Formula 1. A video I filmed with Formula 4 cars, one I really liked — but we wanted Formula 1. We designed the cars in Higgsfield and told Genjutsu to change only the vehicles. The video was respected completely; it transferred the cars. Nothing else.

The F4 cars I shot, turned into F1. Camera, framing and cut: untouched.

Joaquín and the pizza. We shot it at the cafeteria of INDIE, the film school, on the FX5. We changed wardrobe and hair while keeping his expression, the camera move and the lighting. Then the real problem: we had our “Japanese” actor dressed, but the cafeteria didn’t look Japanese. We built the set in GPT Image 2 inside Higgsfield and dropped it in. The light direction holds more or less; the space and the wardrobe, completely.

Tattoos in my film. One character’s tattoos go on in post because in the story they move slightly. They’re designed afterwards, not on real skin, and they respond to the exact position of the muscles, the skin and the shadows. I can redesign them and try variants without calling the actor back.

What this changes in production

The first thing they tell you in film school is: don’t write what you can’t produce. Today you can write whatever you want. A Formula 1 race? You do it with very little money. And yes, AI isn’t free — but neither was getting two Formula 1 cars. That’s where it becomes accessible.

The read for anyone producing here is direct: high-end VFX in the hands of low-budget productions. You no longer have to find, pay for and build a space. You focus on your actors, light them well, pick the lens and the frame — the part we love — and build the rest over here. Your period film noir with cars, gunshots and 1940s New York streets no longer depends on a budget that doesn’t exist in LATAM.

Same actor, same expression, same camera: different wardrobe and a different cafeteria.

About the Semana AMC backlash

There were very nervous people at the talk, and I get it. My read is that most of them don’t know these specific tools: they’ve seen rehashes and weird soap-opera clips on social media and concluded “AI looks bad.” Look at Pau’s jacket and tell me why it looks bad.

Where does the production designer go? I’d treat it like an animated film. I’ve shot two animated features as a DP and I didn’t show up with cameras or lights: I designed light sources, contrast ratios, colors, all virtual. There are concept artists, production designers, a full crew working with tools that resemble a set’s but live in software. That keeps happening, and it will.

My personal opinion — Daniel Blanco A.M.C.

I’m genuinely impressed. Two years ago none of this existed, not even close, and changing a wardrobe took weeks; today it takes ten minutes. But what excites me isn’t today’s model, it’s what’s coming in two or three months. There’s a lot of discussion and very little action: I’m not seeing films where this is applied end to end without it being annoying to watch. I’m using it to my advantage in my own films, as a VFX tool inside my workflow, not to generate everything from scratch. What’s going to matter now is creativity.

Try it on your own footage

Don’t test it with the brand’s demo: upload a take of yours, ask for one small change — the color of a garment, the weather — and judge on your own monitor. With that result in hand, the conversation in the comments becomes a different one.

Images: frames from Daniel Blanco A.M.C.’s video.

Original source: Higgsfield Genjutsu — video-to-video, on YouTube

Extra for you: if you’re experimenting with AI for video, my Higgsfield link has discounts and promos to get started — and using it supports this content.