Hailuo Recast: Put Yourself in Any Video With a Single Photo

October 7, 2026 · 6 min read

Hailuo Recast is the video-to-video half of MiniMax's Hailuo H3 Max — the #1 ranked image-to-video lineage — and it does one thing better than anything else available right now: it takes a clip of someone moving and re-shoots it with you as the performer, from nothing but a photo of your face. No filming, no green screen, no editing timeline. The motion, the camera work, the cuts and the original sound all stay; only the person changes. It is the engine behind the recast videos filling your feed — and it is the default recast backend inside Hermosso's Motion Transfer, where the whole job is three steps: photo, clip, generate.

What Hailuo Recast actually is

Recasting is different from face-swap filters, and the difference matters. A filter pastes your features onto the original performer's face. Hailuo H3 Max Recast rebuilds the scene: the model understands the clip — the person, their movement, the camera, every cut — and regenerates it with the person from your reference photo instead. Your body proportions, your look, in motion that was never filmed of you.

The details that make it the current quality bar:

• Tracked across shots. Most earlier tools fall apart at the first cut. Hailuo Recast follows each person through every shot they appear in (up to 15 seconds per shot), so multi-angle clips — the ones that actually go viral — survive intact.
• Multi-person in one pass. One photo per person, and the model maps them onto the people in the clip from left to right. A group scene with your friends works the same way a solo clip does.
• The original sound carries over. Music, dialogue, ambient audio — the recast keeps the clip's soundtrack, which is half of what makes these videos land.
• Real resolution. 768P by default, 1080P at the pro tier — the previous generation of recast models topped out at 720p.

It took the crown from Genjutsu

If you read our explainer on the recast trend, you met Genjutsu — Higgsfield's model, the engine that made identity-through-motion look real instead of rubbery and kicked this whole format off. Genjutsu is still good, and it still runs the "Change the video" (restyle) half of our Motion Transfer tool. But for recasting people, Hailuo H3 Max Recast is simply the better model: identity holds harder through fast motion and cuts, multi-person scenes work in one pass instead of one at a time, the sound survives, and the output is sharper. When we tested the same clip through both, the choice was obvious enough that Hailuo is now the default engine for every recast on Hermosso.

How to put yourself in any video (three steps)

1. Add a photo per person. One clear, front-facing photo each — the first photo replaces the leftmost person in the clip, the second the next, up to four people.
2. Add the clip. Anything 5–30 seconds with clearly visible people: a dance, a movie scene, a sports highlight, a meme. Multi-shot clips work — no single shot over 15 seconds.
3. Transfer. Draft at 768P to see the result fast, render the winner at 1080P. The price is per second of clip and shown before you run — a 5-second recast at 768P costs 2,250 credits, a 10-second one at 1080P costs 7,000.

Two tips from our testing: turn on Realism boost for close-ups (it adds the model's skin-texture and blending pass), and if the result drifts, the fix is usually a sharper, better-lit photo rather than a different clip. That's the whole workflow — open Motion Transfer and the first recast is minutes away.

The formats that are going viral

The winning recipes, if your feed hasn't already suggested them: dance challenges performed by people who demonstrably can't dance; group recasts of iconic scenes — one photo per friend, everyone in the clip at once; sports and movie moments re-starred by the group chat; and the eternal meme formats, which recasting refreshes faster than they can burn out. The trend's mechanics — and the consent line that keeps it fun — are in our recast trend explainer: your face, your friends' faces with their blessing, fictional characters. Label it AI when you post; the transparency is part of the bit.

Where it still struggles

An honest list. The clip needs clearly visible people — abstract footage or a tiny distant figure comes back with "no people found" (a refund, not a charge, on Hermosso). Heavy occlusion — the performer passing behind objects constantly — can break identity for a frame or two. The 5–30 second window with a 15-second ceiling per shot rules out full songs and long scenes. And likeness is strongest from a sharp, front-facing photo; a grainy side profile gives the model little to hold onto. Within those lines, this is the best recast tool we have tested — and we test all of them.