# MiniMax H3 Recast Canonical page: https://fuser.studio/models/minimax-h3-recast MiniMax H3 Recast in Fuser swaps up to four people in a 5-30 second clip for your reference photos, keeping the motion, camera, cuts and original sound. ## Specs | Spec | Value | | --- | --- | | Inputs | Video, People (up to 4 images), Prompt (text) | | Output | video | | Resolution | 768p, 1080p | | Cost | 413–37,196 credits per run | Drop new faces into a finished clip while the original timing, camera moves, cuts and soundtrack play on untouched Creator: MiniMax [Recast the clip](https://app.fuser.studio) ## How MiniMax H3 Recast works You bring the performance and the faces. Order your photos to match who is where on screen, nudge the mapping with a short prompt only if you need to, and let the clip keep its own length and sound. ### Load your source clip Add a 5 to 30 second video with the right performance and final sound, with no single shot running past 15 seconds. ### Line up your reference photos Add one clear photo per new person, up to four, ordered to match the screen positions from left to right. ### Test, then render Add an optional prompt to remap who becomes whom, check identity at 768p, then render the final recast at 1080p. ## What MiniMax H3 Recast is good at A recast tool, not a generator: it holds on to everything the footage already does well and changes only who is in it. ### Cast up to four, left to right Supply up to four photos and they replace the main people from left to right in every shot they appear in, with identity tracked across cuts. Add a short prompt to override the order. ### Performance stays, people change The source clip is the performance. Head turns, timing, expressions, lighting and lens movement carry over while faces and outfits follow your reference photos. ### Cuts and sound carried over Edit rhythm and the original soundtrack come through intact, so a recast is a drop-in replacement for the finished clip, not a rebuild. ### 768p test, 1080p final Check casting and identity on a short 768p pass, then commit to 1080p once the faces hold. Output length matches the source, so cost follows clip duration. ## Made with MiniMax H3 Recast Each clip starts from a different source performance: a music-video close-up, a barista ad, a fencing duel, a skate launch and a dawn fish-market scene, with the people swapped and the motion, cuts and sound inherited. ### A performance reused with a brand-new lead ### One spot, a personalized cast ### Two-person choreography, recast left to right ### A viral skit, new face ### Naturalistic and grounded, like the source ## What people build with MiniMax H3 Recast Directors, ad teams, short-form creators and fashion and e-commerce crews reach for it when the take is right and only the casting is wrong. ### Film directors swapping stand-ins Shot the scene with a stand-in? Replace them with final cast while blocking, lens movement, timing and sound stay exactly as filmed. ### Advertisers making personalized spots Produce one ad once, then recast it for each market or audience with up to four new people per version. Edit and soundtrack stay identical. ### Short-form creators reusing a skit Keep a performance that already works, such as a dance, a gag or a reaction, and put new characters into it without reshooting. ### Fashion teams recasting lookbooks Use a single runway or turntable walk and recast it across different models. With no prompt, hair and outfit tend to follow the reference photo, even a cropped headshot (our test, 23 October 2026, one run). To keep the source styling, prompt it to replace only the face. ### E-commerce teams localizing demos Take one product demo clip and recast the presenter for each regional storefront, keeping the demo's script, timing and sound intact. ## MiniMax H3 Recast questions ### What does MiniMax H3 Recast actually do? It replaces the people in an existing video with the people in your reference photos. The source motion, camera work, cuts and audio are kept, so it's a casting swap on finished footage, not a new generation. By default, photos replace the main people from left to right in every shot they appear in. ### How long can my source video be? Between 5 and 30 seconds, and no single shot may run longer than 15 seconds. Split long takes first. The recast keeps the length of the source. ### How many people can I recast, and in what order? Up to four, one photo per person. Photos replace the main people from left to right by default, so order them to match screen positions. Add a prompt such as 'the woman on the right becomes person 1' to override the order. ### What kind of reference photos work best? One sharp, front-facing, evenly lit photo per person. With no prompt, Recast tends to bring in the photo's hair and clothes, even from a cropped headshot: in our test on 23 October 2026 (one run, 768p), a forehead-to-chin photo still turned the source's silver hair black and her rust cardigan sage green. To swap only the face, add a prompt such as "Replace only the face of the woman with the face from the photo. Keep her hair and clothes exactly as they are." In the same test, a prompt along those lines kept the hair and outfit and swapped only the face ([see the face swap walkthrough](https://fuser.studio/articles/ai-video-face-swap)). Avoid sunglasses, hats, hands over the face, group photos, and low-resolution or heavily filtered portraits. ### Should I use 768p or 1080p? Test at 768p, then render the final at 1080p. A short 768p pass confirms the casting and identity before you pay for the higher resolution, and cost scales with clip duration. ### When should I not use it? Skip it when you need new motion, new camera moves or a changed environment, since the source dictates all of that. It also struggles with busy crowds, small or distant faces, and fast action or occlusion after hard cuts. Identity can drift back toward the original actor in long shots or profile turns. ## Try MiniMax H3 Recast on Fuser Drop new faces into a finished clip while the original timing, camera moves, cuts and soundtrack play on untouched [Recast the clip](https://app.fuser.studio) ## Related models - [MiniMax H3 Reframe](https://fuser.studio/models/minimax-h3-reframe.md) - [MiniMax H3](https://fuser.studio/models/minimax-h3.md) - [Hailuo 2.3](https://fuser.studio/models/hailuo-2-3.md) - [Hailuo 02](https://fuser.studio/models/hailuo-02.md) - [Kling 4.0](https://fuser.studio/models/kling-4-0.md) - [Kling 3.0 Video](https://fuser.studio/models/kling-3-0-video.md) Node reference: [MiniMax H3 Recast docs](https://docs.fuser.studio/docs/nodes/video/minimax-h3-recast.md) ## More articles - [AI Video Face Swap: Recast and Kling O3 Edit Tested on One Clip](https://fuser.studio/articles/ai-video-face-swap.md) - [How to Change Video Aspect Ratio With AI: 16:9 to 9:16 Without Cropping](https://fuser.studio/articles/change-video-aspect-ratio-ai.md) - [Jean Phil AI Character: How to Make Your Own (and Make It Dance)](https://fuser.studio/articles/jean-phil-ai-character.md)