---
name: video-edit
description: "Route edits of a video you already have and new camera or style shots on MiniMax H3 Max tools (2026-11-03): reframe to another aspect ratio, relight from a lighting sphere, recast the people from photos, insert a new scene, extend, a camera move from one picture, and five looks. Inputs, limits and checks. Only Turbo Extend is measured so far."
---

These creative references describe external tools as listed on 3 November 2026. Use your own supported account and tools; this download installs no automation. Apart from Turbo Extend, which the AI Talking Avatar guide measures on talking clips, none of these tools has been measured in our experiments yet: for those, the guide gives inputs and limits, not tested settings. It is not a claim of current FlashBoards availability.

# video-edit · edit a video with MiniMax H3 Max

Work on your own video or picture. Say which tools are not measured (all but Turbo Extend), run one take, check it and show it before a batch. A new shot of your character from references is a different job (`animate`); a bigger or sharper copy of a finished cut is `upscale-video-final`.

## Before you start

- **Trim the source first.** Reframe, relight and recast process every second of it.
- Pictures: a static PNG, JPEG or WebP, 256–8192 px a side, aspect 0.4–2.5. Pad a smaller picture; do not scale it up.
- File sizes: a source video up to 50 MB, a picture up to 10 MB, audio up to 15 MB.

## Classify

| The job | Tool | Inputs | Limits |
| --- | --- | --- | --- |
| Another aspect ratio (a 16:9 clip to 9:16), the edges painted in | MiniMax H3 Max Reframe | `video_url`; `aspect_ratio` 16:9 · 9:16 (default) · 1:1 · 4:3 · 3:4 · 21:9 · 9:21; optional `prompt` | source up to 15 s; 480p or 768p (default) |
| New light on the same shot | MiniMax H3 Max Relight | `video_url`; `reference_image_url` = a lighting sphere (a ball lit the way the shot should be); no prompt; `aspect_ratio` adaptive (default) or fixed | source 2.33–15 s; 480P · 768P (default) · 1080P · 2K, the last two refined from 768P |
| Other people in the same shot | MiniMax H3 Max Recast | `video_url`; `reference_image_urls` 1–4 photos of the new people; optional `prompt` (up to 2,000 characters); `enhance_realism: true` for a more photographic look | source 5–30 s; 768P or 1080P (default) |
| A new scene inside the clip, then back to the clip | MiniMax H3 Max Insert Video | `video_url`; `prompt`; `start_time` (where the new scene starts) before `resume_time` (where the source resumes), both inside the source; `duration` 5–13 s; `color_match` (on) | source 1.625–60 s, aspect 0.4–2.5; `start_time` at least 1.625 s; 480p or 768p (default) |
| More of the same clip | MiniMax H3 Max Turbo Extend, or MiniMax H3 Max Extend | `video_url`; `prompt`; up to 3 `reference_audio_urls`, 2–15 s each and 15 s combined; `duration` 1–15 s; `output: "continuation"` | source 1.625–60 s, aspect 0.4–2.5; 480P · 768P · 1080P · 2K |
| A camera move from one picture | MiniMax H3 Max Camera Controls | `image_url` (the first frame); `free_camera_preset` **or** `camera_trajectory`; optional `prompt`; `prompt_expansion_mode` (balanced by default) | `duration` 1–15 s (the output may run about 0.7 s longer); 480P · 768P (default) · 1080P |
| A shot in a look: VHS, hand-drawn pencil, 1970s painted cartoon, 16-bit pixel art, retro low poly | MiniMax H3 Max Styles | `prompt`; optional first frame `image_url`; `aspect_ratio` for text-only runs (16:9 default); VHS: `damage_level` light · medium (default) · heavy | 5–15 whole seconds; 768P only |

## Notes per tool

- **Extend.** Turbo Extend first (measured on talking clips in the AI Talking Avatar guide); H3 Max Extend, new on 3 November 2026 and not measured, when its continuation is not good enough. Send `output: "continuation"`, the default of both: it returns only the new footage after a 1.625 s repeat of the source's end, which you trim off before joining. `"extended"` returns the whole clip re-generated and softer. Talking clips in pieces: `avatar-long-speech`.
- **Camera controls.** Presets: `dolly_in`, `dolly_out`, `truck_left`, `truck_right`, `pedestal_up`, `pedestal_down`, `pan_left`, `pan_right`, `tilt_up`, `tilt_down`, `roll_left`, `roll_right`, `s_curve_forward`. A path is `camera_trajectory`: 2–12 keyframes `{time, azimuth, elevation, distance}`, with `time` 0–1 across the clip, `elevation` −90 to 90, `distance` above 0. Send one of the two: with both, the preset wins and the path is ignored. For your own clip, take a frame from it and use it as `image_url`.
- **Styles** make new footage from a prompt and an optional first frame; they do not repaint an existing video. To put a shot of your clip into a look, give its frame as `image_url`; the frame sets the aspect ratio.
- **Recast** keeps the motion, camera, cuts and sound of the source and swaps the main people for the photos. Use your own characters or people who agreed; never a real person's face without their consent.
- **Insert** keeps the source before `start_time` and from `resume_time` on; only the new scene is generated. Reference images or videos for the new scene are not offered.
- **Reframe and relight** keep the timing and the sound of the source.

## Not covered

H3 Max lip-sync, H3 Max 3D-to-video and the H3 Max director are outside this guide. Lips to a given voice: `animate-audio-ref`.

## Check

Put each take next to the source and ask: reframe, "Is anything stretched or cut at the new edges?"; relight, "Where does the light come from?"; recast, "Are these the people of the photos in every shot?". Compare the cuts against the source for insert and recast, and a transcript where the source had speech.

Read: `animate`, `avatar-long-speech`, `upscale`, `upscale-video-final`, `animate-audio-ref`. Use your own approved media.