Writing a multi-shot video as one prompt means describing every shot in the same paragraph. It gets long, it’s hard to tell who is in which shot, and changing one shot often means rewriting the whole thing.
The Multi-Shot Editor turns that paragraph into a timeline, with one card per shot. Each card has its own duration, its own reference image and its own camera motion. You can change one shot without touching the others, and switch video model without rewriting anything.
This tutorial covers the whole process: preparing consistent reference images, cleaning them up, building the sequence, then rendering it with several models.
Preparing your reference images
The video is only as consistent as your reference images, so the work starts in Image Gen.
Build the concept with snippets
Compose your character, outfit, pose, and lighting as snippets rather than as one sentence. Swapping an outfit or a lighting preset is then one click, and nothing else in the prompt moves.
Generate a contact sheet
Turn on contact sheet variations, set the grid to 2x2 and the mode to Angles + Poses. One generation gives you four camera angles and poses of the same subject.
Split out the tiles you want
Open the sheet, enter Split mode, select the shots worth keeping, and extract. Each tile is saved as its own asset in your gallery.
Cleaning up the extracted tiles
Tiles from a contact sheet have a lower resolution than a full generation, so upscale them before making the video.
Drag them into Upscale and process the batch at 2x. On wide shots, where the face is small and loses detail, use the Small Faces Fix template in Image Edit: the wide shot as the source, and a detailed close-up as the face reference. It rebuilds the face from the reference instead of guessing it.
Building the sequence
Once your references are ready, switch to Video. The Multi-Shot Editor is where you link each reference to a shot and add its camera motion.
Pick a model that supports multi-shot
Every video model with multi-shot support in Hylope works, like MiniMax H3, Kling Omni v3 or Wan 3.0, the three used in the video. The Multi-Shot badge appears on the prompt bar when the active model supports it.
Set the durations
Open the builder and choose a total length, six seconds for example, then split it between the shots on the timeline.
Link a reference to each shot
Drag an image onto a shot card, or type @ in the card to pick from your active references.
Add camera motion
Use the camera motion snippets instead of describing the movement: Long Track into Wrap, Cinematic Orbit, Defocus Slow Zoom. The snippet is added to that shot’s prompt.
Improving the prompt and trying other models
Improve Prompt reads everything: the shot cards, the snippets and the linked references. It rewrites the prompts into one clear structure, written for video.
Each model expects its own prompt format for multi-shot. The editor keeps the shots separate and writes the right format for the active model, so you can render the same sequence on any model that supports multi-shot without rewriting anything. The video tries MiniMax H3, Kling Omni v3 and Wan 3.0. It’s the quickest way to see which model handles your idea best.