Introduction

Ditto is a new video recasting tool from Melius, the AI creative-canvas startup. It launched on November 11 (cheekily timed for 11/11 at 11:11), and its job is simple to say: take an existing clip and swap the people, props or background for an avatar, a photo or a new scene, while keeping every move, camera turn and beat of the original.

The big number is 30. One Ditto request can carry up to 30 replacements, and they're all applied together in a single generation, with one price and one output video. It's a closed, credit-based tool you can open in the Melius app today, and Melius says it's also available through its API and MCP server.

How Ditto Works

Melius says Ditto isn't just a wrapper around one video model. The team built its own processing pipeline that mixes generative models with classic video transformation software, aiming for swaps that are cheaper and sturdier.

In the app, it's three steps:

  • Pick a video. Upload your own clip, or start from a trending video in Ditto's template library. You can trim a longer clip down to the part you want.
  • Choose who and where. Pick exactly which person, thing or background to replace, then give Ditto a photo, one of your avatars or a curated Melius avatar.
  • Generate. Ditto renders the new clip with the original choreography, framing and timing.

You can swap a background on its own, or together with the people in the shot. The examples on the Ditto page include a new customer in a barber's chair, a street dance rebuilt in a blocky game style, and a phone call turned into a retro console cutscene.

The API, for Builders

The Ditto API docs show how it works under the hood. You upload an MP4 or MOV, then send an ordered list of replacements. Each one has a reference image or an avatar, a targetKind of character, prop or background, and an optional description of what to replace:

curl -X POST https://api.melius.com/api/v1/ditto \
  -H "Authorization: Bearer mel_..." \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: <saved UUID>" \
  -d '{
    "sourceAssetId": "<videoAssetId>",
    "replacements": [
      { "referenceAssetId": "<personImageId>", "targetKind": "character" },
      { "referenceAssetId": "<mugImageId>", "targetKind": "prop", "targetDescription": "the mug on the desk" }
    ],
    "maxCredits": 500
  }'

A few handy details from the docs:

  • maxCredits caps the cost, so a pricier render is refused instead of charged.
  • sourceRange lets you recast just part of a clip, in milliseconds.
  • audioMode keeps the original audio by default, or mutes it.
  • additionalInstructions adds direction that applies to every swap.
  • You poll the job as it moves through preparing, rendering and finalizing, and get a video URL when it's ready.

One catch: the API is in beta and rolling out gradually. Teams outside the beta get a DITTO_BETA_REQUIRED error and can join the waitlist from Ditto in the app.

Price and a Gentle Reminder

Melius is running 25% off Ditto through the end of the year as a launch deal. It hasn't posted a per-second rate for Ditto, since renders are priced in team credits.

Swap tools this good come with responsibility. Recast yourself, your own avatars, or people who've said yes, and label edits that could fool a viewer.

Conclusion

Character swaps used to mean one face at a time. Ditto goes after the whole scene at once, with up to 30 swaps in a single pass and the original motion intact. If you make trend remixes, ad variants or quick parody cuts, it's worth a spin while the launch discount lasts.

Original Source

—Aurelia ♡