Baselight 7 – Testing Auto-Resize Matching (with Machine Learning)

April 16, 2026

Rebuild broken conforms with Baselight 7's machine learning transform matching - from plain resizes to keyframed and stabilised shots.


Quick Summary

Baselight 7’s machine learning transform matching automatically rebuilds resizes from a ganged offline reference clip, handling keyframed moves, speed-affected shots, and stabilised plates through the Transform and Perspective operators – a conform rescue for XML or AAF deliveries that arrive without usable sizing data.

A real-world test on keyframes, speed effects, stabilised shots, and the one case that still belongs with VFX

You have been handed an offline with a lot of resizing in it, and an XML or AAF that either drops the transform information entirely or interprets it wrongly. Manually rebuilding those resizes from the reference burns hours, and it burns the kind of visual attention you really want to reserve for the actual grade.

Baselight 7 introduces machine learning transform matching. You gang a reference cursor to your conform, drop a Transform or Perspective operator at the bottom of the stack, and ask Baselight to “match all frames”. In this Insight, I put the feature through a gauntlet of real-world shots – plain resizes, keyframed moves, a speed effect, stabilised plates, and a picture-in-picture composite.

The results are excellent on most of the material. I also show you where the feature hits its limit, why that particular shot belongs with a compositor rather than a colourist, and a bonus technique for rebuilding a better camera stabilise from area-track data when you want to keep some of the original plate motion.


“Manually matching the resizes from a conform takes quite a lot of time, quite a lot of patience. It utilises all of your visual capabilities that you should be reserving for actually grading.”

Kali Bateman, Colourist

Key Takeaways

By the end of this Insight, you should understand how to:

  • Pull an XML or AAF into a Baselight 7 scene and link it to your rushes without applying the (potentially broken) image transforms.
  • Gang cursors between two scenes to confidence-check a conform against an offline using side-by-side and difference displays.
  • Use the Transform and Perspective operators to match a resize from a ganged reference cursor, including keyframed and speed-affected shots.
  • Recognise when a resize belongs in the grade and when it should be handed off to a compositor or motion graphics artist.
  • Composite a picture-in-picture in Baselight 7 using the machine learning transform match as a starting point.
  • Rebuild a better camera stabilise from area-track data, with filtering to preserve some of the original motion.


Member Content

Sorry... the rest of this content is for members only. You'll need to login or Join Now to continue (your career will thank you!).

Need more information about our memberships? Click to learn more.

Get Answers, Join Now!
Member Login

Members, enter your details here. You will be returned to this page.


Is your career calling out for help?

Answer the call with a Mixing Light Membership. Gain client-tested tips, workflows, and add new skillsets from our pro Contributors!


JOIN NOW! You don't have to do this alone!
Loading...