petermercell

petermercell / CutieRoto

Public

AI-tracked roto for Nuke: roto a few keyframes, propagate the mask across the clip. Native TensorRT + libtorch port of Cutie

10
0
100% credibility
Found May 26, 2026 at 10 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
C++
AI Summary

CutieRoto is a native plugin for Foundry Nuke that uses AI to propagate hand-drawn roto masks across video clips. Instead of rotoscoping every frame by hand, you draw the mask on just a few keyframes, and the AI fills in the rest. The results are cached to disk for smooth playback, and you can fine-tune the matte with simple controls like Gain and Gamma. It's a from-scratch port of the Cutie video object segmentation model, optimized to run directly inside Nuke with no external services needed.

How It Works

1
🎬 You have a long clip to roto

You need to track a complex shape across hundreds of frames, and the thought of hand-rotoing each one makes you wince.

2
🔌 You connect your footage and a Roto node

In Nuke, you wire your plate into input 0 and an animated Roto node into input 1 — just like any other node.

3
✏️ You draw the mask on a few keyframes

You animate the Roto by hand on just a handful of frames — maybe frames 1, 48, and 96. It takes only a few minutes.

4
🚀 You hit Process and watch it work

You type your keyframe numbers into the Keyframes box and press Process. The AI studies your keyframes, then fills in every frame between them automatically.

5
💾 Everything is saved to disk

Each frame's matte is written to a cache folder on your drive, so you can scrub back and forth instantly without waiting.

✨ You scrub through your clean matte

The mask follows your subject beautifully across the entire clip. You adjust the Gain or Gamma knobs to crisp up the edges, then send it downstream.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 10 to 10 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is CutieRoto?

CutieRoto is a native Nuke plugin (C++) that uses AI to propagate rotoscoping masks across a clip. You animate a roto on a handful of keyframes, hit Process, and the tool fills in every frame between them. Under the hood, it's a port of the Cutie video object segmentation model onto a hybrid TensorRT + libtorch pipeline. Four stages run on TensorRT for speed; one stage (the pixel fusion and transformer) runs on Nuke's bundled libtorch. The plugin is self-contained — TensorRT is statically linked, so artists just drop the `.so` into Nuke 17 and need only an NVIDIA driver.

Why is it gaining traction?

The pain CutieRoto solves is real: manual roto is tedious, and generic AI tools struggle with fine details. Cutie was trained specifically for video object segmentation — it tracks *this* mask, not *an* object. The port is also unusually well-documented: the README walks through every decision, the dead-ends, and the architecture choices. There's a pure-libtorch backend as a fallback if you don't want to deal with TensorRT engine builds. For a studio already on Nuke 17, the deployment story is clean — no extra installs, no CUDA toolkit needed on artist machines.

Who should use this?

VFX compositors and roto artists working in Nuke who need to track masks across long clips. If you're doing feathered edges or complex shapes that blur tracking data, this could save hours of manual work. Studios with Nuke 17 and NVIDIA GPUs can deploy it immediately with the pre-built libtorch backend; studios with CUDA expertise can build the faster TensorRT variant for better performance. It's not a general-purpose tool — you're expected to know what roto is and why you'd want to propagate it automatically.

Verdict

A promising niche tool with unusual documentation depth, but the credibility picture is thin: 10 stars and a 1.0% credibility score mean this is essentially one person's project with limited community validation. The codebase appears complete and well-structured, and the author clearly knows their domain, but there's no track record to lean on. Worth evaluating if roto speed is a bottleneck for your team — but treat it like production tooling: validate on your actual shots before committing.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.