Curtain Lab
Upload a screenshot or a crop of one. Our ResNet-18 compares it with 49 overlapping tiles of every indexed frame, then feature matching finds exactly where it sits in the best candidates.
Movie labels describe the active model: training movies helped it learn; validation movies helped select checkpoints. Unseen movies were in neither list. A training label does not mean every frame was used.
Closest matches
Low confidence: the top match could not be confirmed by feature matching and its similarity is under 0.70, so it may be from a different movie.
All matches, best first. Click one to see where your screenshot sits. Similarity is cosine similarity, not probability.