Training-free match-cut synthesis: run two text-to-video generations from one shared noisy latent for the first K denoising steps, then let each follow its own prompt, so the outputs align in structure and motion.
Diffusion illusions: Hiding images in plain sight
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
MatchDiffusion: Training-free Generation of Match-cuts
Training-free match-cut synthesis: run two text-to-video generations from one shared noisy latent for the first K denoising steps, then let each follow its own prompt, so the outputs align in structure and motion.