Painting Change Map guides video generation of painting processes
Original title: Painting Change Map as guidance for generating painting processes
The method encodes localized structural change across 1,000 timesteps for forward painting and reverse unpainting control.
Researchers introduce the Painting Change Map, a spatio-temporal control signal for video models that generate painting and unpainting processes. It tracks normalized structural changes across a 1,000-timestep continuum and encodes localized visual differences in RGB. The authors say it beats existing control baselines and is competitive with state-of-the-art painting methods, with larger gains under oracle conditions. Code is at github.com/Markus-Pobitzer/painting-change-map.
- Control covers where an artist acts and when those actions occur.
- A 1,000-timestep continuum is used to capture stroke-level change.
- Oracle conditions are reported to beat all baselines by a wide margin.
- Code is publicly available on GitHub.
From the original article
1 min read at huggingface.co
Digital art generation has undergone a massive paradigm shift, moving from static image synthesis to dynamic video generation that can mimic the human painting process. While existing models can synthesize visually stunning time-lapses of paintings being created or "unpainted," they largely treat the creative process as a black box.
By Markus Pobitzer, Chang Liu, Niccolò Biondi · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.