Adobe stealth-drops TransPixar to generate transparent AI video
Adobe Research and HKUST just open-sourced TransPixar, a new model that solves a major VFX headache: generating videos with built-in alpha channels. Instead of solid backgrounds, it generates transparent RGBA elements like smoke and portals ready for compositing. It is an early-stage research project, but the code and interactive demo are already live.
TransPixar is an open-source video generation model from Adobe Research and Hong Kong University of Science and Technology. It does one specific job that VFX artists have been begging for: generating video with a native transparent alpha channel.
Anyone who has tried using existing AI video tools in production knows the frustration. Every model spits out fully opaque, solid backgrounds. If you just need a puff of smoke or a glowing orb, you are stuck spending hours rotoscoping edges that still end up looking crunchy.
What it does
TransPixar cuts out the cleanup work entirely by rendering RGBA video directly from simple text prompts. Because the alpha channel is built right into the output, elements composite straight onto live-action plates without matte lines.
In the official demos, the model successfully pulls off tricky semi-transparent physics. Examples include rising smoke plumes, an electrical sci-fi portal, and a swirling asteroid belt.
Beyond generating clips from text, TransPixar can also take a standard still image and animate it with transparent floating effects layered on top.
How it works and availability
As Adobe Research project lead Yijun Li noted in the research paper, alpha channels are non-negotiable for VFX compositing. However, training models to produce RGBA video has historically hit a wall due to the lack of transparent video datasets.
TransPixar tackles this by adapting diffusion architectures to predict transparency and RGB values simultaneously. This allows delicate details, like ethereal reflections and dissipating smoke edges, to blend naturally across frames.
Adobe and HKUST have made the source code available on GitHub for researchers and engineers to build upon. There is also an interactive demo hosted on Hugging Face if you want to test prompt results yourself.
Keep in mind that this is still an early-stage research prototype. It has not been packaged into a production-ready After Effects or Cinema 4D plugin yet, and resolutions remain experimental.
Why it matters
For indie VFX artists, solo game devs, and motion designers, TransPixar is worth keeping on your radar right now. Testing the demo shows how much tedious rotoscoping work could vanish from post-production suites in the near future.
While most AI video labs are chasing cinematic trailers, Adobe is tackling actual pipeline bottlenecks. Moving from flat pixels to compositable transparent layers is exactly what will make generative video viable in professional studios.
Sources (1)
- creativebloq.com(opens in a new tab)Originalcreativebloq.com