For mature composers: the DAWN model for keys and hair relighting (and more).
British start-up PXLD has launched the basic spatial harmonization model DAWN (Global Harmonization Foundation model), capable of combining real video footage of actors into freeform videos, 360-degree panoramas and 3D scenes with dynamic re-lighting.
While diffusion video generators necessarily redraw the entire scene, distorting the actor's face, facial expressions and original performance from frame to frame, the DAWN architecture works on the principle of absolute preservation of the source material.
The system leaves the captured video untouched, isolates the character and adjusts the geometry and light physics to the new environment without the need for text instructions, manual rotoscoping or 3D tracking.
Unlike open source crap and plug-in sticks, DAWN was designed as a proprietary base model for cinema. It supports sources of any quality - from smartphone photography to raw footage from IMAX cameras, and has no limitations on the length of footage.
In summary:
Sub-pixel matte: perfect highlighting of complex transparent textures and moving hair without ghosting.
Zero deformation principle: preservation of the game and the original face without neural network morphers.
Dynamic Relighting: Automatic physical adjustment of light sources, shadows and highlights to the target scene.
Cinematic scale: video processing in any resolution (from iPhone to IMAX) and unlimited timing.
Sign up for the waiting list here:
pxld.ai.