SolarRenderer in enderfga/solarrenderer is a research implementation that turns structural videos made by game engines or agents into footage with a visual style applied. According to the repository description, it renders in four stages using two kinds of LoRA based on MiniMax-H3. The code and weights, training data, and benchmarks are published. No separate latest release version is given in FACTS. The repository's language is Python and its license is Apache-2.0.
The structural videos used as input are videos that carry scene information, such as semantic maps, skeletons, and depth. So this is closer to applying a desired visual style on top of structural information than to general-purpose video generation. The README also describes a path that generates videos block by block in a causal configuration. The public evaluation bundle includes 208 difficult cases, 53 held-out videos, and the scoring code with result materials. Covering both reference rendering and the block-by-block generation path, and providing the implementation along with the evaluation materials, is what distinguishes this repository.
That said, the composition of the public evaluation materials alone does not allow any judgment about rendering quality or superiority over other approaches. FACTS gives the number of difficult cases and held-out videos and the presence of scoring code and result materials, but no concrete rendering performance figures. So within what can currently be confirmed, it is more accurate to view this as a research repository offering an implementation and evaluation materials for converting structural videos into stylized footage.