The ready-made envs' camera observations (camera=True: 64 x 64 pixels, 4 samples per pixel, no denoiser, the
default studio light) come out bright and with visible sampling noise.
On a table, cube and ball scene seen like the env camera: mean brightness 214 of 255, and 3.9 grey levels RMS noise
against a 256-sample image (95 % of pixels within 5 levels).
More samples cut the noise: 1.5 levels at 16 samples, 0.7 at 64. Each step costs about that factor in render time.
Build your own alloy.camera.Camera(..., spp=16) over the env's sim for cleaner images, and pass
render(..., exposure=-1) (one stop darker) when the images look washed out.