Qwen Pagodas — one prompt, three models

Same canonical voxel pagoda prompt · thinking on · no output cap · all rendered locally on DGX Sparks. Pick a tab, drag to orbit.

Flash-Next scene: 55,208 voxels, five-tier pagoda, koi pond, torii, bridge, day/night + four seasons, 16K sakura petals. Generated by SGLang on DGX Spark GB10 with the SM120 trtllm-gen sparse-decode patch. One-line render fix applied (season-color index bug in the model's own code); stats and raw output preserved in the run archive.