Rough practice, not approved production assets. Authored fixtures use each assignment’s picture 1 overall reference. Source FBX files were frozen before texture exercises. No target-engine captures or human review rounds exist. $0 new metered generation cost; no priced-credit assumptions.
Drag to orbit · Scroll/pinch to zoom · right-drag to pan. This Three.js diagnostic loads the exact original FBX bytes, not the preview GLB. Browser lighting is not Blender lighting or a target-engine baseline. #6 compares incorrect alpha-as-roughness with 1−alpha routing. #7 is a synthetic normal-Y test, not an identified production defect.
Matched Blender Comparisons
Same camera, lighting, exposure and shader settings within each image pair, except the explicitly tested routing variable. Exposure-v2 proofs retain the original, overly bright WIP renders in each package.
Picture 1 / AI appearance reference![Overall reference]()
Original / synthetic baseline![Original material proof]()
Current / draft attempt![Current material proof]()
What Passed—and What Did Not
Independent saved-PNG checks found zero changed pixels outside the authorized masks, zero protected-pixel changes, and zero unnamed-channel edits in all ten first attempts. All ten primary FBX hashes were unchanged. These are technical scope checks, not visual acceptance.