Evaluating Image-to-3D Beyond the Input View: Checking Backs and Bottoms
rohanpaul_ai · x · 2026-10-27
Part 2 of rohanpaulai's Hi3D evaluation thread, on methodology:
- Most image-to-3D tests only grade the view shown in the input image; this team also inspected the back and bottom of every result
- Example: on a winged armored figure, TRELLIS.2 collapsed into a flat panel while Hi3D v1.0 kept the wings and limbs as connected parts
- Caveat: since one photo can't reveal hidden surfaces, those views remain the model's best guess rather than exact reconstruction
Related event: Hi3D Beats Eight Rivals in 83.5% of 562 Image-to-3D Comparisons(2 posts)→
More from Multimodal
- Prompt Motion: Free Gallery of 200+ Claude Opus Motion Videos with Exact Prompts — victor_explore · 2026-10-27
- Depianoer: a joke app to strip the piano Seedance adds to every scene — umpherj · 2026-10-27
- kohya's kv-cache Control LoRA Implementation Cuts MOSS-Image Overhead to Half of Concat — bdsqlsz · 2026-10-27
- Opus 5.5 codes music, 3D stage and lighting for an anime MV with 38 synced dance moves — majidmanzarpour · 2026-10-27
- Redditor argues full HDR workflows, not SDR-to-HDR conversion, kill the "AI slop" look — robertwellesley · 2026-10-27
- $1M AI Film Award Returns: Entries Must Be 70% Seedance-Generated — HashemGhaili · 2026-10-27