Image-to-3D review
Kollektive first look, 18 October 2026: six Source images given to TRELLIS-image-large and TripoSR on one rented RTX 3090, 12 executed cases.
Drag to rotate, scroll to zoom. The first viewer in each card is the 50,000-face member-budget mesh with vertex colours; the second is the model's raw output before simplification.
Catalog photographs: Amazon Berkeley Objects dataset, CC BY 4.0, Amazon.com. The teapot is a synthetic control image. Temporary review page.
Cost analysis
Nominal figures from the quoted rate of $0.2633/hour on RTX 3090 (France, FR)
and the measured timings of this session. They are not a provider invoice; the actual charge posts later and transfer prices are the host's quoted rates.
Session item Value Instance lifetime 17.2 min GPU rental $0.075 Inbound transfer (runtime 3.7 GB + models 6.5 GB + sources) 10.3 GB · $0.027 Allocated storage (60 GB) $0.005 Nominal session total $0.11 (ceiling was $2)
Model Inference / object Post-processing Cold load Peak VRAM GPU time / object (warm) Cost / object 100 objects, one session triposr 2.9–3.6 s 0.5–0.8 s 50 s 3.2 GB 3.7 s $0.0003 $0.05 trellis-image-large 9.1–12.7 s 1.7–3.2 s 34 s 15.7 GB 13.0 s $0.0010 $0.12
Fixed costs dominate short sessions: instance start, the runtime transfer, the model download and the cold load. TripoSR fits an 8 GB GPU, which widens the pool of cheap offers; TRELLIS needs a 16 GB class card. TRELLIS ran at its 25-step defaults here; its 12-step README setting roughly halves inference time and is an untested variant in the case matrix.
Amazon Brand – Rivet Modern Geometric Pattern Decorative Stoneware Vase, Large Centerpiece, 12.25"H, White VASE · Amazon.com, Amazon Berkeley Objects Dataset
Source trellis-image-large · pipeline defaults {"faces": "50000", "seed": 1, "slat_guidance": 5.0, "slat_steps": 25, "sparse_guidance": 5.0, "sparse_steps": 25, "texture": "color"}
member-budget GLB raw geometry (14412 KiB) inference 12.7s, load 33.8s (cold), faces 491890 → 50000, cutout u2net, peak VRAM 15.7 GB
triposr · upstream defaults {"faces": "50000", "resolution": 256, "texture": "color", "threshold": 25.0}
member-budget GLB raw geometry (2358 KiB) inference 3.6s, load 50.1s (cold), faces 80396 → 49965, cutout u2net, peak VRAM 3.2 GB
Rivet Pineapple Ceramic Table Lamp, Modern, 18" H, White and Gold LAMP · Amazon.com, Amazon Berkeley Objects Dataset
Source trellis-image-large · pipeline defaults {"faces": "50000", "seed": 1, "slat_guidance": 5.0, "slat_steps": 25, "sparse_guidance": 5.0, "sparse_steps": 25, "texture": "color"}
member-budget GLB raw geometry (18279 KiB) inference 11.0s, load 0.0s (warm), faces 623904 → 50000, cutout u2net, peak VRAM 15.7 GB
triposr · upstream defaults {"faces": "50000", "resolution": 256, "texture": "color", "threshold": 25.0}
member-budget GLB raw geometry (3468 KiB) inference 3.0s, load 0.0s (warm), faces 118240 → 49986, cutout u2net, peak VRAM 3.2 GB
Amazon Brand – Stone & Beam Parson Armless Dining Room Chairs, 40 Inch Height, Set of 2, Beige CHAIR · Amazon.com, Amazon Berkeley Objects Dataset · multiple objects in frame
Source trellis-image-large · pipeline defaults {"faces": "50000", "seed": 1, "slat_guidance": 5.0, "slat_steps": 25, "sparse_guidance": 5.0, "sparse_steps": 25, "texture": "color"}
member-budget GLB raw geometry (9601 KiB) inference 9.2s, load 0.0s (warm), faces 327680 → 50000, cutout u2net, peak VRAM 15.7 GB
triposr · upstream defaults {"faces": "50000", "resolution": 256, "texture": "color", "threshold": 25.0}
member-budget GLB raw geometry (2837 KiB) inference 3.0s, load 0.0s (warm), faces 96728 → 49978, cutout u2net, peak VRAM 3.2 GB
Amazon Brand – Rivet Mid-Century Modern Cone Nightstand Table Desk Lamp with LED Light Bulb - 15 Inches, Black and Gold LAMP · Amazon.com, Amazon Berkeley Objects Dataset
Source trellis-image-large · pipeline defaults {"faces": "50000", "seed": 1, "slat_guidance": 5.0, "slat_steps": 25, "sparse_guidance": 5.0, "sparse_steps": 25, "texture": "color"}
member-budget GLB raw geometry (11526 KiB) inference 9.1s, load 0.0s (warm), faces 393376 → 50000, cutout u2net, peak VRAM 15.7 GB
triposr · upstream defaults {"faces": "50000", "resolution": 256, "texture": "color", "threshold": 25.0}
member-budget GLB raw geometry (2157 KiB) inference 2.9s, load 0.0s (warm), faces 73556 → 49971, cutout u2net, peak VRAM 3.2 GB
Amazon Essentials Onore Fashion-Sneakers, Evelina, 8 Toddler SHOES · Amazon.com, Amazon Berkeley Objects Dataset
Source trellis-image-large · pipeline defaults {"faces": "50000", "seed": 1, "slat_guidance": 5.0, "slat_steps": 25, "sparse_guidance": 5.0, "sparse_steps": 25, "texture": "color"}
member-budget GLB raw geometry (21367 KiB) inference 11.0s, load 0.0s (warm), faces 729364 → 50000, cutout u2net, peak VRAM 15.7 GB
triposr · upstream defaults {"faces": "50000", "resolution": 256, "texture": "color", "threshold": 25.0}
member-budget GLB raw geometry (2912 KiB) inference 3.0s, load 0.0s (warm), faces 99252 → 49983, cutout u2net, peak VRAM 3.2 GB
Controlled synthetic ceramic teapot TEAPOT · OpenAI imagegen, generated for Kollektive 3D evaluation on 2026-10-17 · controlled synthetic image; exclude from claims about performance on real photographs
Source trellis-image-large · pipeline defaults {"faces": "50000", "seed": 1, "slat_guidance": 5.0, "slat_steps": 25, "sparse_guidance": 5.0, "sparse_steps": 25, "texture": "color"}
member-budget GLB raw geometry (15618 KiB) inference 10.6s, load 0.0s (warm), faces 533024 → 50000, cutout u2net, peak VRAM 15.7 GB
triposr · upstream defaults {"faces": "50000", "resolution": 256, "texture": "color", "threshold": 25.0}
member-budget GLB raw geometry (2756 KiB) inference 2.9s, load 0.0s (warm), faces 93964 → 49982, cutout u2net, peak VRAM 3.2 GB
