Zero123-XL
The scaled-up Zero-1-to-3: novel view synthesis trained on 100M+ renders from the Objaverse-XL dataset.
About
Zero123-XL is what happens when the Zero-1-to-3 recipe meets a genuinely huge 3D dataset. The same team that built Objaverse-XL — over 10 million deduplicated 3D objects — trained the view-synthesis model on more than 100 million multi-view renders and showed that zero-shot generalization to new objects keeps improving with data scale. The result handles harder cases than its predecessor: people, cartoons, paintings, sketches.
It has become a standard reference point in the field: later papers like Cascade-Zero123 and Zero-to-Hero evaluate against the public Zero123-XL checkpoint, reporting PSNR 18.68, SSIM 0.883 and LPIPS 0.189 on the Objaverse test split. It is a research artifact rather than a product — no hosted API, no company behind it — but the checkpoint is publicly used and the paper (July 2023) is one of the most referenced in image-to-3D.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Independent evaluations in later papers: PSNR 18.68, SSIM 0.883, LPIPS 0.189 on the Objaverse test split (Cascade-Zero123; Zero-to-Hero)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: July 2023 paper from a six-institution collaboration; checkpoint used as a baseline in multiple independent papers (Cascade-Zero123, Zero-to-Hero) with reported Objaverse numbers. Strengths: open weights, strong zero-shot generalization, a de-facto standard baseline. Docked for: (1) research artifact with no maintenance or hosted access; (2) license terms for the checkpoint are not prominently documented; (3) superseded by newer view-synthesis methods. License: checkpoint publicly available; confirm terms before commercial use.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 8.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Paper July 2023; Allen AI, Columbia, UW, Stability AI, LAION, Caltech
- · Trained on 100M+ renders from 10M+ Objaverse-XL objects
- · Baseline in independent papers: PSNR 18.68 / SSIM 0.883 / LPIPS 0.189 on Objaverse
- · Checkpoint license terms not prominently documented