Testing ref2va: Accurately Recreating Cat Details with Just 3 Reference Images
smb3d · reddit · 2026-08-10
A developer shared their experience testing the ref2va model. By providing just 3 photos of their cat as references, the model was able to accurately capture and reproduce fine details, such as the cat's black chin, which many other image models tend to skip over.
More from Multimodal
- Testing World Models: AI Video Subjected to Brutal Ragdoll Physics — bilawalsidhu · 2026-08-10
- Testing MiniMax to Generate 'Malcolm in the Middle' Sitcom Clips — Yoshi8460 · 2026-08-10
- After Two Years, We Finally Have a 'Local Sora' — MustBeSomethingThere · 2026-08-10
- Testing MiniMax H3: Building a coherent image-to-video workflow on a single 4090 — Singingmute · 2026-08-10
- 3DGS Training Breakthrough: 1M Gaussians in 20 Seconds — janusch_patas · 2026-08-10
- Creator Uses Minimax H3 to Generate Warhammer-Style FPS Game Concept Demo — Far_Cast_Far_Wide · 2026-08-10