Unverified50% confidenceFactExact time

H3 的 reference-to-video 画面比 image-to-video 更柔和，原因是两个 checkpoint 接受了不同的后训练

1

Sources

50%

Confidence

Medium-term (~90 days)

Relevance

8/12/2026

First Seen

Valid until: 11/10/2026

Sources

MiniMax H3 Team AMA Full Breakdown: What's Coming Next and What's Still Broken

Related Claims

VerifiedMiniMax-H3 features two capabilities: image-to-video generation and reference-based generation.67% similar UnverifiedBecause H3 is essentially a video model rather than an image model, it falls short in the detail quality of static images.64% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/735489

API

curl https://kongchang.com/api/v1/knowledge/claims/735489

MCP

get_claim(id=735489)