AI

We’re MOSS, the team behind MOSS-Image, MOSS-TTS, and more — Ask Us Anything!

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 30/10/2026, 22:47:26 UTC.

Briefings

MOSS, a research team based in Shanghai, announced the release of the MOSS-Image model, a text-to-image system with a 5B DiT backbone and 3.7M Hugging Face downloads. The team also introduced MOSS-TTS, a suite of speech synthesis models including a 100M-parameter variant, and MOSS-VL for video understanding. The AMA was held on October 30, 2026, from 8-11 AM PT, with the thread continuing for 48 hours.

Key points

• MOSS-Image is a text-to-image model with a 5B DiT backbone, Chinese and English text rendering, and 3,702,490 downloads on Hugging Face.
• MOSS-TTS includes a lightweight 100M-parameter variant for CPU-friendly streaming inference, alongside other voice synthesis and audio understanding models.

Context

MOSS is a research team from Shanghai, with roots at Fudan University and the Shanghai Innovation Institute, focused on AI that follows real-world events through speech, audio, and visual understanding.

Limitations

The evidence is a Reddit AMA thread from October 30, 2026, with limited coverage of the team's broader research scope and specific technical details beyond the listed models.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 31/10/2026, 00:40:07 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeifv3bqc66so6577uc4hdolg4oweh5f6awr43lm3o52eu4n54ywbfa