What is DreamActor-M1 and how does it compare to Act-Two and Wan-Animate?

The question

DreamActor-M1 was named in the wan-animate paper's human-evaluation comparison alongside Runway Act-Two as a closed-source competitor. The survey didn't characterize it. What is it (vendor, architecture, capabilities, availability)?

Why it matters

If Wan-Animate's authors picked DreamActor-M1 as a baseline, it's likely a credible system in the same workflow niche as runway-act-two. Worth understanding before recommending one over the other.

What we currently believe

Nothing concrete. Only that it appeared in Wan-Animate's human-eval comparison and lost to Wan-Animate on the authors' tests.

Evidence we have

• 2026-05-07-ai-avatar-motion-mimicking-models-survey mentions DreamActor-M1 only in passing as a Wan-Animate baseline.

Evidence we need

• A primary source for DreamActor-M1 (paper, vendor page, GitHub).
• Vendor identity, model architecture, input format, output capabilities.
• Whether it's open or closed.

How to resolve

• Targeted web search or a follow-up /autoresearch pass scoped to "DreamActor-M1".

Related

• wan-animate
• runway-act-two
• independent-avatar-benchmarks