OmniSenterStep (Omni SS) — the multimodal + music + agentic¶

Status: ⏳ planned · HF target: sovthpaw/omnisenterstep

Identity¶


Full name	OmniSenterStep (abbreviated: Omni SS)
Type	OmniSenter + AceStep Darwin fusion
Modalities	text + vision + audio + speech (in + out) + music (out)
Self-evolution	no (Senter, not Senter Ohm)

What it is¶

OmniSenter + AceStep Darwin fusion. So you get: - The Senter agentic core (function calling, notebook, plugin routing) - The multimodal vision (from Cosmos/Omni) - The music generation (from ACE-Step) - The notebook-keeper - All in one model

The most capable single model in the family short of the flagship (Senter Ohm).

Naming¶

OmniSenter = Omni + Senter agentic core
Step = ACE-Step music generation
Omni SS = the shorthand

Build approach¶

Base: OmniSenter 12B (or similar Senter variant)
- ACE-Step v1.5 XL 4B DiT (Darwin-merged as a routed expert or as a head)
Multimodal heads attached: Whisper (audio in), NaViT (vision in), talker + token2wav (speech out), ACE-Step DiT (music out)
Dense model, no MoE routing (or single-expert "MoE" if Senter Ohm shares weights)

When to use this¶

Use OmniSenterStep when you need everything (function calling + notebook + multimodal I/O + music generation) but don't need the flagship's full 32B MoE capacity or its self-evolution capability. This is the "production" model for the OmniSenter system.