Animo
animoThe bake-off winner, shipped — spring rig + character builder + talking lip-sync
About this approach
The Vector Rig taken all the way to a product: a 24-channel spring rig over original parametric SVG art, orthogonal action x expression catalogs, a slot-based character builder, and TTS lip-sync with a 4-provider fallback chain (ElevenLabs -> OpenAI -> Deepgram -> OpenRouter) with Deepgram-STT timing recovery when a provider returns no word timings. Characters persist per user in Postgres.
- License
- Fully owned original art; MIT runtime deps. TTS is metered per provider.
- Cost
- $0 to render; TTS billed per character of speech
- Effort
- Highest of any entry — this is the one that was carried to a finished product
- Determinism
- Pose is a pure function of the state vector; audio timing follows the provider timestamps
- Customization
- First-class — every slot and colour is data, saved per user
- LLM fit
- Best of all — an LLM directs it AND can edit the art, because the art is code
Pros
- The proven winner: everything below it was evaluated to pick this
- Lip-sync that actually works, with a fallback chain so one dead key is not an outage
- DB-backed saved characters, frictionless auth, zero art-licensing risk
- No runtime dependencies — plain SVG in the DOM, SSR-friendly
Cons
- Every hairstyle and garment is hand-authored art work
- Front view only — turns would need a redrawn 3/4 atlas
- Cut-out joints show hard elbow breaks (no mesh deformation)
- Speech costs money and needs provider keys configured
Verdict — THE SHIPPED ANSWER — what the bake-off chose, running as a live app at /animo. Start here unless you need something it explicitly cannot do.
The highest-polish flat-vector figure in the repo (three review campaigns: doodle calibration, anatomical ROM limits, then eyes/third-tone/waist). The runtime imposes no ceiling — remaining distance to the bar is art iteration.