LordPhenixDeNetra/my3d-dream
A 3D scene plugin for DSH that grounds decisions in a written knowledge corpus and validates declared render costs.
Most 3D generation stops at code that compiles. This plugin drives scenes from a knowledge corpus instead: one decision per file across six domains — lighting, material, composition, motion, budget, verification — where each record carries starting values, the checks to look at on the frame, and the limits it does not cover. Four tools expose that to the model: my3d_search ranks records and returns ids with one-line summaries, my3d_resolve returns a full record, my3d_scaffold writes a runnable Vite + Three.js project at a stated delivery target, and my3d_budget validates a scene's declared render costs. my3d capture takes rendered views for inspection, and the per-task context is roughly a third of reading the whole corpus, which is the point of indexing a growing base rather than pasting it into the prompt.
Install
dsh plugin --profile desktop add my3d-dreamThe README's plugin line. The package also installs a my3d command globally (npm install -g my3d-dream) and can be run without a global install via npx my3d …, which the README uses for its getting-started examples. Several verification commands (my3d search, resolve, scaffold, budget, capture) are documented alongside it.
Compatibility
The README states its measurements openly, including their limits: seven knowledge records and 465 terms in the current corpus, a context cost of about 345 tokens per task against about 2,121 to read everything (−34%), a 4,244 B skill at load, and frame figures (2 draw calls, 14,144 triangles, 1 texture) captured on a SOFTWARE rasterizer (SwiftShader) — which the README says does not represent a GPU, while noting draw calls and triangle counts are hardware independent.
Details
- Repo: LordPhenixDeNetra/my3d-dream
- Category: Vision & Multimodal
- Stars: 0
- Version: 0.1.0 (npm latest)
Alternatives
NeoHotaru/dsh-read-animation · PiyotaHu/muxiva-dsh-voice · corrinehu/dsh-chat-imagine