Multimodal Imagination model stack: Gradio apps, Colab/FastAPI backends, training tooling, and an optional Next.js chat UI. This repo holds versioned runtimes under versions/ plus shared modules/ (weights and assets are not committed—see below).
| Path | Role |
|---|---|
versions/v1.1.2/ |
Gradio app entry app.py, runtime imagination_runtime/ |
versions/v1.2.1/ |
Gradio + optional QLoRA training (train_imagination.py) |
versions/v1.3/ |
Colab/FastAPI backend (colab_backend.py, imagination_runtime/chat_http.py) |
v0-imagination-ui/ |
Next.js frontend (separate git history; submodule-style nested repo) |
modules/ |
Expected on-disk model trees (CAD, reasoning, research, image, …) |
.github/workflows/ |
CI (e.g. validate-v112.yml) |
Agent-oriented notes for Cursor and automation live in AGENTS.md.
- Python 3.10+ for app code under
versions/ - Node.js 20+ (or pnpm/npm) only if you work on
v0-imagination-ui/
Runtimes expect IMAGINATION_ROOT to point at a directory that contains the expected modules/ layout (and temp/ as needed). Without real weights you can use a minimal fake tree for import/UI smoke tests—see AGENTS.md and version-specific READMEs.
cd versions/v1.1.2
pip install -r requirements.txt
export IMAGINATION_ROOT=/path/to/your/root
python app.pycd versions/v1.2.1
pip install -r requirements.txt
export IMAGINATION_ROOT=/path/to/your/root
python app.pyFrom versions/v1.3/ (after dependencies and IMAGINATION_ROOT are set), see colab_backend.py and imagination_runtime/chat_http.py. Typical flow: run Uvicorn on the FastAPI app and point the Next UI at the public URL via NEXT_PUBLIC_API_URL.
cd v0-imagination-ui
pnpm install # or npm install
cp .env.example .env.local # if present; set NEXT_PUBLIC_API_URL to your backend origin
pnpm devImagination is a research/engineering stack. Model outputs can be wrong or unsafe; deploy with appropriate safeguards for your use case.