A fast, multi-file Streamlit toolkit for generating AI images (including a dedicated "This 3D Model Does Not Exist" mode). The project uses the Latent Consistency Model (LCM) for very fast generation (4–8 steps recommended).
- Main enhanced app: app_enhanced.py
- Colab-optimized app: app_colab.py
- Tiny demo app: app_demo.py
- 3D model standalone: this3dmodeldoesnotexist.py
- Random art: thisartdoesnotexist.py
- Example / helper: sample.py
- One-click Colab notebook: RUN_IN_COLAB.ipynb
- Run demo on Windows: run_demo.bat
- Environment: .env (gitignored via .gitignore)
- Requirements: requirements.txt
- Feature guide for 3D models: 3D_MODEL_FEATURE_GUIDE.md
- Changelog / speed notes: UPDATES_SUMMARY.md
- Start & deployment guide: START_HERE.md
- Upload RUN_IN_COLAB.ipynb to Colab.
- When prompted, upload either app_colab.py (recommended) or app_enhanced.py.
- Enable GPU in Runtime → Change runtime type → GPU.
- Run the notebook; the public URL will be printed.
- Create a virtualenv, install dependencies:
pip install -r requirements.txt- Add your HuggingFace token to
.envasHF_TOKEN=your_token(example in.env.example). - Run:
streamlit run app_enhanced.pyFor a tiny demo use: streamlit run app_demo.py or run run_demo.bat on Windows.
- LCM is the fast default. Optimal settings: 4–8 steps, guidance ~1.0–1.5, sizes 256–512 for best speed/quality balance. See UPDATES_SUMMARY.md.
- 3D Model mode generates 4 angles with the same seed for consistency. See
this3dmodeldoesnotexist.generate_random_3d_promptand 3D_MODEL_FEATURE_GUIDE.md. - Model-loading helper functions: see
sample.load_modelsandapp_enhanced.load_models. For CPU-specific loader seethisartdoesnotexist.load_model.
- "ScriptRunContext" warnings: harmless (use
app_colab.pyto avoid them). See START_HERE.md. - Out-of-memory: reduce image size, enable GPU, or use smaller batch sizes.
Include your preferred license or "All rights reserved" as needed.