diegochavez's picture
Upload README.md with huggingface_hub
8539f51 verified
|
Raw
History Blame Contribute Delete
1.46 kB
---
license: cc-by-nc-4.0
tags:
- lora
- wan2.1
- video-generation
- daydream-scope
- krea-realtime
base_model: Wan-AI/Wan2.1-T2V-14B
---
# DayDream Scope LoRAs — Wan2.1 14B
Three custom LoRAs trained for the Krea Realtime Video pipeline on [DayDream Scope](https://daydream.live). Built to pair with a cellular automata preprocessor for real-time AI video generation.
## Models
| File | Style | Description |
|------|-------|-------------|
| `anmdf_mlt_wan2-1_14B_000001000.safetensors` | Animated Melt | Liquid chrome, melt-reform cycles |
| `tdcrtv_000001000.safetensors` | CRT Recapture | VHS degradation, scan lines, analog glitch |
| `slm_mld_grth_v2_000000400.safetensors` | Slime Mold Growth | Bioluminescent shimmer, organic expansion |
## Details
- **Base model**: Wan2.1-T2V-14B
- **Training**: Rank 16, qfloat8 quantization, OstrisAI-Toolkit
- **Datasets**: 14-17 clips each, richly captioned
- **Hardware**: RTX PRO 6000 on RunPod
## Usage with DayDream Scope
Download a LoRA and place it in your Scope models directory:
```bash
# Default location
~/.daydream-scope/models/lora/
# Or on RunPod
/workspace/models/lora/
```
Then select it from the LoRA Adapters dropdown in the Scope UI.
## Plugin
These LoRAs are designed to work with the [Cellular Automata Scope Plugin](https://github.com/diegochavez-io/cellular-automata_plug-in) — a living organism that generates constant-motion video as input for Krea Realtime.