Views
No views yet
.pth (pickle) files for the transformer and VAE checkpoints. This repo replaces them with .safetensors files — no arbitrary code execution, faster loading, same weights.| File | Original | This repo |
|---|---|---|
transformer/transformer.pth (31 GB) | pickle | transformer/transformer.safetensors |
vae/Wan2.1_VAE.pth (485 MB) | pickle | vae/Wan2.1_VAE.safetensors |
JoyAI-Image-Und/ | already safetensors | unchanged |
infer_config.py | references .pth | updated to reference .safetensors |
1git clone https://github.com/SanDiegoDude/JoyAI-Image.git
2cd JoyAI-Image
3python -m venv .venv && source .venv/bin/activate
4pip install -e .
5
6# Models auto-download on first run
7# Default: FP8 transformer + 8-bit text encoder (offload mode)
8python app.py
9
10# Full precision (bf16 transformer + bf16 text encoder, needs ~48 GB+ VRAM)
11python app.py --fullprecision --16bit-vlm --highvram
12
13# Minimum VRAM (~13 GB active, fits RTX 4090)
14python app.py --nf4-dit --4bit-vlm
15
16# CLI inference
17python inference.py \
18 --prompt "Turn the plate blue" \
19 --image test_images/test_1.jpg \
20 --output result.png \
21 --steps 18 --guidance-scale 4.0 --seed 42
22
23# Headless REST API
24python app.py --headless-api 7500 --nf4-dit --4bit-vlm--nf4-dit runtime NF4 quantization (~8 GB), which provides the best quality-to-size ratio for 4-bit since it quantizes directly from full-precision weights.