Downloads · 30 days
0
AKHYui/anything-v5-rknn-512
anything-v5-rknn-512 is a text-to-image model from AKHYui. Use it when you need an image from a text prompt. It is set up for rknn. The card lists the license as creativeml-openrail-m.
Anything V5(二次元)融合 LCM-LoRA 后,转换为 RK3588 NPU 的 RKNN 格式(fp16,静态形状)。
Downloads · 30 days
0
Access
Public
Updated Sep 10, 2026
Repo size
2.2 GB
Likes
0
Public
Click a slice to open those files.
.rknn2.2 GB · 100%
From the Hugging Face model README
Anything V5(二次元)融合 LCM-LoRA 后,转换为 RK3588 NPU 的 RKNN 格式(fp16,静态形状)。
| 项 | 值 |
|---|---|
| 基座 | Yntec/AnythingV5(SD1.5 架构) |
| 加速 | latent-consistency/lcm-lora-sdv1-5(手动融合,alpha=8.0) |
| 分辨率 | 512×512(固定) |
| 转换工具 | rknn-toolkit2 2.3.0(fp16,3 输入 UNet) |
| 板端 runtime | rknn-toolkit-lite2 2.3.x |
text_encoder/model.rknn # CLIP 文本编码器 (~238M)
unet/model.rknn # UNet + LCM-LoRA 融合 (~1.7G, fp16)
vae_decoder/model.rknn # VAE 解码器 (~186M)
scheduler/ # LCMScheduler 配置
tokenizer/ # CLIP tokenizer
app.py # 板端 FastAPI WebUI(负面词 + CFG + 历史)
run_rknn-lcm.py # 命令行推理脚本
# 依赖
pip3 install --break-system-packages rknn-toolkit-lite2 diffusers transformers torch fastapi uvicorn pillow "numpy<2"
# 运行 WebUI(端口 8100)
python3 app.py --port 8100
# 或命令行出图
python3 run_rknn-lcm.py -i . -o ./images --size 512x512 \
--num-inference-steps 4 --guidance-scale 1.0 \
--prompt "masterpiece, best quality, 1girl, long hair, blue eyes"
| CFG | 说明 | 耗时 |
|---|---|---|
| 1.0 | 单遍推理(忽略负面词) | ~36s |
| 1.5–2.5 | 两遍 CFG(含负面词) | ~60s |