Mr-J-369/StableDiffusion-SD1.5-qnn2.28

📱 Now on Google Play — Fancy AI
On-device AI image generation, optimized for Qualcomm NPU phones.

Fancy AI — https://github.com/Mr-J-369/Fancy-Ai feel free to test and enjoy my android app.

Credits (source of safetensors and model guide)

• StableDiffusion — source: https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5

Apps (use these files)

• Local Dream — https://github.com/xororz/local-dream
• LaylaAI — https://www.layla-network.ai/

QNN SD Model Zips

Variants (per model)

• min
• 8gen1
• 8gen2

Use min for the broadest compatibility, which can run on non-flagship chips like 7gen1,8sgen3,etc. Use 8gen1 if you want to support Snapdragon 8gen1 and above (8gen1/2/3/4). If you have a 8gen1 device, this is recommended. Use 8gen2 if you want to support Snapdragon 8gen2 and above (8gen2/3/4). This will be much faster on 8gen2+ devices.

ZIP contents (same for all variants)

• clip.mnn
• tokenizer.json
• vae_decoder.bin
• vae_encoder.bin
• unet.bin
• Required HTP runtime libraries (libQnnHtp.so, libQnnSystem.so, and multiple libQnnHtpVXXskel.so/libQnnHtpVXXStub.so)

Filenames

• StableDiffusion_min.zip
• StableDiffusion_8gen1.zip
• StableDiffusion_8gen2.zip