Stable diffusion failing to load

[GabrielandPeach]

Xiaomi mi 14 with version 0.4.2

Diffusion model fails to load and crashes back to the model menu

[bi4key]

https://github.com/alibaba/MNN/blob/master/apps/Android/MnnLlmChat/README.md#version-040

Try new version 0.4.3 and give feedback

[Juude]

try 0.4.3 please,you can download from mnnchat page

[GabrielandPeach]

0.4.3 updated, model still fails to load (xiaomi 14). It crashes back to the model page. I will try with xiaomi pad 7 in a few days when I have access to it again.

[Juude]

select memory saving mode in settings

[Disonantemus]

I was able to load the model with 0.44 and Memory Saving Mode, the result was abstract/noise, and takes a lot of time, like no NPU acceleration and only using CPU:

Local Dream works good, with CPU only in my phone, but is a lot more flexible with prompt and options.

Local Diffusion has a lot more options, but result is abstract/noise:

[GabrielandPeach]

0.4.3 fails to load

0.4.4 generates garbage output

[bitxsw93]

you can try with memory saving mode in settings,

and try to use lowercase letters prompt, like "a lovely cat"

[Disonantemus]

Observations:

• MNN Chat 0.44
• Settings > Diffusion Memory Mode: Memory Saving Mode
• 1st generation was garbage/noise, when starting (4%-8%) some screen glitches
• 2nd generation was ok (same prompt)
• Other Diffusion Models crash or doesn't load the model
• I can't find a way to save generated image
• I miss:
  • negative prompt
  • smaller resolutions: 128, 256, 384, 256px
  • img2img : denoise strength
  • more advanced settings

---

Machine Translation:

观察：

• MNN Chat 0.44
• 设置 > 扩散记忆模式：内存节省模式
• 第一代图像杂乱/噪点较多，启动时（4%-8%）会出现一些屏幕故障
• 第二代图像正常（提示相同）
• 其他扩散模型崩溃或无法加载
• 我找不到保存生成图像的方法
• 我遗漏了：
  • 负面提示
  • 较小的分辨率：128、256、384、256px
  • img2img：降噪强度
  • 更多高级设置

---

Samsung Galaxy S20FE (Q1 2020)
SM-G780G/DS
Qualcomm Snapdragon 8250 ; SM8258 Snapdragon 865 ; 1 + 3 + 4 cores Kryo 585 (2.84 GHz Cortex-A77 + 2.42 GHz Cortex-A77 + 1.80 GHz Cortex-A55)
Adreno 650 587 MHz (901.6 GFLOPS in FP32)
Hexagon 698 (15 TOPs)
RAM 8GB
arm64-v8a
Vulkan Version 1.1.0
GLES Version 3.2

[github-actions[bot]]

Marking as stale. No activity in 60 days.