DMD2 accelerator Lora model - CivArchive (CivitAI Archive)

DMD2 accelerator Lora model - sdxl_4step_lora_fp16

This is the best acceleration model I have ever used, but I couldn't find it on Civitai, so I am reposting it,Glory to Tianwey, I am only spread it

If your rights have been violated, please feel free to contact me at any time to delete it

form :https://huggingface.co/tianweiy/DMD2#dmd2-model-card

这是我用过最棒的加速模型，Civitai上一直没找到所以转载过来，如果这侵犯了任何人的权益，请联系我我将第一时间删除。荣耀归于作者，我只是搬运工。

下面是一些说明：

It is certainly sdxl, but it also has a very good effect on Pony. Because if it is not marked as Pony, it will not be able to be used online in the creator. I hope this model can reduce the consumption of buzz.

两个版本都是sdxl，但它对Pony也有很好的效果。因为如果它没有标记为Pony，它将无法在在线生图器Creater中在线使用。我希望这个模型能减少在线生图器Buzz的消耗，所以我把其中一个标记为了Pony。

The setting I currently commonly use is cfg1, LCM SGM Uniform,6-8 step。Also, I think its FP32 version performs much better than FP16。So if possible, try to use the fp32 version as much as possible. I marked PONY because I want to use it in the creator to save buzz consumption. Another tip is that the minimum number of steps for the creator in the UI is 10, but in reality, when entering a lower number of steps such as 6, pressing enter can directly generate and consume less Buzz

我目前常用的设置是cfg1，LCM SGM统一，6-8步此外，我认为其FP32版本的性能远优于FP16因此，如果可能的话，尽量使用fp32版本。我标记了PONY，因为我想在创建者中使用它来节省Buzz消耗。另一个小技巧是，Creater的UI中Step的最小步数是10，但实际上，当输入较低的步数（如6）时，按enter键可以直接生成并且消耗较少的Buzz。

CFG:1

Sampler : LCM or Euler a

Step: 4-10

Description

FAQ

LORA

Pony

by civitai

Download (Beta)

tool

Details

Downloads

210

Platform

CivitAI

Platform Status

Deleted

Created

8/14/2024

Updated

8/9/2025

Deleted

8/7/2025

Files

dmd2_sdxl_4step_lora_fp16.safetensors

Size:

375.61 MB

SHA256:

b3d9173815a4b595991c3a7a0e0e63ad821080f314a0b2a3cc31ecd7fcf2cbb8