I couldn't find the DMD2 LoRA on Civitai, so I copied it from <a target="_blank" rel="ugc" href="https://huggingface.co/tianweiy/DMD2">https://huggingface.co/tianweiy/DMD2</a> . It is similar to the LCM-LoRA and Lightning-LoRA in that it accelerates image generation while approximately preserving quality. Apparently DMD2 must be used with the LCM scheduler and CFG 1.0 (can't handle CFG &gt; 1.0). Steps should be 4, but slightly more seems to improve the quality. The absence of CFG (my guess) means that prompt adherence is not perfect.

DMD2 LoRA

dmd2_sdxl_4step_lora_fp16.safetensors

SDXL1.0 4 steps

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 itIf your rights have been violated, please feel free to contact me at any time to delete itform :<a target="_blank" rel="ugc" href="https://huggingface.co/tianweiy/DMD2#dmd2-model-card">https://huggingface.co/tianweiy/DMD2#dmd2-model-card</a>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:1Sampler : LCM or Euler aStep: 4-10

DMD2  accelerator Lora model

sdxl_4step_lora_fp16

This is the official DMD2 implementation LoRA from the Improved <a target="_blank" rel="ugc" href="https://arxiv.org/abs/2405.14867">Distribution Matching Distillation for Fast Image Synthesis</a> paper.<h1 id="dmd2_sdxl_4step_lora">dmd2_sdxl_4step_lora</h1>Original model at <a target="_blank" rel="ugc" href="https://huggingface.co/tianweiy/DMD2">Huggingface tianweiy/DMD2</a>.<h2 id="recommended-settings">Recommended Settings</h2>Steps: 8LoRA weight: 0.7Sampler: LCMScheduler: Normal, SimpleCFG: 1.4 - 1.8I use similar settings on both a low resolution base image, and the upscaled refinement pass (hires fix). The settings above are my personal recommendations, but I suggest you play around with the settings to find what works for you and the model you are using.<h2 id="combining-both-loras">Combining Both LoRAs</h2>It's also been reported to perform better to use both the fp32 and fp16 at the same time at 0.5 weight each.<h2 id="recommended-lora-weight">Recommended LoRA weight</h2>Load the LoRA like any other. I personally recommend a weight of 0.7, but others swear it's better at 1.0 in weight.<h2 id="usage">Usage</h2>You should be able to use it with any SDXL or similar checkpoint. Use the settings above.You can also merge this into a checkpoint and save that, to include the result of the LoRA into the checkpoint directly. This means you no longer need to load this as a LoRA, and you can still use it with LCM and low steps for good result.The LoRA itself doesn't add any speed improvement, it just allows you to generate at fewer steps, with better result.If the results are washed out, desaturated or bland, try increasing the CFG.In the comments below there are a bunch of interesting discussions, so please explore them. Here are a few noteworthy:<a target="_blank" rel="ugc" href="https://civarchive.com/models/1608870/dmd2-speed-lora-sdxl-pony-illustrious?commentId=991371&amp;dialog=commentThread">https://civarchive.com/models/1608870/dmd2-speed-lora-sdxl-pony-illustrious?commentId=991371&amp;dialog=commentThread</a><a target="_blank" rel="ugc" href="https://civarchive.com/models/1608870/dmd2-speed-lora-sdxl-pony-illustrious?commentId=1179404&amp;dialog=commentThread">https://civarchive.com/models/1608870/dmd2-speed-lora-sdxl-pony-illustrious?commentId=1179404&amp;dialog=commentThread</a><a target="_blank" rel="ugc" href="https://civarchive.com/models/1608870/dmd2-speed-lora-sdxl-pony-illustrious?commentId=1034300&amp;dialog=commentThread">https://civarchive.com/models/1608870/dmd2-speed-lora-sdxl-pony-illustrious?commentId=1034300&amp;dialog=commentThread</a><a target="_blank" rel="ugc" href="https://civarchive.com/models/1608870/dmd2-speed-lora-sdxl-pony-illustrious?commentId=1123128&amp;dialog=commentThread">https://civarchive.com/models/1608870/dmd2-speed-lora-sdxl-pony-illustrious?commentId=1123128&amp;dialog=commentThread</a><h2 id="licence">Licence</h2><a target="_blank" rel="ugc" href="https://huggingface.co/tianweiy/DMD2#license">Creative Commons Attribution Non Commercial 4.0</a>I am not the creator of the model, just uploading the model here to make it easier to download and reference in creations. Full credit goes to <a target="_blank" rel="ugc" href="https://huggingface.co/tianweiy">Tianwei Yin</a>.

Files with SHA256: b3d9173815a4b595991c3a7a0e0e63ad821080f314a0b2a3cc31ecd7fcf2cbb8