models/DMD2 Speed LoRA [SDXL, Pony, Illustrious] - dmd2_sdxl_4step_lora

DMD2 Speed LoRA [SDXL, Pony, Illustrious] - dmd2_sdxl_4step_lora

7/17/2025

5:31:01 AM

Related Keywords & Tags

6tz,distribution matching distillation,dmd2,dmd2_sdxl_4step_lora,fast image synthesis,hyper,lcm,lora,sdxl 1.0,speed,stable diffusion,tool,turbo

A cyberpunk female robot crouches on rubble under a stormy night sky illuminated by bright lightning with glowing eyes and a futuristic cityscape.

Cute happy orange tabby cat standing with arms up, smiling with closed eyes, an orange camera above it, with yellow stars on a white background.

Silhouette of Aku Bouzu figure standing in front of a vibrant key lime and deep crimson gradient background, holding a single rose from a thorny bush, surrounded by dripping watercolor and pop art style effects.

Flatline styled vector illustration of a girl in a purple dress sitting barefoot on a concrete docking station by the ocean with a spaceship flying nearby under a clear blue sky.

A girl with very long hair lying on a bed in a blue onesie, reaching surprised for an iPhone 11 Pro next to her in a dimly lit room with a sunset visible through the window and potted plants on the windowsill.

White-haired girl in a sleeveless black long dress eating a donut while leaning against a neon-lit dilapidated shop, holding a long sword behind her back.

A watercolor illustration of a cute cat-like monster with one long leg and one short leg standing on one hand, with multicolored fur and cat ears, interacting with two smaller cats on grass with small flowers.

A neon-colored jackalope character with glowing yellow eyes and long ears, standing in water surrounded by coral and blue foliage in a dynamic pose

A striking android woman with white hair and yellow cape, wearing sleek mecha armor with robotic limbs, standing confidently against a light blue background.

A vibrant graffiti-style mural featuring a surreal, detailed closeup of a large eye surrounded by pink, fleshy swirling patterns with intricate designs, set against a blue background.

A vibrant graffiti-style mural of a female cyclops with a long patterned neck and a single large eye, surrounded by colorful splashes and urban skyline.

Stylized minimalistic Japanese art of a girl with black spiked hair, pale skin, red eyeshadow, wearing a black choker and black top against a red background with a yellow circle.

Recommended Prompts

A closeup of a SFW

Recommended Negative Prompts

NSFW, naked, nude, watermark, text, simple background

Recommended Parameters

samplers

LCM, lcm

steps

cfg

1.4 - 1.8

other models

Checkpoints\Z\ILXL\WIP\Perfection_ILXL_Realistic_3.2.safetensors (83a0ec21ca), Checkpoints\Checkpoints\SDXL\Photorealistic\juggernautXL_v9Rundiffusionphoto2.safetensors (c9e3e68f89)

Recommended Hires (High-Resolution) Parameters

steps

Tips

Use a LoRA weight of 0.7 but feel free to adjust it.

If results appear washed out and bland, try increasing the CFG value.

You can merge this LoRA into a checkpoint for more convenient use without loading LoRA separately.

The LoRA enables generation at fewer steps with improved results but does not itself speed up generation.

Works with any SDXL or similar checkpoints.

Version Highlights

dmd2_sdxl_4step_lora

dmd2_sdxl_4step_lora_fp32

This is the official DMD2 implementation LoRA from the Improved Distribution Matching Distillation for Fast Image Synthesis paper.

dmd2_sdxl_4step_lora

Original model at Huggingface tianweiy/DMD2.

Recommended Settings

Steps: 8

LoRA weight: 0.7

Sampler: LCM

Scheduler: Normal, Simple

CFG: 1.4 - 1.8

I use similar settings on both a low resolution base image, and the upscaled refinement pass (hires fix).

Usage

Load the LoRA like any other. I recommend a weight of 0.7, but feel free to play around.

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 and bland, try increasing the CFG.

Licence

Creative Commons Attribution Non Commercial 4.0

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 Tianwei Yin.