If you’re like me and have been playing around with Stable Diffusion a lot lately, you might be wondering what all the different sampler options are for. There are so many samplers to choose from now, it can be quite overwhelming! The model defaults on Euler A, which is one of the better samplers and has a quick generation time. The sampler can be thought of as a “decoder” that converts the random noise input into a sample image.
Changing the sampler can often get you a completely different image, that’s why it’s important to experiment with them. You can see a few comparison images below to check out how each sampler affects the image generation process.
It might be difficult to see the difference since the images are small and compressed, but I’ll list a few of my personal findings after using different samplers for the past month.
- The default Euler A sampler is a good all-round choice, but it often produced softer images compared to the others. If you’re trying to create an oil painting or relaxing painting, this could be a good choice.
- DDIM is one of my personal favorites and usually gives drastically different results compared to other samplers. Works well for abstract images.
- LMS is one of the fastest at generating images and only needs a 20-25 step count.
- DPM++ 2M Karras takes longer, but produces really good quality images with lots of details. Can be good for photorealistic images and macro shots.
- Heun is very similar to Euler A but in my opinion is more detailed, although this sampler takes almost twice the time.
These are the samplers which I usually use, although I’d recommend you to try out the others and experiment! You can also try to find two or three samplers you really like and then stick with those, this is usually my tactic with Euler A, DDIM and DPM++ 2M Karras.
Choosing a best sampler in Stable Diffusion really is subjective, but hopefully some of the images and recommendations I listed here will give you an idea of which ones you should try out!
