From f657079bf6c3712cfd82579535b03be581fdce76 Mon Sep 17 00:00:00 2001 From: GoldMath Date: Sun, 12 Mar 2023 01:37:10 +0100 Subject: [PATCH] added detail about "dpm_adaptive" --- How-to-Use.md | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/How-to-Use.md b/How-to-Use.md index 01d1462..abf2f40 100644 --- a/How-to-Use.md +++ b/How-to-Use.md @@ -22,7 +22,8 @@ Here is a rough overview of the behaviors of different sampler implementations: Most samplers are deterministic. If re-run with the same seed and settings, it will generate the same image. You can also increase the step size to see if artifacts in the image clear up, but it may be a slightly different image. -Note that **Ancestral** sampler, `euler_a`, `dpm2_a` and `dpmpp_2s_a` are *not always* deterministic. They can create different images when run in different batch sizes. +Note that **Ancestral** sampler, `euler_a`, `dpm2_a` and `dpmpp_2s_a` are *not always* deterministic. They can create different images when run in different batch sizes. +`dpm_adaptive` use an automatic steps count algorithm, ignoring the user Inference Steps setting. [![SD2](https://user-images.githubusercontent.com/7282547/222918227-052d5c8f-f6db-4bcb-bfe8-2902228171db.jpg)](https://user-images.githubusercontent.com/7282547/222918227-052d5c8f-f6db-4bcb-bfe8-2902228171db.jpg)