Sign In

LEOSAM HelloWorld Turbo + LCM

492
5.2k
78
Verified:
SafeTensor
Type
Checkpoint Trained
Stats
4,606
Reviews
Published
Nov 29, 2023
Base Model
SDXL 1.0
Hash
AutoV2
97EBE87867
default creator card background decoration
LEOSAM's Avatar
LEOSAM

🖥️Welcome to try out the open-source GPT4V-Image-Captioner, developed by my friend and me. It offers a one-click installation and comes integrated with multiple features including image pre-compression, image tagging, and tag statistics. Recently, we also launched the webui plugin version of this tool, everyone is welcome to use it!

🌍欢迎加入QQ群"兔狲·AIGC梦工北厂",群号 :780132897 ;"兔狲·AIGC梦工南厂",群号 :835297318(入群答案:兔狲)。Telegram群聊“兔狲的SDXL百老汇”,链接:https://t.me/+KkflmfLTAdwzMzI1

This model is a run-accelerated version of the HelloWorld SDXL base model, combining both SDXL Turbo and LCM technologies. Paired with the Eular a sampler, it can generate images within 6-8 steps, which is 3 times faster than the original SDXL version.This model is optimized for the Eular a sampler and it is recommended to only use the Eular a sampler for output.

After multiple rounds of testing, we identified the optimal integration ratio of the SDXL Turbo and SDXL LCM models. The current test results show that for the same 8-step image generation, the effect is: Turbo+LCM dual fusion > Turbo single fusion > LCM single fusion.

The image quality of the 8-step output from the Turbo+LCM dual fusion version is very close to the HelloWorld original model!

The memory usage of the Turbo+LCM dual fusion version is consistent with the HelloWorld original version. Therefore, if you have enough memory, it is recommended to enlarge the direct output image by 1.5 times (still within 6-8 steps).

The recommended parameters for generating images with this model are:

Sampler: Eular a (Important! The model is specifically adapted to Eular a, other samplers may not yield as good results)

CFG scale: 2 (Important! It is recommended to have a CFG scale between 1.5~2.5)

Sampling steps: 8 steps (6~8 steps are acceptable)

Hires algorithm: ESRGAN 4x (Other upscaling algorithms can also be used, not a mandatory option. Please ensure that your GPU memory is sufficient)

Hires Upscale factor: 1.5x

Hires steps: 8 steps

Hires Denoising strength: 0.3

本模型为HelloWorld SDXL原版结合SDXL Turbo和LCM技术的运行加速版本。搭配Eular a采样器,可以在6-8步内生图,是原sdxl版本的三倍速。本模型针对Eular a采样器进行效果调优,只推荐使用Eular A采样器出图。

最新版经多轮测试,得到了SDXL Turbo以及SDXL LCM两种模型的最佳融合比例,目前的测试结果是同样8步生图,效果上:Turbo+LCM双融合>Turbo单融合>LCM单融合

Turbo+LCM双融合版本的8步出图画质已经非常接近HelloWorld原版模型

Turbo+LCM双融合版本在内存占用上与HelloWorld原版一致,因此如果内存足够,建议对直出图进行1.5倍放大(同样6-8步),加速版模型可以用与xl原版大模型直出1024分辨率图像相近的时间,实现1024分辨率出图+1.5倍放大

本模型推荐的生图参数:

采样器:Eular A(重要!模型针对Eular a专门适配,其他采样器效果不佳)

采样步数:8步(6~8步均可)

CFG scale:2(重要!CFG scale建议1.5~2.5)

放大算法:ESRGAN 4x(其他放大算法也可以,非必须选项,请确保GPU显存充足)

放大倍数:1.5倍

放大步数:8步

放大降噪系数:0.3