RealModelBase02-LCM
This is a realistic model with roughly the same concept as RealModelBase01, but I came up with a new technique and constructed it in a different way from 01.
While 01 was mostly centered on photorealistic materials, the ratio of anime-style and semi-realistic materials has increased.
RealModelBase01とだいたい同じコンセプトですが，新しい手法を思いついて01とは違った形で構成してみたリアル系モデルです。
01がほぼフォトリアル系の素材を中心にしていたのに対して，アニメ系，セミリアル系の素材比率が上がっています。
The goal is to cover a wider range of themes while trying to reduce the number of scenes that look like paintings as much as possible, but there are still some paintings in sci-fi and fantasy scenes.
より広範なテーマに対応することを目標としつつ，絵画っぽくなる局面をできるだけ減らすことを目指していますが，SF,ファンタジー系のシーンではまだ絵画っぽさが残っています。
This is an LCM model, so please use LCM for the sampler. Steps 10 and CFG scale 2 are the standard.
LCMモデルですのでサンプラーにはLCMを使用してください。Steps 10，CFG scale 2が基準です。
All samples have been made to look older using L_Cute-Mature LoRA to accommodate the incomprehensible TOS. Please note that the actual output image may differ.
サンプルはすべて，意味不明なTOSに対応するため，L_Cute-Mature LoRAを使用して見た目の年齢を上げています。実際に出力される映像とは異なることがあることをご了承ください。
Description
FAQ
Comments (5)
I was surprised to suddenly receive so many buzz. Does this mean there is still a need for SD1.5? Thank you.
yes!, there is a need, please keep updating :)
LCM models can be used with only CPU, will be some audience for sure.
This model produces unique images which quite good looking and artistic, it follows prompt same good as anime models. Though LCM requires some tweaking, initial renders were really low quality, but it is fixable with correct values. My preset for diffusers pipeline:
image_size=(512, 768), steps=8, cfg=1.5, lora="", vae="LittleApple-fp16/vae-ft-mse-840000-ema-pruned", scheduler="lcm",It's been a while since I started using LCM regularly, but I recently realized that I was making models based on CFG 2.0. Burns that occur when there is a tendency for overfitting often appear at high CFG. Even before applying LCM, there were models that could only be used up to CFG 2 with other samplers, and I was regretting that I had packed them in too much. This is also why there are more elements in it than in other models.
However, is CFG1.5 appropriate for diffusers? I think the image has changed a little in ComfyUI as well. I checked everything with A1111, and I don't feel any breakdowns with CFG2.0. I'm training it to the very limit, so please adjust this in your execution environment. Thank you for the information.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.