Sign In

Real Model Base01

77
1.8k
16
Updated: Jun 13, 2024
base model
Verified:
SafeTensor
Type
Checkpoint Merge
Stats
757
Reviews
Published
Jun 6, 2024
Base Model
SD 1.5
Hash
AutoV2
B65B9CD3A7

Real Model Base01C LCM

Version B was a bit half-baked, so I made it a bit more semi-realistic and increased the age with the cute-mature slider. I didn't really want to do this because it could have some drawbacks. This is the end of the realistic model. I don't think I'll be using this as a base model, so I've made it LCM.

In the future, I'll consider a semi-realistic model, which I don't think will get many complaints.


Real Model Base01B

Although Real Model Base01 does not incorporate any age manipulation, for unknown reason it tends to produce images that look younger than any of the merged materials.

CIVITAI prohibits the posting of photo-like images of minors, so a large number of images have been deleted.

Real Model Base01B makes the characters look slightly older and has a slightly higher proportion of Asian characters, as well as (slightly) improving the hands and fingers, and enhancing some SciFi and other elements. The adjustment of the picture may not be enough. I'll leave Real Model Base01 as is, so feel free to use whichever you prefer.

Real Model Base01に年齢操作を行うようなものは組み込んでいないのですが,なぜかマージしたどの素材よりも若く見える画像が出る傾向にあります。

CIVITAIでは未成年者の写真風画像の掲載が禁じられており,それでかなりの画像が削除されてしまっています。

Real Model Base01Bでは,見た目の年齢が少し上がり,アジアン比率も少し高くなっているほか,手指などが(ほんの少し)改善され,SciFi系などの一部の要素が強化されています。絵柄の調整は十分ではないかもしれません。Real Model Base01のほうもそのまま残しておきますので,お好きなほうを使ってください。


Real Model Base01

This is a realistic base model. It is a non-LCM model, but has only been tested with LCM. When using LCM LoRA, use Steps = 8-10, CFG scales = 2. When not using LCM, Steps = 20, CFG scales = 7 should be fine (probably).

If using Hires.Fix, apply it very weakly. I recommend R-ESRGAN 4x+, Hires steps = 2, Denoising strength = 0.15, Upscale by 2.0. I think the image is less likely to be destroyed with R-ESRGAN than with UltraSharp.

リアル系のベースモデルです。非LCMモデルですが,LCMでしかテストしていません。LCM LoRA使用時は,Steps=8~10,CFG scales=2程度で利用してください。LCM非使用時は,Steps=20,CFG scales=7で大丈夫だと思います(たぶん)。

Hires.Fixを使う場合はごく弱くかけてください。推奨は,R-ESRGAN 4x+,Hires steps=2,Denoising strength=0.15,Upscale by 2.0です。UltraSharpよりもR-ESRGANのほうが絵は壊れにくいと思います。

Well, I haven't made many photo-based models until now. R-Fantasy, R-SciFi, Good Old Days, and Twinkle are also realistic models, but they are specialized in one style and are not general-purpose base models. The next versions of R-Fantasy and R-SciFi will be derived from a common realistic model, and I am currently creating a base model for it. This model will not necessarily be used as is, but it shows the current direction.

さて,私はこれまで写真系のモデルはあまり作っていませんでした。R-FantasyやR-SciFi,Good Old DaysやTwinkleもリアル系のモデルではありますが,1つのスタイルに特化したもので,汎用のベースモデルではありませんでした。次期バージョンのR-FantasyとR-SciFiは共通のリアル系モデルから派生させる方針であり,現在,そのためのベースモデルを作成しています。これがそのまま使われるとは限りませんが,現状の方向性を示すのがこのモデルです。

-The drawings aim to be photorealistic, but may be more like 2.5D illustrations.

-The faces are mostly Lowropian, with not many Asian faces.

-The characters appear slightly younger (although this was not intentional).

・描画はフォトリアルを目指していますが,2.5Dイラスト的かもしれません

・顔はローロピアン中心でアジアンはあまり出てきません

・やや年齢が低めに出ます(意図したものではありませんが)