YAFM: Yet Another F Merge

#포토리얼리즘

업로드 날짜: 2026. 6. 10. 오후 8:17
이용수: 182
리뷰: 훌륭함 (2)
샘플링 단계: 40
샘플링 방식: Heun
CFG 스케일: 7
제외 조건: cartoon, painting, illustration, (worst quality, low quality, bad quality, normal quality:2), nsfw, lowres, anatomical nonsense, artistic error, bad anatomy, interlocked fingers, extra fingers, text, artist name, signature, bad feet, extra toes, ugly, poorly drawn, censor, blurry, watermark, simple background, transparent background, old, oldest, glitch, deformed, mutated, disfigured, long body, bad hands, missing fingers, extra digit, fewer digits, cropped, very displeasing, sketch, jpeg artifacts, username, censored, bar_censor, mosaic_censor, conjoined, bad ai-generated, nsfw, long neck, skin blemishes, skin spots, acne, the wrong limb, error, black line, excess hands
권한: 이미지 생성 및 공유 허용
다른 사용자가 내 모델을 다운로드하도록 허용합니다
생성된 이미지의 상업적 이용 허용

설명

https://civitai.red/models/155444/yafm-yet-another-f-merge

Pro 18

YAFM: Yet Another F Merge 159

1.6k

Updated: Jun 11, 2025

base model photorealistic lora-friendly realism

Remix 10657-338187537-Heun-7-40-cybermix last.fp16.jpg

Remix 10645-978827126-Heun-7-40-cybermix last.fp16.jpg

Images hidden due to mature content settings

Download

1 variant available

fp16 SafeTensor

Half precision, best balance • 4.87 GB

Verified: 3 years ago

Details Type

Checkpoint Merge Stats

1,343 0 1 2 3 4 5 6 7 8 9 , 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9

Reviews

Very Positive (150)

Published

Oct 2, 2023

Base Model

SD 1.5

Hash

AIR

civitai: 155444 @ 174359 default creator card background decoration Followers - 12 12 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9

Likes - 159 159 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9

charless's Avatar charless

Joined May 3, 2023

License:

CreativeML Open RAIL-M Addendum

Hide

EDIT For V2: Continued testing/merging/going forward then backward then forward, tweaking etc. Felt like sharing a new version after all this time. I like it better than the old YAFM, that's the reason I'm sharing it.

For v2 illustration: I basicaly just used my wildcards to generate a group of pics. Grabbed everything (except 2 of them that where really awful) and posted all of them here. Made this to avoid cherrypicking. So if the pics aren't great, that's normal.

V1: I'll be quick, I had only 2 goals merging this thing:

it will be somewhat realistic

it will use my lora/lyco and not render awful (or not if you like them) Mars Mutants Imported from Verhoeven's Mind.

(I'm using it everyday so I guess it didn't fall short of my expectations)

It's been posted on civitai because someone asked for it. I'm not providing any update or anything. Just the merge as it is now.

So basically, it can render your favorite girl/celebs lora/lyco. I trained mine using the standard sd-v1-5-pruned-noema-fp16.safetensors and it's doing well. I don't know for others.

Now, all the Cheers and Kudos and Love and Kisses go to the original models creator of checkpoints listed here:

cyberrealistic_v31

https://civitai.com/models/15003?modelVersionId=89680

fantasticmix_v40

https://civitai.com/models/22402?modelVersionId=59385

almosthumanMix_v20

https://civitai.com/models/107132/almosthuman-mix

epicdream_lullaby

https://civitai.com/models/125189?modelVersionId=136739

photon_v1

https://civitai.com/models/84728?modelVersionId=90072

Concerning the illustrative pics:

It was quick gens, between 1-3 tries for each one. I didn't want to spend hours generating them. I don't see any point cherry picking images or posting inpainted one. It's all '1 click on generate and I wait' pics.

Concerning Prompts:

Positive:

Keep them simple, cf what I posted here. You can see some more examples of what i call simple in photon_v1 creator's gens.

(photo portrait:1.0), a woman wearing tennis top and shorts, smiling, brown hair, dark eyes, on a tennis court

Negative:

Just use photon's one as well. Nowaday models don't need a full novel quantity worth of words in prompts to gen correct pics.

cartoon, painting, illustration, (worst quality, low quality, normal quality:2)

I just added EasyNegative embedding on my everyday usage (plus more_details lora and the dynamic thresholding extention). I find results better fit my taste (I didn't use them for the illustrative pics though. I wanted them to be raw).

As usual for every checkpoint, use hires fix. And if you want awesome perfect faces from distance, use Adetailer (I don't, it's not working on my rig anymore. AMD powered):

https://github.com/Bing-su/adetailer

Of course if you like to write novels..... feel free. It's your usage, not mine. I don't mind what you do. Experiment, post and share, or not. :3

So... that's it I guess.

Hugs, or trout slaps.