Flux V2.

V2 already, you ask? well don't expect another one for a long time for Flux dev if ever

Better realistic style when needed / better cum

LoRa strength 0.7 -1.

.

.

Flux V1

LoRa strength 0.7 - 1

Flux Dev works best on naturalistic language, but can be specified with tags to clearer detail

V7- Chained2

1. Trained on WD14 style captions, limited to solo frontal images of men, like all previous models, but within this restriction, caption like crazy, my personally trained captions are listed in the tags

2. Cfg is variable but try higher first, LoRa strength start with 1, then work down to 0.65

3. "Hyper penis" is the style tag, then try "a naked man" then the penis type, "erect penis" or "hanging penis"

4. sometimes like to make solo images of penises, remedy this by the "full body" tag or by specifying what limbs are missing

5. Maybe the most style flexible

V6 - Chained

• use LoRa strength 1 and cfg 5 as a start.

• Works nearly the same as Final V2 but was trained a little softer, This has allowed it to be more flexible, have better prompt understanding, clearer faces, clearer images in general, better style flexibility, more dynamic images, better understanding of limbs, and sometimes better penises

• Might be less consistent and less detailed than its immediate ancestor but still more consistent than its other ancestors besides. Backgrounds are a little better than "FinalV2"

• Named so to represent the restricted and chained up nature of this model and all of these models due to the limitations that online LoRa training spaces and my own GPU have placed on this LoRa due to the relatively enormous sample size I use to train. I fear a 4090 may be the only cure.

FinalV2

• "Hyper penis" to prompt the token style tag, "a naked man" works as well and is maybe even better, then specify the penis type, "erect penis", "hanging penis", "foreskin" and then the style tag, "bara style", "hyper body style", "anime style", "skinny"

• "hyper body style", "bara style", "anime style", "skinny", "foreskin", "realistic style", "hairy" are new tags I would like to bring attention to, foreskin and realistic style work but are a little unreliable

• Try LoRa strength 1 as a start work down to 0.7

• More consistent than the original Final version but also possibly a little more

  overtrained

• ...backgrounds seems hard to prompt..........

• mostly consistent

V4

• Male is the trigger world

• Kind of undertrained

• LoRa strength 1

V3

• More balanced weights in comparison to its ancestors

• Balanced anime/realistic style

• First time to feature "full body" prompt

• Mostly Consistent

• CFG 1 - 0.7

V2

• Heavy weights (overtrained but still good)

• Medium penises

• Anime style

• Normalish body sizes

• Consistent

• Use Low CFG in general for this model

• LoRa strength 1~0.7 but I think lower is better

• "a naked man" is probably the most important tag to add, use it by default or if you're getting unsatisfactory results.

• try "uncensored" if still getting wonky results

• If using "hanging penis" try "handsome face" or something to indicate their face, otherwise results are pretty inconsistent

• Was trained on "foreskin" but it seems to be really bad at making them, try it anyway it might work

• Try "anime style" or "realistic, realism" for big changes if using base SDXL model :)

V1

• Soft weights (undertrained but can make more life like images)

• Medium penises

• Realistic style

• Hypermale bodysized more focused

• Inconsistent

• LoRa strength 1 and CFG 10

checkout my suggestion resources (these models are fantastic)

Probably best to avoid naturalistic language with theses model try the tags instead except V7 and V2

Select images have been inpainted to restore faces

Post your images!