Sign In

Zab's Male Mix

53
455
1k
35
Verified:
SafeTensor
Type
LoRA
Stats
197
644
2.6k
Reviews
Published
Sep 19, 2024
Base Model
Flux.1 D
Usage Tips
Clip Skip: 1
Strength: 1
Hash
AutoV2
2C8B0B12E7
default creator card background decoration
Zabadu's Avatar
Zabadu
The FLUX.1 [dev] Model is licensed by Black Forest Labs. Inc. under the FLUX.1 [dev] Non-Commercial License. Copyright Black Forest Labs. Inc.
IN NO EVENT SHALL BLACK FOREST LABS, INC. BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL.

I decided to give it one last try, but this will end with V1. The sample images were heavily cherry picked, as the model still gave inconsistent results for the femboy/feminine and androgynous images. You will likely need to add the word "Male" to the prompt, to ensure it outputs a male, especially for 'femboy' prompts.

After v1, I still don't really care for Flux. Yes it's more consistent with text and hands, and I will admit the quality is nicer than I first thought, but it's still just not capturing everything that I was trying to train. Perhaps if it was separate training data specifically focused one the feminine and androgynous male types it would've worked better, but other models handled thist data set well. And, again, the size of the Flux models just makes them practically useless to me, because I can't run them locally, and effeciently, to finetune or better train them.

Flux Model

I haven't been able to do much with Flux, but I figured every model could use a little more male body types in the mix.

Since I can't run Flux decently on my personal computer this was trained with CivitAI, so I couldn't use my full dataset, and it didn't capture all of the body types properly.

Not really happy with the outcome myself, but feel free to try it out and see if it works for your own needs.

v0.1 was my first attempt, using an uncurated dataset with minimal captions. The results were okay, but not very consistent. It mostly leans toward a digital art style, especially as prompts get more complex.

v0.5 had more curation and better captions in the dataset, but it still seems like it didn't capture all of the proper body types. It should be able to handle both artistic and realistic/photo styles though.

v1.0 the sample images weren't good, so I don't know where it stands. Doubtful there was any real improvement but I figured I'd give it another try.

v2.0 I only made this because I made a new dataset for the MaleMix model I am working on for illustrious and figured I'd see if it did anything for Flux.

I hope these models can at least help some people out there, but I just can't keep pursuing Flux. I will wait for something better, until then I'll stick with Pony.