If you like my contributions to the community, please consider Buying Me A Coffee or check out my other sites on AllMyLinks
Differences between versions:
v5.1 - Adds more custom training to v5 while remaining very flexible.
v5 - Recommended version. Flexible with more variety of faces, bodies and settings.
v6 - Less flexible and more of a specific look to the faces and bodies. It's had some custom training done to it, and while it still works well, it was created more for my personal work than for public release. It leans more towards studio backdrops and my custom training.
About RyanBlend:
The RyanBlend models are designed primarily to create photorealistic images of women with a high-end glamour/boudoir photography style, with a very subtle hyper-realistic anime style for facial features and body proportions. It's NSFW friendly, but doesn't have to be used that way.
It's excellent at creating closeup skin texture and high quality faces without using a face restoration model.
I've been a professional photographer since the mid-2000's and specialize in a high-end retouched look for glamour and erotic photography. So I try to make my RyanBlend models close to my own photography and editing style.
The base of the merge consists of:
Deliberate, phDMix, and Realistic Vision v2
...plus a little:
Analog Madness, HARDblend, URPM, and Perfect World
v5.1 and v6 also have a lot of my own custom appearances for faces and bodies.
I recommend using VAE ft mse 840000
Description
v6 adds 10,000-15,000 training steps of my own custom appearances for faces and bodies. Because of the training, backgrounds often lean towards a studio backdrop. But it's flexible enough to obey most background/environment prompts you may give it.
FAQ
Comments (7)
Pruned Model fp16 for v5 is 5.28GB too!
Could you please fix it? Thanks.
oops the pruned one was right but the 5.28gb one was just labeled wrong. fixed now
model aside, i can tell you are a professional from the example you hand picked :D
;D
This looks pretty solid. I've also been a model photographer for just slightly shorter amount of time than you have. I believe the influence often comes through in my generations as well.
A question I'd have about the model here... it does look like a lot of the expressions in the images you chose for the showcase, they're very similar. Does the model, to your knowledge, often default to that sort of facial expression.... or is this just a matter of the prompting style you used and also possibly the selection bias in choosing which images to feature?
I'm sure it would be no problem either way, as expressions are easy enough to change with prompts. Just sometimes, I don't always remember to make that extra effort, especially if the prompt itself is getting long or complex.
for RyanBlend v6, I trained 4 LoRAs and merged them with RyanBlend v5. So v6 is heavily influenced by the LoRAs and the facial expressions in them. For the LoRAs, accurate likeness was more important to me than flexibility, hence the similar expressions. v5 has no custom training done to it, it's just a merge of all the models listed above, so I'm sure most people would get more usage out of v5 than v6.
@rjox That's a good explanation, I understand what you mean exactly. Thank you!