XaiJu
cwhk
cwhk

patreon


[Model Test] Vane (Assorted)

Hey all,

A few posts back I mentioned I would be changing my workflow for generating images. I did change my workflow, but I also decided I wanted to change the diffusion model as well. I have been using one specific ILXL checkpoint for a long time and while it makes great male bodies, the creator has not updated the ILXL version, which means the training data is OUTDATED (e.g., lack of new concepts, not as good anatomy, outdated character base).

Meanwhile, a new base model, a complete retrain of the original Illustrious, was released. It not only has an updated dataset, but the author figured out a novel way to prevent concept bleeding (which is a huge issue in the model I was using before). This allows me to create certain images far more easily without the use of constant inpainting.

Because this is a new model, the male body looks a bit different. I spent the last two days testing prompts to figure out which combination would generate me bodies similar to the old model and I think I found a combination that does do that (hopefully!)

Here are some example pictures I generated, where I tested different brightness, lighting and gamma settings, as well as a different art style.

[Model Test] Vane (Assorted) [Model Test] Vane (Assorted) [Model Test] Vane (Assorted) [Model Test] Vane (Assorted) [Model Test] Vane (Assorted) [Model Test] Vane (Assorted) [Model Test] Vane (Assorted)

More Creators