hello everyone, finally decided to start sharing some of my models here and decided to start with my current semireal illustrious base model as i dont see a whole lot of illu models that go towards this direction, for the sake of simplicity and making it easier for the average joe to replicate the outputs i made all the examples using only face adetailer and fairly simple prompts and no highres fix/upscales of sorts using webui tho the model is capable of way more than that

Prompting

Positives: i suggest starting with "masterpiece, best quality" and going from there as the model does ok with any prompts really, the only thing of notice is that the longer and more complex your prompts get i feel the more it strays away from realism so you may want to use "realistic" or "photorealistic" to drag it back to that direction, you will notice in the examples i uploaded i used a few different quality tags across them including one example with no possitives and negatives quality tags for the sake of science that ended coming out ok

Negatives: ive been using "lowres, (worst quality, bad quality:1.2), bad anatomy, sketch, jpeg artifacts, username, censored,extra digit" for pretty much all my models nowdays, you may want to try using some negative embeddings as i had great results with a few of them too (namely Stable_Yogis_PDXL_Negatives-neg) , but the model should do fine with whatever negatives you are used to working with

Final thoughts

hope the model is of use to some of you, its pretty good for cosplay and things of the sorts but generally speaking its just a easy to use 2.5d model, ive encountered an issue while messing around with it in comfy where upscalers where stretching the faces of outputs but i assume its related to my workflow being scuffed as i didnt have those issues in webui but its something worth noting let me know if you run into the same issue

Kindly help by liking, it helps a lot with being noticed

描述:

v2 finally here after a lot of headache and trouble shooting

should have a little bit better anatomy and hopefully cause less mangled fingers, better shading/lighting imo and now it should also respond a bit better to loras tho style lora support is still wonky (i notice style loras make the model look way too anime so you may want to use the tags "realistic" or "photorealistic" to balance things out if needed)

训练词语: