Merged 11 Character Loras from the same creator. No key words, just drop it in and see. These came from wyomee so you can use the key words from those SDXL models to push it in the direction you want to go
Description
Comments (14)
@jrrytng467 I'm going to try it out, but just curious, is this intended to replicate one singular character that has the traits of the 11 character LoRAs -or to cycle between those 11 characters, or something in between? Either way, it seems like a neat idea.
cycle between I hope. I pulled a name from one of the Loras to see if it would trigger that image but it didnt. Was just playing around, maybe see if it fixes a bad face model. I dont know
That's an interesting coincidence. Was just playing around with Frisky Dingo by prodajie tonight and loved everything about it, other than the facial quality I was getting. Amazing backgrounds, textures, lighting/contrast/depth, body/anatomy, everything looked great, but kept having trouble adjusting or improving the faces using various character LoRAs (sometime even stacked using multiple characters, which has been working well for many other models).
@jrrytng467 Just tried your LoRA and it wouldn't load for me and just gives the error message "Error: Could not load the lora model! Reason: Could not find text_encoder_2.text.projection in the given object!"
Not sure what would cause that. I know my software doesn't support DoRA, but ~218MB seems right for a LoRA, so probably not the issue. Were any of the original LoRAs something other than SDXL-based, like maybe Flux or regular non-XL SD?
@AFD_0 no, wyomees Loras are the ones I used. I used it on a few models i had to see the reaction. Well, It was my first one and was a fun experiment. If I could get other people to comment like you do, i would just take it down if i doesnt work
@AFD_0 letting you know. The last 2 images i posted, I created 60 I'm using to train my own with the guidance from GPT start to finish. Fingers crossed
@jrrytng467 Best of luck! And not sure if you're using local tools to train your LoRA or CivitAI, but in my experience (I've only used CivitAI to train so far), but using the standard/original SDXL checkpoint to train usually gives me better results as opposed to using a specific custom model (even if that's the model I want to use the LoRA with). I also get better results this way, even if I intend to use the LoRA for Pony or Illustrious checkpoint (rather than using a base Pony or Illu model to train a character LoRA). Really not sure why that is and your experience may differ, but just thought I'd mention that in case your results aren't as expected. I've also had pretty good results mixing source images of different (but similar) characters to create a singular character LoRA. Mixing 2 or more very different characters is a bit more unpredictable.
@AFD_0 Everything I do or use comes out of my stable diffusion. I run it local so nothing ever cost me anything and all the extensions built for SD is opensource so those are free as well.. When I built my new PC I upgraded to the newer version of SD and newer version of python which opens the door for so many more types of images I couldn't do before, I'm sure a lot came from the new card as well. Running a NVIDIA RX 5060 ti 16gb and I love it. I really want to start to create stuff and not as much merging. A lot to learn which means cutting back on play time :)
@AFD_0 One more thing, so you use buzz then to create stuff?
@jrrytng467 Yeah, at one point I probably had 30 or 40,000 buzz saved up and used some of that for training a few LoRAs. Whatever I can save up now I just use for making videos. All the images I create are 100% local using Easy Diffusion on my old 2080 Ti.
@AFD_0 well i just gave you all i have. I have no use for it and to me your probably more deserving for it. Nothing to get excited about
@jrrytng467 Thanks! Hopefully I'll be able to put it to use on another video from Test_ v4 or something. That's such a crazy-good model and I keep coming back to it constantly. Really glad you're experimenting with LoRAs for a bit, cause that's just giving us all a little more time to enjoy v4 :)
I dont understand why people put emojis of smiley face with tears or crying but wont comment on the image. Someone educate me.
I hate that too! Can never understand if it's just someone's way of saying "I really don't like that image" or "it makes me feel sad for some reason you'll never know" XD
Pretty sure it's just a "dislike" button for those that use it. Btw, I put the one on your comment, but just for lulz/irony.