Merged (base SDXL) model with my own LORAs to make photorealistic and fantasy portraits.
I suggest using in ComfyUI due to its faster speed than Automatic1111.
The WF I am using should be sewn into the files.
Support Paweł Tomczuk on Ko-fi! ❤️. ko-fi.com/aderek514 - Ko-fi ❤️ Where creators get support from fans through donations, memberships, shop sales and more! The original 'Buy Me a Coffee' Page.
Visit my DeviantArt page: Aderek - Hobbyist, Digital Artist | DeviantArt
Description
DPO merged with MIX XL V5 and loras.
FAQ
Comments (8)
i still have to compare this to v3, but I don't really remember so many people and robots. It seems all of my tests prompts forces a person a robot in battle armor in almost every image. Most of my things have no people, they are empty rooms. I did get some interesting results on some that did have people. But i'd rather it didn't fill something in that I didn't ask for.
It's strange that he behaves like that with you. I have no problems with generating what I want. It's a matter of what model strength you're using. Give an example of a prompt where the image is cluttered and I will try to find a solution. Thank you for testing my model.
This is the prompt block using Easy Diffusion (not sure if the generator makes the difference:
a close up of a bejeweled watch, clockwork using steel gears, brass gears, shiny, decorated metal, chinoiserie theme, a watch fit for a king the finest quality, many jewel bearings, exact details on the gear teeth, detailed gears.
Seed: 394902962,
Dimensions: 1600x1024,
Sampler: dpmpp_sde,
Inference Steps: 30,
Guidance Scale: 7,
Model: UPDATED REVIEW aderekSDXL_dpoV2,
Negative Prompt: ((((ugly)))), (((duplicate))), ((morbid)), ((mutilated)), [out of frame], extra fingers, mutated hands, ((poorly drawn hands)), ((poorly drawn face)), (((mutation))), (((deformed))), blurry, ((bad anatomy)), (((bad proportions))), ((extra limbs)), cloned face, (((disfigured))), gross proportions, (malformed limbs), ((missing arms)), ((missing legs)), (((extra arms))), (((extra legs))), (fused fingers), (too many fingers), (((long neck))), (((watermarks))),(((malformed gears))), (((poor gears))), lights in bottles, (((cross eyes))), (((wall eyes))), too many fingers,twins++++,over exposed,overexposure,(malformed hands, poorly drawn hands, mutated fingers:1.4)
Normally this creates a pocket watch of some kind. The negatives are often different still trying to force some models to give me human fingers. But I get a watch, its to test if it can do gears, many models can't.
But what I got was a lady holding a shield with wings on her head. Which isn't even close to what I was thinking. Right now i'm at a ratio of 1 out of 3 are usable and it followed my concept. But those other 2 are either a robot, a knight, a person, or some random character, I didn't ask for. I like the old version, I plan on testing it with the new block of prompts i've been using. I just don't remember it giving me all these extras.
i just compared this to v3 and it also seems to not follow prompts like I wanted a city I get a horse. If I wanted an apple it gave me an eagle. So i'm not sure how to use these. I do like the ones that I wanted to make. Just not sure why it adds its own thing for some prompts.
@frankmike helo, cfg 7.5, remove ALL negative prompt, i press three times and got free watches. only first stage from my wf, look to my post.
@Aderek514 I'll give it a shot again later on. I can't imagine why that negative would give so many robots... Ai is weird.
hey Aderek, I use your mixXL all the time, one of my favorite niche checkpoints. What is the difference in scope/purpose between mixXL and AderekXL?
Aderek SDXL based on clear SDXL DPO model and my own loras. MIX XL is mix of many models from civitai and my own loras. So, you get most diferent images betwen Aderek SDXL DPO and MixXL.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.