Type | |
Stats | 882 |
Reviews | (184) |
Published | Oct 30, 2023 |
Base Model | |
Usage Tips | Clip Skip: 1 |
Trigger Words | d3qck artstyle |
Hash | AutoV2 E25C520BF1 |
Hi again and welcome to my experimental Dominic Qwek's Art Style XL LORA.
First comes first and this may be personal but this is a special time since anyone who knows me already should point that I use to avoid training on one single artist on his own style and I really prefer being more creative with the AI, if possible. Being said this, this doesn't mean that I don't find enjoyment training like this, I surely do and it's a thing that privately I do in my free time or for the sake of learning but giving the chance, I prefer mixing styles and/or artists with synthetic data in the way too, yeah.
So yeah, now to the main thing. I choose Dominic Qwek cause I loved the monster/alien sculpts he did in the past but I loved even more his last style from 2021-2023 because I find his compositions absolutely stunning.
His works is oriented to female portrait (or faceless sometimes) in a marblish style with gold ornaments but at the same time he "fuses" them with the background making an impossible statuesque surrealism style. As I say, impressive. If you have time, check his works. Really underrated figure in the scene.
This LORA has deep (but complex) flexibility but you will have to play with the CFG scale as I pointed on my latest guide to focus more on the input data than settings and so on.
I have been doing testings on different scenes and works well creating different concepts as interiors, rooms, structures but it shines on portraits (cause the dataset), creatures and so on.
Main token is d3qck artstyle . This time I trained in a different way since I wanted to test things, why not? So you can prompt as "d3qck artstyle restoftheprompthere".
You can refer to the original images uploaded to see the prompts (I used ComfyUI and different models but the prompt was jus that). I have started to train in a different way than before (cause just learning purposes and try to get better outputs) and making more "complex" prompts in the inferencing but preserving the main style.
Important: This LORA has subtokens and those are "over black background", "over white background" and "fuse/fused with/to". The more subtokens more Qwek's aesthetic so use with caution (even backgrounds).
Obvious thing: Of course if you use marble and gold the style trigger will get activated
but the LORA is designed so you don't have to in 80-90% of time. Feel free to experiment.
How to use this LORA?
*************************
You can refer to the sample images I upload in the post but:
- For characters d3qck artstyle portrait of concept/character over white/black background (optional this last)
- For landscape and objects d3qck artstyle concept or a "concept" over white/black background (optional this last)
- For fused aiming refer to the post to see the examples.
- Negatives as cartoon/anime will help also the LORA.
- It's important that because how the captions were done, aspect ratio is VERY important to succeed on the creation.
Which are the best values?
*******************************
- You can go from 0.65 onwards till 0.8 (recommended) and even 1 for more "Qwek's artstyle" but I suggest to stay in the range of 0.65-0.8. Depending on the model and CFG Scale you can alter the numbers.
- Detailed further info: Values 0.6 onwards till 1 you are free to go with the recommendation of 0.6 to 0.8. For example for more "spacial" faces you will need to go 0.7 onwards while for more "normal" faces (while preserving style) you should use 0.6-0.69 more or less. Experiment.
Which models are good for this LORA?
********************************************
- My recommendation to do the testings at first is base model (1.0 XL Base) and after you can go Crystal Clear XL, Juggernaut XL, etc.
All the images were done with Stable Diffusion base model so you can do similar images with prompts.
IMPORTANT!
I wrote an article here and it's like my finetunning diary for the ones interested with releases upcoming, trainings done, exclusive patreon models previews, personal thoughts, etc. So if you want to see what I'm planning or what I'm into, feel free to take a look.
As always thanks to everyone reviewing, giving feedback , commenting and supportin us in Patreon. Keep supporting us so we can keep creating and delivering quality content.
David.