This is a DreamArtist Textual Inversion Style Embedding trained on a beautiful AI image (my own) of a tropical South Pacific island beach. You can get stunning ocean waves, beach and palm tree effects in your images by calling Style-Moana in the positive prompt, and if you want to temper the effect, also call Style-Moana-neg in the negative prompt.
If you'd like the effect to come through stronger (some models need it), try bumping up the emphasis on Style-Moana - see the images below.
Please consider joining my Patreon! Advanced SD tutorials, settings explanations, adult-art, from a female content creator (me!) patreon.com/theally
Description
FAQ
Comments (14)
Just a thought can someone with experience in creating embeddings or checkpoints try to train a model with high-res pictures 2048px or more of hands, fingers and eyes to solve that problem of bad mutated hands and ugly eyes once and for all?
It's your lucky day! https://huggingface.co/datasets/Nerfgun3/bad_prompt (works, somewhat), but truly, you're better off just inpainting the bad details.
If you use high resolution pics of hands, then it will give you back images of hands. The issue comes from hands in medium and long shots are low resolution, while having the context of the scene you are building with the image.
Looks awesome! Are you thinking of possibly creating an SD 2.1 based embedding as well? Otherwise it is not compatible with my other embedding that I'm using on 2.1 based models.
Hey! This started as a 2.x embedding, but I have had THE WORST luck training those - I'll keep trying. Eventually I'd like to port all my embeds over!
Oh I see. Well I hope your luck changes soon because I'd love to see this on animal art. It would be perfect. Good luck!
Awesome! Is it a LORA trained embedding? Asking to keep my embedding list organised!
I think I'll categorise this type of asset as "Contextual style" alongside with style-kitchen, as it describes a scene.
This is trained with DreamArtist extension for Web UI, thanks!
Loving all your DreamArtist Textual Inversions. What sort of learning rate and steps do you typically use? I've been trying my hand at it and don't seem to get great results. I have to disable reconstruction due to only having 12gb of VRAM but I suspect it's the rate playing a part in my poor results.
I'd kill for 12gb :) I have 10! I can train with reconstruction if I set my source image to 384x384, but I haven't seen an advantage to doing so. I am using the default LR, sometimes dropping it after the first 500-1000 steps. I run it for 100k steps, then go through and test a bunch to see which has the most consistent effect! Thanks :)
Thanks so much for the reply! Good to know it's not the reconstruction setting and more so likely my source images and number of steps I'm using. Cheers!
It looks very good, but when I try it myself I don't get similar results. What kind of model do I have to load to achieve this style?
The v1-5-pruned-emaonly.ckpt or something completely different?
Thank you for the help
Ah yeah, 1.4 and 1.5 are the base SD models, and they're not great, to be honest. They're very rigid in their styles, and hard to prompt. If you try downloading my "TheAlly's Mix II: Churned", or Elldreth's Vivid mix, you'll get much better results!
Let me know if you need help with anything!
Thank you so much for the quick response!!
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.