Beach ~ ☼ ~ Sunset (Scenery)

IMPORTANT:

In the following text I give examples, on how to use the LoRA properly for different styles or scenes.

Please, click "like", if you find the LoRA useful and feel encouraged to write a feedback or "follow" me. If there's enough usage/feedback, I'll try to create some more LoRAs for sceneries, soon.

This LoRA is trained to create specific sunset scenery at a beach, including atmospheric lighting (see left or the test images). The trigger words (beach sunset, sandy beach, big red sun, ocean horizon, ambient sunset lighting) are added to the prompt as you activate the LoRA. You can adjust them depending on the image you want to create.

☼ STYLES ☼

Even though I trained it with various different styles of sunsets (photos, art, anime), it tends to create high quality sunsets. In order to change the styles accordingly, you simply can ADJUST the added TRIGGER prompts and REDUCE the LoRA's WEIGHT. Examples:

☼ for ANIME: anime-style drawn beach sunset, sandy anime beach, anime ocean horizon. LoRA weight: about 0,0 to 0,5.

☼ for COMIC: comic-style drawn beach sunset, sandy comic beach, comic ocean horizon. LoRA weight: about - 0,5 to 0,0 (less than - 0,5 might also work).

The LoRA creates similar styles of sunsets based on the trigger words. If you don't want to have this exact setting, you can also change or delete some of them to vary the outcome.

☼  SCENES  ☼

☼ more beach:

 Simply ADD a prompt text, like "large beach" or "broad beachline".

☼ no beach:

If you want to create a picture with a sunset out on the ocean, simply DELETE the triggers "beach sunset" and "sandy beach" and add some other specific prompts (e.g.: "out on the ocean" or "open water"), so that there won't be sand/beach, where the main objects or characters are located.

If this isn't enough, additionally REDUCE the WEIGHT of the LoRA to about 5,0 or less.

☼ no ocean horizon:

If you don't want to show the ocean on the horizon, for example when you want to create a picture, where there is water in the foreground and the mainland in the background, simply delete the trigger "ocean horizon" and add something else (e.g.: "large mainland with city skyline on the horizon").

If this isn't enough, additionally REDUCE the WEIGHT of the LoRA to about 5,0 or less.

☼ darker atmosphere:

If you wish to have rather a darker, moody atmosphere, you can CHANGE the trigger "ambient sunset lighting" accordingly (e.g.: "gloomy sunset lighting").

☼ big red sun:

Sadly, it appears to have been a mistake to add "big red sun" as one of the triggers. It seems, there haven't been enough pictures with big red sunsets in the training pool. I already tried creating pictures by changing it to "very big red sun" or "big crimson sun", but it comes out rather like a golden-orange sun (but getting slightly larger). You can try to REDUCE the WEIGHT of the LoRA to less than – 5,0 (minus!). But this does not seem to change any much. So, maybe you better ADD prompts to change the atmosphere (like above) in order to get overall more red color into the picture.

I think, the greatest problem is that AI generally has some difficulties in connecting colors to their specified counterparts (you most probably have noticed when trying to create a picture where a specific person has differently colored clothes or haircolor). That's why you actually have to use A LOT MORE color-descriptions in your prompts and not just rely on LoRA (especially the ones, which don't add any useful trigger words to the descriptions). No matter if the AI doesn't get it perfectly right. The more often you add such prompts, the more the AI will learn, the more precise the images will get.

So, even if you use and also if you create LoRAs: add DESCRIBING (trigger) words. In the long run, this will greatly improve the AI's accuracy.

☼  MODELS  ☼

According my tests, this LoRA should work with every base model. I already tested the following ones with solid results: Hassaku, RealisticVision V51, Anything V5, Midsummer, ReV Animated, SeasideMix, PixAI Anime Real. So, it should work fine with every model and size (portrait, normal, landscape).

I also uploaded some of the created test images, so you can see different examples of the outcome. See below...