The Dark Atmospheric LoRA was trained to produce a style characterized by shadow and mist with a slightly diffuse effect.
Description
FAQ
Comments (7)
TBH this lora is subtle with z-image. i generated 3 examples with and without lora. the difference is hard to tell. I feel that base model already does a good job at this type of scene and it just depends on prompting. I posted 3 examples i made using a prompt from Gemini since i am lazy to come up with my own. Both of each image is with same seed and settings, only change is using lora or bypassing lora. i used .95 strength on the lora.
Thanks for your feedback. In my testing with ComfyUI, I saw a noticeable difference with and without the LoRA. It may have been the prompts I used. Not sure what to suggest. I produced and tested LoRAs at various steps and thought the 4k worked best. The differences I see are similar to testing with the other models for which I produced the LoRAs.
That mistiness and gloom around the your wolf image is typical of the kind of effect I expect to see. Sometimes the base models already do a fine job, as you commented. I've skipped producing LoRA for some models as I felt I couldn't improve on them.
@CreativeEdge yea wolf one was most obvious I think. Maybe if someone doesn't prompt well for it, then the Lora has a better affect for them. Or Lora makes it easier to achieve the desired result. i find this model so far has been able to do almost anything. I've been able to reproduce most art styles and lighting just from trial and error messing with prompting. but prompt engineering takes time to get right for each unique image. I tend to start with AI made prompt then tweak it a lot until i get the desired effect. I am not discrediting your Lora. It's just when i seen the Lora i was thinking i can likely do that with the right prompts with no Lora. but like i said it probably is helpful for people to save time.
@zengrath Couldn't agree more. The right prompt can achieve much of what many LoRAs aim to achieve. If nothing else, the LoRA adds some additional style. Then its up to you whether your satisfied with the model.
@zengrath I will just add that long, complex prompts can detract from the purpose of the LoRA. Simpler prompts are often more capable of drawing out its effect. However, as you have mentioned, a sufficiently detailed prompt may produce the sort of results you seek anyway. So its a case of how you prefer to prompt and what result you're aiming for.
@zengrath Try this LoRA at a strength of <= 0.5 and play around with the sampler and scheduler. I really love this style and this LoRA for Flux. I've been testing it out for Z Image and haven't had as much success. I have to agree with the author - Z Image is already really good with photography and lighting. I also have a LoRA that's focused on dark scenes and photography. I love where my LoRA ended up...after training it like 6 times...But the effect is sometimes subtle. Other times it's rather dramatic. I made a post with more detail you can read at the bottom of it, the images should be ordered so that the first has no LoRA at all, the next series of images use this Dark Atmospheric LoRA and then the last few use my AfterDark LoRA. I left some notes about key words in prompts that may be helpful. https://civitai.com/posts/24912855