WAN 2.2 Fantasy Theme LoRA
A fantasy-themed LoRA trained on artwork featuring elves, witches, and mystical creatures, with light NSFW elements.
Trigger Wordfantasylora
About
This LoRA focuses on a rich fantasy aesthetic with glowing environments, ethereal lighting, and mythical tones. Perfect for fantasy character portraits, enchanted landscapes, and magical storytelling visuals.
Usage Tips
If image quality degrades, adjust LoRA strength for the low noise model.
Increase high noise strength for stronger style alignment.
Decrease low noise strength to improve detail.
Experiment with different samplers and schedulers for unique looks.
Recommended Settings (ComfyUI)
High Noise Model
Sampler: LCM
Scheduler: Beta
Steps: 0–14
CFG: 3.5
Low Noise Model
Sampler: Euler
Scheduler: Beta
Steps: 14–24
CFG: 3.5
Example Prompt
“fantasylora, a young woman with silver hair that shifts between silver and deep sapphire, her skin like polished river stones, standing at the edge of a river glowing with liquid starlight. She holds a staff carved from a single moonbeam, its tip pulsing with blue light as shimmering bridges form over the water.”
Community Contribution
I’m working on future versions of this LoRA and would love your help improving it! If you create great renders using WAN 2.2, please upload them to help build a better training dataset.
→ Submit your images here
If you’d like to support future work and new LoRAs, you can:
Use my RunPod Cloud GPU referral link. RunPod offers a wide range of cloud compute options for all kinds of open-source AI projects. When you sign up using my link and deposit $10, we’ll both receive an extra $5 in RunPod credits. Every bit of support helps a lot with training these LoRAs!
https://get.runpod.io/3cbr613qjf6gJoin my Patreon. I share a variety of installation guides and one-click installers to simplify setting up local AI projects—including image and video generation, text-to-speech tools, and more.
https://www.patreon.com/cw/TheLocalLab
Enjoy the LoRA and share your creations!
Description
FAQ
Comments (8)
Please upload the safetensors files separately so that LoRA metadata is available via the civitAI API. Thank you for the LoRA.
Thanks for the comments but it appears that you can't upload two of the same files types with the same size which WAN 2.2 LoRAs tend to come with.
I keep getting this message "Duplicate file types
There are multiple files with the same type and size, please adjust your files", when I attempt to upload both.
Do you guys know how to bypass this or would I have to create an entirely separate post?
@TheLocalLab - I would simply look at others' uploads. I think the method everyone uses is to treat each base as a "version". You upload the low and name the version accordingly, then upload the high and name the version accordingly as well. I use musubi-tuner and train single LoRAs in dual mode so I have never had to upload a low and a high, but everyone is doing it essentially the same way. It's nice to be able to use civit's API for metadata retrieval in all my software.
Thank you for your time and I appreciate your willingness to address my concerns and of course I totally appreciate you sharing with me!
Thanks for this, it looks very nice.
As someone else wrote, in future please try to post the two safetensors separately, because it makes much easier for lora managers to retrieve the metadata
Thanks for the comments but it appears that you can't upload two of the same files types with the same size which WAN 2.2 LoRAs tend to come with.
I keep getting this message "Duplicate file types
There are multiple files with the same type and size, please adjust your files", when I attempt to upload both.
Do you guys know how to bypass this or would I have to create an entirely separate post?
Hi, the other user probably already answered about this point : the easiest way is to consider the high noise and low noise loras as versions of the model, so they stay under the same model page. This is how pretty much everyone is doing it, so I would say it became kind of standard
@nekotxt Ok I made two new versions, one for each model where I uploaded the models directly. Is this better?
Perfect, thank you!