Another LoRA XL model – Vextria for you.
I think you’ll like it! The model is quite flexible and easy to use, but you can further guide it with prompts like “sketch,” “colorful,” or “detailed,” depending on the elements you want to emphasize. Even without additional prompts, it works well in most images.
Trained and tested on: sd_xl_base_1.0
Have fun!
Description
FAQ
Comments (6)
I tried everything to get this LoRA to work, even with the same prompts and Guidance:7, Steps:35, Sampler:DPM++ 2M. Even tried different Clip skips.
Nevertheless the generated images aren't even close to those one you posted. Any further hints? Thank you.
I have no idea what might be causing this. Sometimes I receive feedback that my LoRA isn’t working well, even though the descriptions and settings remain the same. I believe the issue could be a different image generator or a different version of it.
I train and test my LoRAs locally. For XL testing, I most often use Automatic1111 with the following setup:
Version: v1.10.1
Python: 3.10.9
Torch: 2.0.1+cu118
Xformers: 0.0.20
Gradio: 3.41.2
GPU: RTX 3090
NVIDIA Studio driver: 566.36
I usually test XL LoRAs on the base model sd_xl_base1.0. This model performed quite stably during tests, without noticeable errors, and showed considerable flexibility. It also didn't require many attempts to generate the sample images I shared.
I’m sorry that it’s not working well for you. I guess I should start testing my LoRAs on other generators more often, not just on Automatic1111.
Thank you for your quick and concrete answer.
Indeed, I am using
a different Version of Python (3.11.8)
a different Version of Torch (2.3.0+cu121)
a different Version of Xformers (0.0.26.post1)
and NVIDIA Driver: 572.16 on a RTX 4080
So, I think you don't owe any apology. It's just a difference in our setups. Plus I am using ComfyUI instead of A1111. Your published LoRAs for Flux always worked well, so I guess this is another materpiece 😉
@Standspurfahrer I just posted a few that I generated with ComfyUI with the JSON embedded. Give one of those a try. 👍 Good luck 🍀
@Daalis Thank you so much. Switching to sampler "euler" with a clip skip of 2 did the trick.
Pictures will follow! @grafmix421 Sorry to bother you and thanks for your support.
@Standspurfahrer Fantastic. Glad it worked for you. 👍❤️
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.