Full body metal chains with a collar and shackles, suitable for rendering BDSM scenes.
I'm still new to Stable Diffusion, so feedback on how to improve the model would be welcomed.
Description
FAQ
Comments (18)
Which model did you use? And parameters?
I think I used NightVisionXL and RealVisXL for the test images. There was nothing fancy about the parameters except that I used quite a low CFG value (3). Just don't forget to use the trigger words and try to adjust the strength according to the result.
@fallenmystic Thank you for your quick reply! I will try them. If there are sample prompts, it would be very helpful. Your lora looks nice
@Kirishimanoa4 Thanks! I didn't use any special words to depict the restraints except for the trigger words in my prompt. So I think you'll be fine as long as you include the correct trigger word, which is "slave_restraints outfit". (Not sure if the "outfit" part is needed but just to be sure.)
I failed to produce any acceptable image. Especially, the head is always cropped and the body is distorted. Please provide the parameters used to create your examples, including
- Positive an negative prompts
- Steps
- CFG scale
- Scheduler
- Seed
- Width and height
- Lora weight
Those examples I posted weren't of great quality and I constantly tweak those settings, so I don't think it'd mean much to share my settings.
But the problem you described sounds familiar to me. It seems that the issue is more frequent with higher LoRa weight (> 0.8) and with certain base models. From what I've tested, using NightVisionXL with the LoRa with 0.6-0.8 weight seems to work most of the time.
Hope this might help.
I just uploaded a few more images I rendered using the LoRa. Some of them contain generation data, so I thought you might find them helpful.
Thanks, but I'm still missing one setting : What's <lora:wegg:1.0> ? I've not found any lora named wegg.
@Globule It's just a style LoRa I use for some of the renders. You can safely ignore that.
Across several tries, I haven't had an image in a batch turn out with the shackles or chains visible, after including things like 'watch' and 'bracelet' in the negatives. Images that contain more than wrists don't show metal at all. Any tips on how to get results?
In that case, I highly suspect something might be wrong with your setup, like missing or incorrect trigger words.
My Lora may sometimes depict restraints incorrectly (especially chains) or deform the body. Still, I've never seen it fail to add restraints entirely with a high enough weight value (e.g. 1.0) from my experience of trying it with a few different checkpoints.
Try checking if you have downloaded the model correctly and are using the right trigger words.
Any possibility we could get a non xl version too?
I'm sorry, but I don't use SD 1.5 any more so it's unlikely I'll make a version for it.
@fallenmystic Pity. But not worries.
what is the trigger???
It's in the model description "slave_restraints outfit". I'm not sure if the "outfit" part is necessary though.
Just let you guys know, this LoRa works amazingly with AnimagineXL3.1
I'm glad to know that my LoRa is working fine with a style that I hadn't even initially intended. Thanks!