weight: -8.0 to 8.0 ( or way more or less )
positive: large DOF (sharp background)
negative: narrow DOF (blurry background)
UPDATE:
The original file had an sdxl meta tag that makes automatic1111 web ui only show it for sdxl. I have uploaded one with the correct meta tag. If you are having issues, download and replace the previous one with the new one and refresh your network modules.
Weight depends on starting photo, if it already has a shallow depth of field, you may need to go pretty high to remove it (10.0 to 15.0), but -5.0 to 5.0 works for most images. The image stays pretty stable even at extreme values, though it will change the composition.
Description
No trigger word needed
FAQ
Comments (34)
For some reason, LORA is not recognized in "my environment". Another LORA recognizes it, so it must be compatibility or something.
Not only the author, but the usual dandy has posted a comparison chart, so others should be able to use it. Thus, it's a problem with my environment.
Too bad, because it looks interesting. I'm looking forward to the next slider.
What environment are you using? Is it not doing anything or throwing an error? My weight key names are a little unconventional, but works in most platforms. I need to fix that.
AAAAH, I accidently put an sdxl flag in the meta. So auto1111 will think it is for sdxl and wont show it. Ill reupload in just a second. THANK YOU! for this comment, I would not have noticed.
Ok, I reuploaded the file. If you redownload, replace the file, and refresh network modules, it should work, let me know if you have any issues.
@Ostris Never mind. I would be honored if I could help correct any mistakes.
There seems to be some problem with SD recognizing this lora. When I load a 1.5 checkpoint in A1111, this lora does not appear among the others. But when I load an SDXL checkpoint, it does appear. However, since this is a 1.5 lora, it does not work with SDXL. So to use it I must either manually type it in the prompt or switch to SDXL, select the lora, and then switch back to 1.5. Strange!
Yeah, sorry. It had bad meta. I just replaced the file with one with a 1.5 tag. The other one was set to sdxl. If you redownload and replace, and refresh the network modules, it should work now
@ostris hi there, i just download the latest version, it didn't show when i load sdxl checkpoint, but like when i switch to 1.5 checkpoint, it can be showed in the lora network modules, how can be that , it likes reverse though hahaha...
@cyptoRabbit it is a 1.5 LoRA
@ostris ok make sence! thx~
Funny, the background depth is supposed to change, but the character's costume changes directly!
More exposure!
They say it is more exposed and more graded. They may also be given wings. They may be Red Bull drinkers.
When Weight- is set, the background is blurred without any problem, but when Weight:10 is set, the background becomes a monochromatic background and the character's shape is changed (not deformed or "Transmogrify"). I think this is a failure of LORA, as it is a function that is probably not intended for LORA.
However, I welcome this change. The chaos invites trial and error, and we can get untested results.
Try Weight:10 with "Indomitable Creativity" and you will discover something new and fascinating without being shocked when you use it for WAIFU.
Wrong usage (in ANIMEModel)
Weight: up to 5, which can be used without hesitation
Weight:up to 10, which erodes the picture but makes it interesting
The upper limit is Weight:20, which does not destroy the picture.
10~20 is difficult to strike a balance between the interestingness of the picture and the pattern due to the erosion of LORA.
Can the same thing be done for SD XL?
I have one trained already. I just need to do some testing on it and generate some example photos. Just need to find time.
@Ostris It would be so cool in SDXL.
@ostris how far is your testing? I'd love to have this slider for SDXL!
can you make a human-to-animal slider? Like the more positive, the more the person looks like an anthropomorphic animal
I have tried previously and has some issues, however, I have some new methods, I'll try it again
Great Lora ! Thank you, very helpful
Hi Ostris -- I've been loving your slider LORAs. As use a number of them all the time in my renders nowadays.
I wanted to throw out one idea for a future slider, if you were open to it: an eye direction slider, to provide guidance for SD to make a subject's eyes look left or right (independent of head orientation). Perhaps a second one for looking up/down as well.
Cheers, and regardless, thanks for the quality work you've shared to date. Thank you!
I have been working on this and head direction as well. Both are difficult issues to crack. Eye direction is very difficult, especially in non close up images as most of that happens in the VAE and is not controllable within the the latent space. I will keep working on it and hope to solve it soon.
i ve used the sdxl base model like base 1.0, refiner 1.0, but the lora still did not work like i 've slide to '8' tho.
This LORA only works on SD 1.5
Hello, N00b here, how are these installed? I'm using auto111, I saved the file in a custom folder I made in the LoRa directory. rebooted the UI and nothing. clicked on it so it would make a prompt entry and generated a set with it in the prompt field and I get nothing. I'm sure its an ID10t error on my part but figured I'd ask.
As long as the lora slider is placed within the lora folder it should work. Are you making sure you are changing the weight number in the prompt? It's the number after the colon (ex. <lora:depth_of_field_slider_v1:-5>). For this lora, negative numbers will ensure the subject is in focus and the background out of focus. Higher positive numbers will ensure everything in the shot is in focus.
@djanks No, I entered no parameters, I was working on the theory that if it was in the LoRa folder it would somehow put a slider tool into the interface and changes could be worked using that. I have custom folder for it but that custom folder is in the LoRa directory, should I just delete the custom and store it naked in the LoRa directory?
@MxProteus Ah I see. Yeah, even though it's called a "slider Lora" it doesn't actually place a visual slider for you to click and drag in the user interface of Stable Diffusion. You have to manually change the number in the prompt itself from either positive or negative to "slide" the weight of the lora to your desired look.
@MxProteus And to answer your other question. No, your custom folder should not affect your loras so long as the custom folder or folder(s) are inside the main lora folder itself. I myself have plenty of sub folders to help organize my loras.
@djanks Cool, I appreciate the info and thanks for the details on folder hierarchy, I was a bit curious about it. I'll play with it and see what it does for me.
Hm, I couldn't find the SDXL version, but I thought it should be available?
tested it on Fooocus and it didn't work on it :D
I am new to this slider business. Is there any guidance on how to train such a slider?
Hello, your skills are truly impressive. I have used several of your creations, and the results have been excellent. I wanted to ask if you have considered making tools for “PONY” in the future? I am very much looking forward to you creating a “Lora” to enhance the details of “PONY.” This message was translated using translation software, so there may be inaccuracies.
does the denoise setting affect the rate of the sliding?
Details
Files
depth_of_field_slider_v1.safetensors
Mirrors
depth_of_field_slider_v1.safetensors
depth_of_field_slider_v1.safetensors
depth_of_field_slider.safetensors
depth_of_field_slider_v1.safetensors
depth_of_field_slider_v1.safetensors
景深滑块_v1.safetensors
depth_of_field_slider_v1.safetensors
depth_of_field_slider_v1.safetensors
lradepthfield.safetensors
depth_of_field_slider_v1.safetensors
depth_of_field_slider_v1.safetensors
depth_of_field_slider_v1.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.