getting UnboundLocalError: local variable 'h' referenced before assignment

[deathtome1998]

I'm trying to make some different expressions for the same character in the img2img batch tab. using ADetailer and scripts prompts from file or textbox, with my 25 total expressions. as well as the ultimate SD upscale extension with the 4x-ultrasharp model. which I maybe should be leaving off the upscale for now, but eh. I keep ending up getting this error after starting my generations, UnboundLocalError: local variable 'h' referenced before assignment. I'm not sure what it means or how to fix it, so any help would be appreciated. in case it matters what my current model is it's ponyRealism_v21inpaint11VAE. and I made the generation I'm trying to img2img with ponyRealism_v21MainVAE, I also attached a text file of the full error traceback.
error code.txt

[w-e-w]

when using DPM++ 2M SDE make sure your Sampling steps is >= 2

[deathtome1998]

what? I changed the setting you suggested changing in the web-ui's setting tab for img2img. which did let me move past the error I was initially having before. now I'm getting a different error. I don't know why you're saying it's still set to 0 or 1 when I literally turned on the setting that you said would be helpful "but "enabling" the do exactly the amount of steps the slider specifies will disable denoising scales sampling steps". I guess you misunderstood me or something I don't know.

[w-e-w]

you siad above that "I have DPM++ 2M SDE with karras schedule type, 30 steps, a CFG of 6, and denoising of 0."

---

TLDR: to avoid the issue cause by denoising of 0
just use some other sampler and scheduler that doesn't throws an error
it doesn't matter which one you use as denoising of 0 because you're literally asking you to do nothing to the image

---

the issue is happening because there is some Edge case bugs in how low Samplers and schedulers are implemented
in that if they are presented with these values they will fail to work

you want unknowingly trickling it one of these edge cases

you have you were sampling step value set to 1 due to, even though you think it was 30 due to With img2img, do exactly the amount of steps the slider specifies,
enabling or disabling the setting only changes your ability to control the values

analogy: like changing your keyboard, it only affects your typing experience but it doesn't affect what's actually being typed
it only helps you to know what your inputs are but does not, actually changes anything about what happens next
if your input would cause error it will still cause error

you can see that in my XY grid DPM++ M2 SDE produces a black image (Error) on all combinations when denoising strength is 0

in the case of DPM++ M2 SDE it will will error when Step i<= 1 or Denoising strength = 0

it will trigger different errors based on the input values
when setp size is <= 1 -> UnboundLocalError: local variable 'h' referenced before assignment

which was the case for you as you annoyingly have you set setp set to 1
after you enabled With img2img, do exactly the amount of steps the slider specifies you know avoided setp size is <= 1 but you still have Denoising strength = 0 which triggers a different error input.numel() == 0

and oversimplification of what stable diffusion is and what denoising value does is that

stable diffusion is a "denoiser" it remove noise from image
it basically looks at an image and remove what it considered as noise returning you an image
for txt2img it basically hallucinates that there is a image in a given random noise generated by the seed
it adds noise to the input image an attempt to remove the noise in this process generating a new image

how much noise is added and removed is controlled by the denoising strength (a value from 0 to 1)
when denoising strength is set to 1, it basically doesn't care about your origin input image
on the other hand if you set denoising strength to 0 you're basically telling it to give you the image as is and there is no noise to add or remove you want it to do nothing

how much noise is added and removed is related to the Sampling and Scheduling method
but if you set the denoising strength to 0, suddenly the method use is completely irrelevant because you just wanted to do nothing

denoising strength 0 = give me the origin image without doing anything
the only issue is that some sampler / scheduler are implemented with bugs that prevent it from doing so
but since you just want the origin image it doesn't matter which one you use
just use any other sampler / scheduler that don't have this issue

---

analogy
imagine sampler / schedules are stores you can go in the stores and do stuff
some stores would have rules that you must have a minimum minimum consumption while others don't
when you're set the denoising strength to 0 is like going into the store and say that you want to no buy anying
DPM++ M2 SDE is basically a store that will throw a fit when you don't buy anying
so the easiest way to avoid the confrontation is to just go to some other store if you don't want to buy anything

[deathtome1998]

alright, so I did go ahead and just stop using denoising of 0. now I'm trying a denoising strength of 0.5. however when I use the img2img inpainting tab, with everything else the same as before, with or without Ultimate SD Upscaler. with prompts from file or textbox scripts. 3-4 of my images come out looking weird AF. when I do all 25 of my prompts in the prompts from file or textbox script, the 1st and 3rd image always have something weird looking from where I drew the inpainting mask. which I'll throw some pics of all three of the images I just made in here. on the last of the messed up pics the hand doesn't look like it changed at all, and the face has red lips with a green outline and some really weird eyes, eyebrows, and nose. it also doesn't matter if I only use those three prompts in the prompts from file or textbox script, it's always the same 3 images consistently being messed up, I have had another one mess up once or twice, but just redoing fixed that one. unlike these 3 pics
pic 1

pic 2

pic 3

[deathtome1998]

@w-e-w sorry for the ping, just wondering if you had seen my response, (sorry it had been a bit late, it took me a couple days to realize I had a message from you). I hope you or someone else if they look into here can help with why those images are genning like that.

[w-e-w]

I believe you have a bad combination of Masked content and Denoising strength
depending on what you want to achieve you want to use different type of Masked content and different Denoising strength accordingly
the second example image looks to me like the results of Masked content set to latent noise but haveing a too low of a Denoising strength

(yes 0.5 is too low for latent noise I think if would need something above 0.85 to get image out of that)
I don't know what you're trying to achieve but you may not want to use Masked content latent noise

---

at this point you getting bad image is basically out of scope of bugs about UnboundLocalError: local variable 'h' referenced before assignment

and as mentioned thats is a know issue with certain sampler schedule and with edge case parameter combinations

whitch is an issue that can be fixed but no one has devoted the energy to fixing it mostly because there's not much point

---

if you actually want someone to help you or debug a certain issue then you need to provide the information that actually allows people to do so, enough information as to let other people recreate your exact results if possible

depending on case by case basis it may include information such as
your prompt
all of your input parameters to all of your modules extensions sectips ext
you actual input image
your mask
what models you're using
also include information on values that are you think are defaults as not only people that may have different defaults defaults may get changed

note if those info contains nsfw content, best ot finde other inputs that can produce the same issues and that people debug those

---

to be honest I really don't want to reply as the origin issues already solved and you did not really include enough information for someone to "replicate" your situation if I want for experience and guess work
like it might not be obvious to you but only you know what is in your from file or textbox

the more information you missed the more people have to guess which exponentially increase the difficulty for anyone who's trying to debug
up to a point people will just give up debugging and even don't reply

me when starting into this issue we have to think is the issue with from file or textbox is it with your sampler scheduler .....
is it a bug or is it expected behavior they're literally countless possibilities
me being rather experience are able to know what all the likely culprits, but you're not making things easy

---

people may have different opinions on how you should use discussion threads
but since the origin question is sold I would suggest you start a new thread with a relevant title so that people who are knowledgeable about that can see the post and join the discussion

you should probably start a new post something like how to use inpaint or how to use ADetailer
the orignal issue you have is about webui
but your last message is either about from file or textbox or ADetailer
if you're asking about ADetailer then you should probably go to https://github.com/Bing-su/adetailer/discussions

you should probably look there in regardless because they might have some valuable information or experience that you can just look up that might solve your issue without asking anyone

I believe you issues falls under "you don't know how to to use inpaint" which resulting you having trouble with ADetailer

ADetailer is essentially automated inpaint