why does my stable diffusion make diffrent image with same seed and settings?

[CryptoSoltan]

I think the models are not showing the correct promt
like i ask for a carpet and it shows me a dog
and the images are very bad
I have downloaded many models to see what's going on and they all are showing garbage.
I have tried short prompts with negative prompts too. 512x512 - 20 step - scale 7 - sampling models has been checked.
the last changes I remember i did that F this up:
create a VENV folder in stable-diffusion-webui
I think I have updated gradio too
(Im not sure because im a noob and ive followed some tutorials on youtube for that lol)
please tell me what have I done to me stable diffusion

[AlexPlom]

Can you please provide examples of specific ones you have tried and what the original output + your output looks like? The issue might be a hundred different things

[CryptoSoltan]

the output of the same hash:
I've tried many of my older prompts and compared them to newly generated ones and get the same garbage results every time
I can't quite tell what's wrong with it

here are the main model example image prompts

model:
Protogen x3.4 (Photorealism) Official Release

generation data:
modelshoot style, (extremely detailed CG unity 8k wallpaper), full shot body photo of the most beautiful artwork in the world, medieval armor, professional majestic oil painting by Ed Blinkey, Atey Ghailan, Studio Ghibli, by Jeremy Mann, Greg Manchess, Antonio Moro, trending on ArtStation, trending on CGSociety, Intricate, High Detail, Sharp focus, dramatic, photorealistic painting art by midjourney and greg rutkowski
Negative prompt: canvas frame, cartoon, 3d, ((disfigured)), ((bad art)), ((deformed)),((extra limbs)),((close up)),((b&w)), wierd colors, blurry, (((duplicate))), ((morbid)), ((mutilated)), [out of frame], extra fingers, mutated hands, ((poorly drawn hands)), ((poorly drawn face)), (((mutation))), (((deformed))), ((ugly)), blurry, ((bad anatomy)), (((bad proportions))), ((extra limbs)), cloned face, (((disfigured))), out of frame, ugly, extra limbs, (bad anatomy), gross proportions, (malformed limbs), ((missing arms)), ((missing legs)), (((extra arms))), (((extra legs))), mutated hands, (fused fingers), (too many fingers), (((long neck))), Photoshop, video game, ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, mutation, mutated, extra limbs, extra legs, extra arms, disfigured, deformed, cross-eye, body out of frame, blurry, bad art, bad anatomy, 3d render
Seed: 105259061, Steps: 30, Sampler: DPM++ SDE Karras, CFG scale: 10

the wrong image its making now:

the correct one:

[econundrum]

doesn't that happen with xformers?

[CryptoSoltan]

I dont really know that

[AlexPlom]

In the Webui-user.bat there is a line called:
set COMMANDLINE_ARGS=
I have edited mine to be the following:
set COMMANDLINE_ARGS=--xformers

You can try this. Furthermore on the settings tab in the ui I have set the Eta noise seed delta to 31337

[CryptoSoltan]

Im still having the problem
generated images are very poor and it seems it uses completely different seed to create them, they are mostly not related or very little related to the prompts I write for it
I have installed SD in another place in my computer - I have RTX 3060 ti and I have tried to install cuda 12 but it didn't install it (tried several times and it didn't work)
here's my log:

venv "F:\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug  1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Commit hash: ea9bd9fc7409109adcd61b897abc2c8881161256
Installing xformers
Installing requirements for Web UI
Launching Web UI with arguments: --xformers
Loading weights [44f90a0972] from F:\stable-diffusion-webui\models\Stable-diffusion\protogenX34Photoreal_1.safetensors
Creating model from config: F:\stable-diffusion-webui\configs\v1-inference.yaml
LatentDiffusion: Running in eps-prediction mode
DiffusionWrapper has 859.52 M params.
Applying xformers cross attention optimization.
Textual inversion embeddings loaded(0):
Model loaded in 69.8s (load weights from disk: 4.5s, create model: 1.0s, apply weights to model: 62.2s, apply half(): 0.6s, move model to device: 0.6s, load textual inversion embeddings: 0.9s).
Running on local URL:  http://127.0.0.1:7860

[CryptoSoltan]

have you added downloaded models recently?
I have added some Civitai models , could it be the reason of it?

[AlexPlom]

1. Do you have the settings I recommended?
2. I am using mostly CivitAi models. Dreamshaper for example works perfectly

[bmc84]

I've noticed the same issue today.

When using the exact same everything (such as using the Image Browser, then "Send to txt2img") - I'm getting different images.

This wasn't happening yesterday when doing the same thing with the same images. Yesterday they were identical, today they are not.

I haven't changed anything at all. I have absolutely no idea how this could happen, when it was working yesterday.
I can't post my examples since they're not quite SFW.

Does running the webui-user.bat file check for updates or perform any updates each time it's run? Perhaps something has happened there, because I have been doing the same process & this is the first time this has ever happened.

[CryptoSoltan]

yes I think this can be possible because I've use git pull in the command arguments
but it must be something related to the system of the PC , since I've installed SD on another folder perfectly and still having the same issue.
If anyone knows anything please help us!

[bmc84]

I accidentally figured out why mine was doing this.

Cause of my issue: an extra 'noise' value has been introduced under the Automatic 1111 settings. This value was populated with a non-zero value, meaning that previously I had a seed of 123, this 'extra' noise setting was being added to that seed - causing images to be different.

Fix: Goto Settings and choose the Sampler Parameters menu on the left. At the bottom of this page, there's a textbox called Eta noise seed delta. If there is a value in this box (mine was set to something like 31337) ... remove it and apply the settings, re-start the UI etc. Try again. This fixed the issue for me. The value of 31337 was being added to all my seeds. No idea how this was introduced, but now it's gone & everything is back to normal.

[Mich-666]

ENSD is there ever since NAI leak. People actually used it with CLIP skip 2 to generate replicable results, not the other way around.
It shouldn't really matter if you are using random seed and it shouldn't matter at all if you are not using ancestral sampler (like Euler a)

Edit: Latest commit. Seems like I'm getting bad results with DPM++ SDE Karras too now. No matter what, the mentioned Protogen Photoreal images look bad and out of focus. Tried both compatibility options (Use old karras scheduler sigmas and Use old emphasis implementation) but to no avail. I'm not using xformers either. ENSD makes no difference, and also, there is no such option in Samplers now (I only have it because I have it pinned on main page).

This might be relevant but I can't see any reason why above-mentioned prompt should be broken:
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Seed-breaking-changes

[Neweh]

Edit: Latest commit. Seems like I'm getting bad results with DPM++ SDE Karras too now. No matter what, the mentioned Protogen Photoreal images look bad and out of focus. Tried both compatibility options (Use old karras scheduler sigmas and Use old emphasis implementation) but to no avail. I'm not using xformers either. ENSD makes no difference, and also, there is no such option in Samplers now (I only have it because I have it pinned on main page).

I am having the same problem with my installation, tried clip fixing models, compatibility settings, etc. to no avail
Suspect it's the same problem as #7340 where no solution has been found either.
Not convinced this is OPs issue however, the difference in generated images is profound in their case - there's barely any similarity between the prompt and the generated image, or the image it was supposed to generate. It seems doubtful to me that it's down to comparability settings, xformers, or anything even in that vein

[coolbei]

this works. Thanks man.

Fix: Goto Settings and choose the Sampler Parameters menu on the left. At the bottom of this page, there's a textbox called Eta noise seed delta. If there is a value in this box (mine was set to something like 31337) ... remove it and apply the settings, re-start the UI etc. Try again. This fixed the issue for me. The value of 31337 was being added to all my seeds. No idea how this was introduced, but now it's gone & everything is back to normal.

[CryptoSoltan]

The problem got fixed by deleting the cash folder contents
C:\Users\yourcomputer.cache
re-launch webui it will download the correct file and done!

[CycloneRing]

I just noticed for me it's caused by using some LORA models, When they get loaded they change the base generation seed and after that even generating previous result will be a little bit different, By restarting Stable Diffusion and don't touching LORA it will be same again.

Weird...