For some reason today I get different images using yesterday's prompts. What can be the case?

[Bra2ha]

I didn't change the hardware, didn't update the drivers.

Yesterday's image

Today's image

[Bra2ha]

Also, I get different results each time I press Generate, with locked seed.

[iqddd]

Same. Same seed, prompt etc - different results.

[iqddd]

Looks like problem occurs only with “Flux realistic” sampler.

[lllyasviel]

i seem to be able to reproduce. will take a look next time when i am on forge

[ReppiksProductions]

I'm having the same issue with the "Flux Realistic" sample. I didn't notice for a long time because I've been doing some testing on my LORA. I thought the difference in my epochs was causing drastic changes to the seed LMAO. Now I'm going to have to do hours of comparison generations again smh, but at least it will be easier to compare.
Are we sure this isn't the expected result when using "Flux Realistic"?

[iqddd]

We don't need non-deterministic samplers. That's for sure. I think there's a non-seeded random being used somewhere.

[djasil]

@lllyasviel I think I found the culprit commit: 1b44ae2

I just restored the previous GOOGLE_BLOCKLY_COMMIT_HASH in modules/launch_utils.py:
1e98997c7fedaf5106af9849b6f50ebe5c4408f1 -> bf36e6fd3750a081f44209ba4f645adb598f7e37

I don't know what this hash is and how it is being used, but this change fixed the issue on my side with Flux Realistic (reproduced the same old gens with LoRA every time).

[Bra2ha]

It works 👍 Thank you

[tristanwqy]

@lllyasviel I think I found the culprit commit: 1b44ae2

I just restored the previous GOOGLE_BLOCKLY_COMMIT_HASH in modules/launch_utils.py: 1e98997c7fedaf5106af9849b6f50ebe5c4408f1 -> bf36e6fd3750a081f44209ba4f645adb598f7e37

I don't know what this hash is and how it is being used, but this change fixed the issue on my side with Flux Realistic (reproduced the same old gens with LoRA every time).

that saved my life!

[grvdggr]

Thanks for sharing, really appreciate it. =)

[grvdggr]

Unfortunately, [Forge] Flux Realistic (1.5x Slow) disappeared when I changed the hash and that's the best one to use along with DDIM as Scheduler according to my tests.
I changed the hash back and will wait for a solution that does not delete [Forge] Flux Realistic (1.5x Slow).

[iqddd]

What are those two modules from ‘~/stable-diffusion-webui-forge/repositories/google_blockly_prototypes/forge’ anyway?
additional_samplers.pyz
icl_v2.pyz
The code is compressed and obfuscated.
They recently added the ‘skip-google-blockly’ startup flag.

parser.add_argument("--skip-google-blockly", action='store_true', help="launch.py argument: do not initialize google blockly modules")

[shaun-ba]

google_blockly_prototype

quick google: https://github.com/lllyasviel/google_blockly_prototypes

Some of our projects will begin to use Google Blockly as a prototyping tool for certain developments. This approach is aimed at enhancing the speed and flexibility of experimental feature creation.

[usigg3]

I put --skip-google-blockly in webui-user.bat and that seems to fix the problem. Not sure if it has other effects... Can't see any so far.

[nektosu]

If put --skip-google-blockly in webui-user.bat
[Forge] Flux Realistic
disappears from the list

[usigg3]

For me it's still there.

[kalle07]

same here ...
any solution ?
that "--skip-google-blockly" can only be a fast fix

[iqddd]

DPM++ 2S Ancestral 2nd Order = DPM++ 2S a

[kalle07]

#2151

[iqddd]

Gosh, what a mess!

[iqddd]

thx i use swarmUI -> forge are OUT ;)

Has lllyasviel said anything about the lack of updates on Forge?

[kalle07]

i dont know him ^^
but he has also a part of fooocus what has ended because mashb1t has left/paused