New Embedding Isn't Showing Up in Dropdown menu In Train Tab

[cre8ivediffusion]

My issue is exactly what the title says.

I'm interested in training textual inversions, have been doing a lot of research into how to properly do it, but I've run into an issue where I can't locate and select my newly created embedding. In fact a bunch of the previously installed embeddings I downloaded from Civitai weren't there either. When I go look at them in my Finder (yes, I'm on a Mac) they are all there, but for some reason only about 8 or 9 of them are showing up in the drop down.

I've tried removing them and adding them in one by one. Same issue. Newly created embeddings don't show up, and the ones that were visible before are also visible when added in solo.

Not sure what else to troubleshoot.

[quasiblob]

Not sure if this is the same thing you are having.

Seems like if you select a model that is based on SD 2.x, embeddings that are created with 1.4 or 1.5 won't be visible in the list:

As soon as I load a 1.5 model (for example), the embeddings list will be populated again.

I guess this is some compatibility thing, 2.x can't use 1.5 embeddings.

[Nacurutu]

Loaded 1.5 models only going to show 1.5 embeddings,

Loaded 2.0/2.1 models are only going to show 2.0/2.1 embeddings

You CANT use 1.5 embeddings with 2.0/2.1 models and vice versa.

[Centurion-Rome]

You CANT use 1.5 embeddings with 2.0/2.1 models and vice versa.

Few weeks ago I HAVE SUCCESSFULLY used 2.1 embedding on 1.5 model... ;)

[Nacurutu]

let us know how... because from what I know, Is just not possible.

[Centurion-Rome]

let us know how... because from what I know, Is just not possible.

Like I said, WAS working some weeks ago (before the limitation was implemented).
You can try it for your self with some old commit...

[Nacurutu]

let us know how... because from what I know, Is just not possible.

Like I said, WAS working some weeks ago (before the limitation was implemented). You can try it for your self with some old commit...

There is no webUI limitation that I Know.. is the way 1.5 and 2.1 are made/train... Again, from what I know, 1.5 and 2.1 were never compatible.

[cre8ivediffusion]

To be honest, I probably wasn't mindful of what checkpoint I had selected up top when creating the embedding the other day. And I do remember reading about that but completely blanked on checking that.

You both had a very solid point, and I thought that could have been the issue, but I just created a new embedding while making sure SD 1.5 was selected, went to the train tab to select it and it still isn't showing up.

I checked that it was in the embeddings folder in Finder, and it is. Yet it doesn't want to show up in the dropdown to allow me to even begin training the embedding.. let a lone try and use it.

Appreciate those suggestions, but I'm still stuck. Have any other ideas as to what could be causing this?

[quasiblob]

I bet you already also tried pressing the refresh button too?

[cre8ivediffusion]

I did try refreshing.

So before I do a fresh install, (which I might need some advice on how to do that efficiently if you didn't mind, as I've literally just started using terminal a few weeks ago when my interest was peaked with all this new AI capabilities), I noticed that my python is at the 3.10.8 version, and in the troubleshooting page of A1111 it says its tested to work up to 3.10.6.. not sure if that could have anything to do with it.

Do you know how to revert back to an older version of python, so I can test that out before completely re-installing?

[Nacurutu]

I have 3.10.9 n everything works properly with the embeddings...

[PeterLorre]

I was having this issue as well. Even after a fresh install of Automatic. I'm using an external SSD and I had to restart Automatic each time I created a new embedding because even after refreshing, they wouldn't appear in the embed dropdown list.

Somehow I'm able to refresh on the automatic install on my C: drive, but no luck on my external drive install. No clue why, same environment too.

Have you tried switching from one model back to the other then refreshing your embeds dropdown? That actually works for me. Huge time saver. Also I don't think it actually matters which model is selected at the time you create your embed file. I've made the mistake of created with random models but it didn't actually matter as long as I trained using the correct model selected.

Side note: training on the original 1.4 model seems to drop better looking results, at least for faces, and they work perfectly with 1.5 base models. You can read a little about it on this reddit post: https://www.reddit.com/r/StableDiffusion/comments/10ddyht/comment/j4llans/?utm_source=share&utm_medium=web2x&context=3

[DJChief]

i still have th issue. any fix?

[Hondymion]

Hello, i have/had the same problem.
Bypassed by adding "--disable-safe-unpickle" to the "set COMMANDLINE_ARGS=" line in the webui-user.bat at the root of the webui folder

I'm training one right now

[Petomai]

wth is a safe unpickle? LOL This worked but idk why.

[natchapol2347]

If you run the program it will mark the created embedding as unsafe. --disable-safe-unpickle will ignore that safety check.
Error verifying pickled file from /content/drive/MyDrive/AI_PICS/stable-diffusion-webui/embeddings/heyyy.pt *** The file may be malicious, so the program is not going to read it. *** You can skip this check with --disable-safe-unpickle commandline argument.

[markhicken]

It was created, by the program. Shouldn't the program assume it's safe?
What's the correctly way to unpickle it then?

[reowasnotavailable]

It was created, by the program. Shouldn't the program assume it's safe? What's the correctly way to unpickle it then?

I'm wondering the same since I tripped over this problem...

[mono8mino]

oh my god. I spend almost 16 hours trying to resolve this. Thank you so much!