itch.io

...once I can implement an easy button to download the default models. :X
With my current internet I'd have trouble uploading the 30+gb package again.

Is their automated installer not working for you?

Can you send a screenshot to Sunija#6598 on discord?

You can update it within the starter. :)

I sadly don't have it uploaded anymore, because the space on my server is limited. :(

If you send me some google drive or dropbox link (via DM), then I can upload it there. :) Should have around 10 GB of free space.

I'm not sure if I understood you completely yet. ^^'

But on the github page of my standalone, there's an instruction on how to create the standalone yourself. It's a bit more "techy", because it requires you to do stuff in the command line. I can send you the compiled starter and batch files via Discord (Sunija#6598), I think I didn't add those to the repo yet. :X

You can find the github page here. Is that what you need?

I think it's called "Denoising Strength". And yeah, that's a really confusing name. :/

Thanks for reaching out! :)

Did you try to start the starter (invokeai_starter.exe) before that? The starter sets some required variables first.

Thanks for reaching out!

If you use the itch.io/app you might be able to download the file in parts. I think. :X

I *could* maybe make a "minimal version" that does not contain the control net models and the SDXL models. Then we can go down to 8 GB again. InvokeAI contains a downloader (it's in the commandline, but kinda usable) so you could download the models after that.

You are running out of memory. Maybe try to make smaller images first, then try to increase the size until it crashes, so you know the limit of your graphics card. :3

False error. :)
Triton is only available on Linux, so this error should not even appear on Windows.

It's uploading and will be available in 2.5h here. :)
(If you start downloading before that, you'll get an incomplete file.)

Yes, that was definitely the wrong link. >.< Thanks for the correction! I switched out the link.

Thanks for reaching out!

You have to open the page, and then delete the local storage (explained here).

Sadly not (yet), there were too many changes. :(
You'll have to download the new version.

Sadly not. :( AMD didn't catch up yet.

Thanks for the report! :)
Do you maybe have comparison numbers (invoke with x loras, vs comfyui with x loras), so I can send that to the team?

Can you go to "C:\Users\Ape\AppData\Roaming\" and rename the "Python" folder to "_Python", and then see if it works?

As background info:
For some reason (that I couldn't find out yet), parts of the standalone will not use standalone-code but python code that is already installed on your PC from other python programs. Renaming that folder will make it not find the wrong code and use the right one.
That's not a great long-term solution though, because it might mess up other python programs that you installed previously. :/

Can you try to reload the black invokeai page with Ctrl+F5? :)

It contains the 4 most used controlnet models (pose, canny, etc). You can easily download more by clicking "Start model installer" in the starter. I left the others out, because that would increase the download by 10 GB.

But I can also make an optional "all included" version. :)

Sorry for the issues!

You can fix your installation, by
1) Download this patch (1 MB)
2) Put it in your invokeai3_standalone folder
3) Unzip it

All new downloads of the standalone already have this patch.

Did you use the update script or download the new version directly? :3
The update script sadly won't work. :( There were too many changes.

Good idea! :)
Btw. first version is up. Would be cool if you could download and try it, so we can see if there are any issues.

I'm compressing the new standalone atm. :) So tomorrow it should be ready.

Downgrading sadly doesn't work. :(
But you can download it again and copy the old output folder to your new downloaded version. Then you should be able to use all your images.

Sorry for the inconvenience. :(

As soon as it's officially released, it will be implemented by the InvokeAI team. :)

Hi, thanks for reaching out! :)

If this is still a problem, can you send me a screenshot of your canvas (and the settings) on Discord? (username: Sunija or Sunija#6598) Then I can check if I see the issue. :)

Hi, thanks for reaching out!

Hi, thanks for reaching out!
There was a broken update. To fix that, download InvokeAI again, move and move your output folder to the new installation.

I'm currently checking if the update script is okay to use again. :) Sorry for the trouble!

Yes, but CPU is really slow (1 image every 10min).
And, in the current version you might have to add "--no-xformers" in the third to last line of helper.bat to run it on the CPU. :/

Edit: Can you install 7zip and unzip it with that (or WinRar, etc)?
Apparently windows doesn't support the zip archive type that we used. :/ Sorry for the trouble.

Yes, it includes LoRA.
When you got the newest version, there is an "open lora folder" button in the starter. Click it, and drop your loras there.

In InvokeAI, there's an "Add Lora" button below your prompt, which you can click to add loras. :)

Thank you for reaching out! :)
Can you send me a screenshot of your outputs on discord (Sunija#6598)?

Thanks for reaching out! :)
In which folder did you try to extract the zip?

(There are some folders, e.g. system files, etc, where windows won't allow the extraction to go through.)

I added the mirror download link again. :)
Thanks for the notice that it's working again! <3

It works exactly like aiimages, so there is no setup. :)

Only exception: If you do not have a nvidia card, it will download the 32bit models, because it cannot use the 16bit models to run on your CPU.

It appears that you are on a 32bit system (e.g. old Win7). The starter itself only supports 64 bit.

You should still get it to run with those steps:
1) Open invokeai_standalone/invokeai/invokai in a text editor (e.g. "Editor")
2) In the lines --outdir="" etc add the paths that you want to use. E.g. --outdir="E:/invokeai_standalone/outputs/"
3) You can disable the nsfw filter by removing the last line (--nsfw_checker)
4) Save the file
5) Double click invokeai_standalone/helper.bat

The most recent. :) (2.3.5)

I think InvokeAI will definitely add it once ROCm properly works on windows. :)

Yep, thank you for mentioning that! :)

Also, the new download is directly 2.3.5 now.

Thanks for the report! :)
There were some patches since then, so the problem might be solved. If not, definitely poke the InvokeAI team about it on their discord. <3