Home Artists Posts Import Register

Content

Besides bug fixes and fixing the memory using Upscaler + Save Vram, there is a few new options.

Scheduler: To change the scheduler used by the model, there is only a few options for now, its been hard finding the source for more schedulers, any link with more schedulers are welcome.

GPU: For those with multiple GPU's, you can select the one you want to use.

Model: It will handle zipped .pt files, users can't generate this files yet, but it will make model loading fast again for the default 2.4 model (it was used in 0.41 and removed on 0.5)

Use Float32: Only useful if you have a card that can't handle the default float16 weights. You can also download the full precision weights and turn this one for the "True" full precision, but I don't think it will affect the results that much.

Inpainting and img2img: Are better

Origin Image: Now you can select multiple images

Prompts: As you can see from the picture, you can now pass some parameters if you like. For this version, the parameters are limited for those on the picture, but it should improve in the future. 

Will try to finish the build today at the end of the day.

Files

Comments

Anonymous

Just an idea, that might be helpful for user experience: The text box to enter prompt is big and has scrollbar. But since it has a very cool feature to support multiple prompts, I very easily get it overflown with text, which is hard to manage. So maybe there could be a way to have a separate resizable window working like a notepad, where You can see a lot more text and organize many prompts into a bigger batch easily - as floating windows popping up with a press of a button [more] or [...] or similar?

Anonymous

more steps when?

Anonymous

Wow! This version is also amazing! A technique called CodeFormer, which uses a Transfomer-based predictive network to enhance low-quality face images, is said to be more stable than the traditional GAN-based model. https://github.com/sczhou/CodeFormer I thought it might be useful for GRisk AI SD at some point, so I thought I'd share it with you just in case you know about it.

Anonymous

Thanks for the update and all your efforts!

Anonymous

There are not enough steps? 50 or 150 have very very similar results, quasi identical. Try it out in a Google Colab. CFG Scale has a lot of impact and y this software GUI has the highest one from 0 to 100

Anonymous

>>Origin Image: Now you can select multiple images<< is this for multiple prompts/outputs or to affect one output(if so how does that work)?

rune hansen

Was this getting released today or what?

Anonymous

why are all my images black ?

DAINAPP

Try turning on "Float32" on the last build. But it will use more vram

Anonymous

I'm checking out the "multiple image" thing, and it turns out that by "multiple images", the selected images are in order, and follows a *edit* single prompt to image process. This is amazing, I am testing a PNG sequence as the image weight now and can already see positive results. You are on the verge of being a deforum rival. Thank you! *I can think of some suggestions, allowing the prompt change sequentially like the image or even better, add a way to be able to change the prompt at specific frames. Also, custom filename saving with sequential option. This would make it easier to import as a sequence into something like After Effects.

Anonymous

or the image sequence is used in its entirety for every prompt it looks like. This could be interesting 😉

DAINAPP

Good call on the name sequences. Need some tools to improve the file names. There is more options for animations soon, also the option to select a mp4.

Anonymous

+1 for more development on the animation side & image sequence input - that would be awesome!

Anonymous

So I can just select multiple images and it will process them in order? Fascinating! I didn't know this was already possible with GRisk..can't way to try it out.

Anonymous

For anyone reading this, is it still normal to get the warning on startup, even for version 0.51? Or was that only supposed to be in the free 0.1 version? torchvision\io\image.py:13: UserWarning: Failed to load image Python extension: torch\_jit_internal.py:751: UserWarning: Unable to retrieve source for @torch.jit._overload function: . warnings.warn(f"Unable to retrieve source for @torch.jit._overload function: {func}.") torch\_jit_internal.py:751: UserWarning: Unable to retrieve source for @torch.jit._overload function: . warnings.warn(f"Unable to retrieve source for @torch.jit._overload function: {func}.")

Anonymous

This is normal start up sequence Huy Nguyen. Ignore it.