Mage.Space: Updates June 2023 (LORA & Textual Inversion)

Profile: www.mage.space/u/PlagueDoctor
Official Mage Discord: / discord
Info about Samplers/Schedulers: stable-diffusion-art.com/samp...
Info about LORAs: stable-diffusion-art.com/lora/
Info about Textual Inversion/Embeddings: stable-diffusion-art.com/embe...
Info abou Clip Skip: github.com/AUTOMATIC1111/stab...

[@martinkay67]

Thanks a lot for this update! The new possibilities with thousands of new layers was very confusing. This video helps to get through with it way more easier!

[@dannarchy-com]

Thanks for watching. I think the process of making the video was also a huge help to me too, since I forced myself to figure out what all this new stuff is.

[@WizzardofOdds]

Glad I found this, I am lost after this update. Thanks for showing what it all does.

[@dannarchy-com]

Thanks for the comment. I was really overwhelmed too, so hopefully I got the main jist of the tech. We're all learning daily with this.

[@brett3732]

Awesome video!!! Thank you so much for the information!

[@MortenNormannAlmeland]

Love your videos!!!!

[@dannarchy-com]

Much appreciated

[@izazael]

On the topic of schedulers, I've found that some work a lot better for anime/animated style images, while others are great for photorealism.

[@dannarchy-com]

I seem to be sticking with UNIPC now since it's good with less steps and it also is convergent so it will reproduce the same image when using the same seed. Some don't due to random step noise

[@arispoli1269]

Great howto Thanks. Can you do a video on uploading an existing image and ‘enhance’ that eg take a holiday photo and turn into a LinkedIn portrait etc

[@dannarchy-com]

Stable Diffusion isn't the best for this type of application. Unlike MidJourney, which will interpret the content and almost act like a single-image textual inversion, SD simply uses an image prompt as a noise layer. Based on the prompt strength, it can range from something identical to the original image, or just a vague composition and color tone. The AI models in conjunction with the text prompt will always want to reach an end goal, and any image fed into it, will help guide it, but ultimately will only re-route it slightly. You'll never get the original face simply from a sample image unless the model and/or prompt would also produce that same face.

[@limpneckmike]

Hi, I’m new to this. Do people submit Models and Lora’s, or is it something that only the people behind Mage Space does? Just wondering where they come from and if there’s a way to create your own custom models and layers.

[@dannarchy-com]

You can't add at the moment. They grab the popular public ones and use those

[@scottpage3679]

how do you get the layer:lora sliders on the home page underneath the base models? mine only has the base models showing.

[@dannarchy-com]

Do you have a paid account? If you do, LORAs and TIs will only appear for the base they were added to, so you'd need to select either 1.5 or 2.1 for them to show. If the new SDXL is selected, then nothing will be there since it has nothing at the moment.

[@Beard_Hood]

is there a way to find out what was used image wise to train different base models?

[@dannarchy-com]

Not so much. The base SD model has an enormous amount, and then the fine tuned models on top of that have their own. In the mage discord server, there is a channel where you can ask the creators questions

[@deanbawdry]

nice work, can you do one showing how to make a tune of our own face to use?

[@dannarchy-com]

I'm not skilled in that at this point. You'd have to create some sort of TI in dreambooth, I believe, and Mage wouldn't support the result currently. Best you can do is choose a starting photo, but it will only loosely use it for generate composition, and not learn the face.

[@malarkey102]

Hi will you be making a video on mage new pro+ plan how to upload base models and Lora’s very interested

[@dannarchy-com]

I plan to, just was waiting for them to iron out the kinks since there's been a few issues with the import process. I've been keeping tabs on the discord, so once I feel things are stable, I'll play around and see what I can do.

[@MortenNormannAlmeland]

Do you have anything about inpainting?

[@user-ki3lk6zj3m]

Are the people providing LoRA ordinary users? How can I learn?

[@dannarchy-com]

Yes, they are just grabbing the top several thousand from civitai. I believe something like dreambooth is used, but I'm not as familiar with the technical aspects

[@CaptianTitan]

Do a tutorial on toonify

[@dannarchy-com]

I'll put that on my to do list

[@nc2988]

THANKS, I wonder when the latest stable diffusion will be available on Mage. I kind of lost interest in AI art,..just too much of the freaky face, hands, limbs wasted a lot of time. I hope this is all fixed soon. Also it gets boring, not being able to create an action scene. Characters or landscapes get boring if they're not doing something except staring at the camera. I try to make a guy brushing his teeth and he's shoving some melted plastic spike into the side of his face,..yuck. Maybe a year from now it will be perfected.