Software Engineering Courses (SECourses)•16mo ago

GPU 2, 90 C* I think it needs maintenance

DEXperience SolutionsOP•9/8/24, 2:28 AM

jajaja

DDEXperience Solutions jajaja

Furkan Gözükara SECourses•9/8/24, 2:43 AM

even 100 is good

FFurkan Gözükara SECourses wow ipndm makes it much more realistic

druhl•9/8/24, 2:48 AM

I agree, it adds a bit of grain to the images. It is also very acceptable for multiple loras. I have used like 5 with it together, and it worked marvellously.

druhl•9/8/24, 2:49 AM

The grain makes images seem more natural, realistic when viewed without zooming in. Zooming in is a different story.

druhl•9/8/24, 2:49 AM

Issues I've experienced with the new sampler: SUPIR upscale removes the grain from its output and smoothens it out again.

Ddruhl Issues I've experienced with the new sampler: SUPIR upscale removes the grain fr...

Furkan Gözükara SECourses•9/8/24, 2:50 AM

are you using latest default configs?

Furkan Gözükara SECourses•9/8/24, 2:50 AM

it is very loyal

DEXperience SolutionsOP•9/8/24, 2:51 AM

is it possible after training my lora, to test it with the flux shnell version?

FFurkan Gözükara SECourses nope should work fine

druhl•9/8/24, 2:51 AM

Training completed using 2 GPUs on Mass Compute. The experience was great, much better than Runpod, I may never go back to Runpod again after this.

FFurkan Gözükara SECourses are you using latest default configs?

druhl•9/8/24, 2:52 AM

Yes, using default configs. But maybe I'm making the images too grainy, old film-like. I have to test more.

xlads•9/8/24, 2:53 AM

file with .tar are good to use inside of swarmui? I downloaded "phhofm_sisr_models.tar.001 & 002" from here https://github.com/Phhofm/models/releases/tag/all_models and put them inside the upscale model file then restarted ui but they are not showing

GitHub

Release All my self-trained sisr Models · Phhofm/models

For convenience, I thought Id provide here a compressed archive file with most of my released (single image super resolution) models in it.
This is meant for inference, and most of these can simply...

DDEXperience Solutions is it possible after training my lora, to test it with the flux shnell version?

druhl•9/8/24, 2:55 AM

I don't think flux S will produce good results. Same goes for the compact flux models circulating around.

druhl•9/8/24, 2:56 AM

Doc @Dr. Furkan Gözükara , I am using default (16 bit) setting for DType on massed compute. When I come back with the final lora on my humble home PC and FP8 setting, will I see any drop in quality?

Ddruhl Training completed using 2 GPUs on Mass Compute. The experience was great, much ...

Furkan Gözükara SECourses•9/8/24, 3:01 AM

amazing