Athis would be nice YuE: Open Music Foundation Models for Full-Song Generation https://huggingface.co/spaces/fffiloni/YuE
Z
Fi don thave SVRF
Fi was reading it right now :d
Fnot that i know
Chello
Fhello
Ajust wanted to say thanks for the batch image captioning software. Took a few tries to find the best way to remove tags: "tagname, " instead of "tagname" but it's a real lifesaver vs. going through one by one and slowly deleting words
Fawesome. yes it takes few tries to understand fully
A
RYouTubeRakia Ben Sassi | TekForge
In this eye-opening video, I share my shocking job interview experience that exposed the hidden flaws in hiring algorithms. An AI system made a biased decision, misrepresenting my identity and raising serious questions about fairness in technology.
Accuracy, transparency, and security in AI systems aren’t just buzzwords—they’re the foundation o...
Accuracy, transparency, and security in AI systems aren’t just buzzwords—they’re the foundation o...
4@Furkan Gözükara SECourses Hey,
As agreed several days ago, I remind you the release of VisoMaster by the creator of Rope Next
https://github.com/visomaster/VisoMaster?tab=readme-ov-fil
As agreed several days ago, I remind you the release of VisoMaster by the creator of Rope Next
https://github.com/visomaster/VisoMaster?tab=readme-ov-fil
GitHub
Powerful & Easy-to-Use Face Swapping and Editing Software - visomaster/VisoMaster
WHas anyone tried to run SUPIR to upscale a batch of images on something larger than a 4090? I was thinking of an A6000, as on the 4090 I managed 13 images in 4 hours, so I was thinking that with 48 I could do more than 1 image at a time or even disable some of the vram saving options to help accelerate without loosing quality?
Ais there any way to set top k val with segment in swarm
Aso it stops changing every face in the image
Aoh hahahaha
Aas always
AJust have to follow this guy and you always get your answer
Aoh but they never really implemented it? I don't have any sorting in regional prompting and if I didn't download the yolo model that feature is not well explained at all 
interesting
interesting
ᴏ@Furkan Gözükara SECourses Im getting black screen when using the Flux fill dev. Any idea why its not creating the rest of image. The image is 1028 x 1028 I notice when I do use a init the ratio goes to 1536x1536
SWhat workflow are you using?
Fwhat was the resolution?
Fthat sounds very slow
Ftiling is the real slowing down thing
Fhopefully today sorry for delay
Fhe didnt implement the request or pull request
Fsadly
Fdid you try lower expansion?
Fmaybe settings are wrong
Fi added latest config to the tutorial's source patreon post check it out
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=40)
D
ZZalso using your lora training scripts in kohya, should we be using float, fp16 or bf16?
Fframe count
Ffloat best fp16 almost same
Fdont use bf16 not tested
Zthanks
ZFtier 1 best
Fall tier 1 equal
Fbatch size 7 slightly worse
Zalso i have the circle filled for float, but my training session says enable full bf16 training
Z
Z
Z
ZZis this a bug? do i need to click off of float and then back on?
