How many do you think would be ideal? (Currently 170)
How many do you think would be ideal? (Currently 170)
Fi would say try to increase 340 and see how it goes
Fas you improve dataset next training should be better
Jokay will try! Took a long time just to gather the 170 though manually picking out from peoples instagram, should perhaps find a faster approach
Fimage to audio no prompt really good
Fyes it is a tedious task
Faim maximum quality
JWould you include people? Or avoid them? Currently 80% of photos had people. I only need the model to generate portraits photos in the end
Jwtf thats so sick
JAnd would you include girls, even though I only want to use the model to fine-tune again on photos of guys?
Fyou can include
Fbut dont include same person
Fyep
Fnot necessarily needed to add girls
Mhow do you guys do that? first you take a base model and finetune it on iphone style images and then you take that same model and finetune it on specific person. Right? but wont the model forget the style it learnt from that iphone photos ?
Mwhat model is that?
Mand can it do video to audio?
Fyou can test few more things i will fix in few hours
FPatreon
Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon
FPatreon
Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon
Fadding step speed
Bfor training the data set for the person lora , what have you focused on in the image discription (prompts) you provided for training person lora on flux ?
where could i find this , what should i focus to make person lora work well with emitions and the persons accuracy -> realism @Furkan Gözükara SECourses
where could i find this , what should i focus to make person lora work well with emitions and the persons accuracy -> realism @Furkan Gözükara SECourses
Fjust ohwx man
Fit works best
Falso we have more proper example dataset
FPatreon
Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon
B@Dr. Furkan Gözükara doesn't captioning the dataset of the person flow make any good sense for ohwx man person lora trainig on the data set , i am not gettting it somewhere in the filex and the blog you shared
Fcaptioning reduces likeliness
Fyour dataset is not good in that case :d
Fcaptioni,ng only helps when your dataset is not consistent
Bconsistent in what sense , like if we have a data set of a person with all the things that you mentioned in the blog about the ohmx man training with diff shots and diff clothes at diff places also considered consistent or not @Dr. Furkan Gözükara
Fit is consistent as long as it is same person
Fand no other persons exists in dataset
Bwhat about he is having beard and diff hairstyle is that inconsitent , if yes then what to take care in captioning so we are not over eng things in the captioning for the lora traning
B@Dr. Furkan Gözükara
Fi think still not needed
Fmmaudio almost ready
F
TThis is my latest NEGATIVE Promt. It is meant for jus a regular realistic video. Wan tends to shift video styles depending on your prompt, lets say you want aliesn in your video, all of a sudden it starts making 3D render or animation instead of realistc video. This prompt will lock you in realistic style and produce less shitty videos by a good margin than |no negative prompt at all. Dont waste your time. Take it ant tweak to your needs:
Overexposure, static, blurred details, subtitles, paintings, pictures, still, overall gray, worst quality, low quality, JPEG compression residue, ugly, mutilated, redundant fingers, poorly painted hands, poorly painted faces, deformed, disfigured, deformed limbs, fused fingers, cluttered background, three legs, a lot of people in the background, upside down, inaccurate, distorted, motion smear, motion blur, motion stutter, video noise, video artifacts, warped, motionless, 3D render, animation, anime, 3D game, cartoon, comic book, manga, stylized, abstract
Overexposure, static, blurred details, subtitles, paintings, pictures, still, overall gray, worst quality, low quality, JPEG compression residue, ugly, mutilated, redundant fingers, poorly painted hands, poorly painted faces, deformed, disfigured, deformed limbs, fused fingers, cluttered background, three legs, a lot of people in the background, upside down, inaccurate, distorted, motion smear, motion blur, motion stutter, video noise, video artifacts, warped, motionless, 3D render, animation, anime, 3D game, cartoon, comic book, manga, stylized, abstract
Fapp ready publishing
Fwill make a video too
FF
Fv3 published all features done
Fplease test asap and let me know 
FPatreon
Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon
Fi will try to make kaggle notebook too now
Jwow you are fast i just tested version 2 and it works great, haha have to test your newest version now! 
Fi think it is perfect now with all features