@Dr. Furkan Gözükara the newest of your ComfyUI install include DeepSpeed, does it improve inference
@Dr. Furkan Gözükara the newest of your ComfyUI install include DeepSpeed, does it improve inference speed or something ?
RHey I made a music video using veo 2, some of you might find it of interest, anyway its all AI (except the music). Hope its ok to post here.
https://www.youtube.com/watch?v=XICNVp7yhg0&t=1s
https://www.youtube.com/watch?v=XICNVp7yhg0&t=1s
YouTubeRuairi Robinson
Music video for Steve Moore's "The Blue Stone"
Directed/edited by Ruairi Robinson
Visuals generated using Google Veo 2
Directed/edited by Ruairi Robinson
Visuals generated using Google Veo 2
.@Dr. Furkan Gözükara Here a gen from a custom fine tune on large images I’ve tiled and cut into 1024x1024 pixel chunk separate pngs to capture every tiny possible detail from the up to 6K large images. It was an experiment and it went well with your config. Original image count was 15 for those large images and then after filing there were over 400 images. So this seems to be the trick to be able to train on any resolution images without losing any details.
RThis is absolutely stunning mate!
Jinteresting! how does the model know the tiles fit together when training?
Jmade a new flux fine tuning model of my friend with 58 images, also gave great and trustworthy results. im wondering how low it can go. my first attempt with 15 images was too glass-skin like. but yea at least now i know that 58 images works well!
.I've written a python script with chatGPT that will tile them with overlap and is also including the original full size image. Plus I'm using flux dev de-distilled.
.@julius so yeah the fine details the model can't train on when only the full image is trained on it's own since the image gets automatically downscaled to 1024x1024 so the training will never see the full details in your dataset. I've taken a step further and have 3 different resolutions for the original image in 2x increments so that it will have 1024x1024 which is the full image downscaled to give the training the full context of the subject then a 2048x2048 version which will serve for 2x Ultimate SD Upscale training on 1024x tiles and then a 4096x4096 size tiled in 1024x1024 tiles to serve as data for training for 4x Ultimate SD Upscale and lastly the original resolution whatever it is to be tiled into 1024x1024 tiles with overlap to not miss any detail from the original dataset.
.Here some more outputs.
.
.
.
NWow, there are so many neural nets that you've taken apart and assembled
DHas anyone used the nee deep face similarity app? It works great but I'm not sure what sort of numbers are considered very similar in terms of comparison. I know lower is better but is .3 good or still very off for instance?
RIt looks great. How long did it take you to complete the training process?
.Looks good on the low resolution. How do the close up renders come out or high res upscales?
VWhere are the models saved that are downloaded the first time I run the video generation for CogVideoX1.5-5B-I2V ?
MPlease, I've been here for two days and I can't do anything here anymore...
can anyone help
can anyone help
Fyes there are
Fhere my experiments
FMedium
Thoroughly experimented with Fine-Tuning / DreamBooth training of Flux-dev-de-distill, PixelWave v03, Verus Vision and base FLUX Dev model…
Fi rather prefer each time train from scratch with improved dataset
Fif they are different class yes can work
Fi trained a person and a handbag
Fthey use different scripts
Fbut if implementation is same should perform same
Fnice
A@Furkan Gözükara SECourses SOS 
I need a list of available stable diffusion classes
I need a list of available stable diffusion classesADoes it work with the T5Xxl?
Fi test manually there is no such thing
Ftest your prompt and see if generates similar thing
AThat's a good approach yeah
Fimprove dataset and train from 0 again - best way
A@Furkan Gözükara SECourses I saw your new upscale gradio , does it reach supir quality?
Fyou can data augment and train - crop differently - i trained with like 3 or 4 images person worked fairly good
Fonly if the model uses
Fmostly llms vlms
Fyes i also find more better than fewer
Fexcellent
F100% here so far 
F
Fi tested all the apps and fixed broken ones due to new libraries
Fit is comparative so good to see ranked. but for values hard to tell
Fif not in same folder than it means it is in the hugging face cache
Fso yes it is hugging face cache
F
Fare you running as administrator?
Fdont do that ever
Funless i written in the patreon post
