Software Engineering Courses (SECourses)•11mo ago

made a new flux fine tuning model of my friend with 58 images, also gave great and trustworthy resul

made 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!

Jjulius interesting! how does the model know the tiles fit together when training?

.tarkan•2/3/25, 7:57 PM

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.

.tarkan•2/3/25, 8:03 PM

@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.

.tarkan•2/3/25, 8:03 PM

Here some more outputs.

.tarkan•2/3/25, 8:03 PM

.tarkan•2/3/25, 8:04 PM

.tarkan•2/3/25, 8:04 PM

FFurkan Gözükara SECourses Click to see attachment

Neuromasters•2/3/25, 9:22 PM

Wow, there are so many neural nets that you've taken apart and assembled

Darth Teethius•2/3/25, 10:35 PM

Has 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?

Jjulius made a new flux fine tuning model of my friend with 58 images, also gave great a...

Renzo•2/4/25, 12:32 AM

It looks great. How long did it take you to complete the training process?

Jjulius made a new flux fine tuning model of my friend with 58 images, also gave great a...

.tarkan•2/4/25, 6:35 AM

Looks good on the low resolution. How do the close up renders come out or high res upscales?

Vegan Apps•2/4/25, 8:38 AM

Where are the models saved that are downloaded the first time I run the video generation for CogVideoX1.5-5B-I2V ?

MeLL•2/4/25, 12:27 PM

Please, I've been here for two days and I can't do anything here anymore...
can anyone help

Ssergio_gonzo Dr Furkan, for kohya trinaing there is another flux dev model that could be used...

Furkan Gözükara SECourses•2/4/25, 6:49 PM

yes there are

Furkan Gözükara SECourses•2/4/25, 6:49 PM

here my experiments

Furkan Gözükara SECourses•2/4/25, 6:49 PM

https://medium.com/@furkangozukara/thoroughly-experimented-with-fine-tuning-dreambooth-training-of-flux-dev-de-distill-pixelwave-13fab8c625bf

Medium

Thoroughly experimented with Fine-Tuning / DreamBooth training of F...

Thoroughly experimented with Fine-Tuning / DreamBooth training of Flux-dev-de-distill, PixelWave v03, Verus Vision and base FLUX Dev model…

RRockerx If I want to further finetune a lora or a model, what is the best approach ? Do ...

Furkan Gözükara SECourses•2/4/25, 6:49 PM

i rather prefer each time train from scratch with improved dataset

�𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮 @Dr. Furkan Gözükara Is it possible to train a style and an object in the same L...

Furkan Gözükara SECourses•2/4/25, 7:02 PM

if they are different class yes can work

Furkan Gözükara SECourses•2/4/25, 7:02 PM

i trained a person and a handbag

Rrob What are the fundamental differences between kohya, onetrainer and simpletuner? ...

Furkan Gözükara SECourses•2/4/25, 7:03 PM

they use different scripts

Furkan Gözükara SECourses•2/4/25, 7:03 PM

but if implementation is same should perform same

AAiInfluence Click to see attachment

Furkan Gözükara SECourses•2/4/25, 7:03 PM

nice

AiInfluence•2/4/25, 7:04 PM

@Furkan Gözükara SECourses SOS

I need a list of available stable diffusion classes

AiInfluence•2/4/25, 7:04 PM

Does it work with the T5Xxl?

AAiInfluence @Furkan Gözükara SECourses SOS 🚨 I need a list of available stable diffusion cl...

Furkan Gözükara SECourses•2/4/25, 7:04 PM

i test manually there is no such thing

Furkan Gözükara SECourses•2/4/25, 7:05 PM

test your prompt and see if generates similar thing

AiInfluence•2/4/25, 7:05 PM

That's a good approach yeah

RRockerx @Dr. Furkan Gözükara can you suggest ? If I want to further finetune a lora or a...

Furkan Gözükara SECourses•2/4/25, 7:05 PM

improve dataset and train from 0 again - best way

AiInfluence•2/4/25, 7:06 PM

@Furkan Gözükara SECourses I saw your new upscale gradio , does it reach supir quality?

Mmmercuri @Dr. Furkan Gözükara What's your recommendation for training with a minimum numb...

Furkan Gözükara SECourses•2/4/25, 7:06 PM

you can data augment and train - crop differently - i trained with like 3 or 4 images person worked fairly good

AAiInfluence @Furkan Gözükara SECourses I saw your new upscale gradio , does it reach supir q...

Furkan Gözükara SECourses•2/4/25, 7:06 PM

nope

Wwardensc2 @Dr. Furkan Gözükara the newest of your ComfyUI install include DeepSpeed, does ...

Furkan Gözükara SECourses•2/4/25, 7:06 PM

only if the model uses

Furkan Gözükara SECourses•2/4/25, 7:06 PM

mostly llms vlms

..tarkan @Dr. Furkan Gözükara Here a gen from a custom fine tune on large images I’ve til...

Furkan Gözükara SECourses•2/4/25, 7:07 PM

nice

Jjulius made a new flux fine tuning model of my friend with 58 images, also gave great a...

Furkan Gözükara SECourses•2/4/25, 7:08 PM

yes i also find more better than fewer

..tarkan Click to see attachment

Furkan Gözükara SECourses•2/4/25, 7:09 PM

excellent

NNeuromasters Wow, there are so many neural nets that you've taken apart and assembled

Furkan Gözükara SECourses•2/4/25, 7:09 PM

100% here so far

Furkan Gözükara SECourses•2/4/25, 7:09 PM

Furkan Gözükara SECourses•2/4/25, 7:10 PM

i tested all the apps and fixed broken ones due to new libraries

DDarth Teethius Has anyone used the nee deep face similarity app? It works great but I'm not sur...

Furkan Gözükara SECourses•2/4/25, 7:10 PM

it is comparative so good to see ranked. but for values hard to tell

VVegan Apps Where are the models saved that are downloaded the first time I run the video ge...

Furkan Gözükara SECourses•2/4/25, 7:11 PM

if not in same folder than it means it is in the hugging face cache

Furkan Gözükara SECourses•2/4/25, 7:11 PM

so yes it is hugging face cache

Furkan Gözükara SECourses•2/4/25, 7:12 PM

MMeLL Please, I've been here for two days and I can't do anything here anymore... can...

Furkan Gözükara SECourses•2/4/25, 7:12 PM

are you running as administrator?

Furkan Gözükara SECourses•2/4/25, 7:12 PM

dont do that ever

Furkan Gözükara SECourses•2/4/25, 7:13 PM

unless i written in the patreon post

Furkan Gözükara SECourses•2/4/25, 7:13 PM

because you see it shows windows

Furkan Gözükara SECourses•2/4/25, 7:13 PM

vp•2/4/25, 7:42 PM

doc - possible for you to upload deepspeed whl and triton whl for python 311 into your huggingface mmorpg repo

vp•2/4/25, 7:45 PM

solved, disregard

AiInfluence•2/4/25, 9:34 PM

@Furkan Gözükara SECourses SOS

https://www.instagram.com/p/DFqmMs_IKyW/?img_index=1&igsh=dmJrbmhxaGdwZG91

acknowledge.ai

Bytedance, TikTok’s parent company, has introduced OmniHuman-1, a groundbreaking AI model that generates highly realistic human videos from just a single image and audio.

Capable of adapting to any aspect ratio and body proportion, OmniHuman-1 pushes the boundaries of AI-generated content.

This innovation could revolutionize digital media, e...

Likes

107

Instagram

made a new flux fine tuning model of my friend with 58 images, also gave great and trustworthy resul

Similar Threads

Similar Threads

Similar Threads