Software Engineering Courses (SECourses)•15mo ago

Does anyone have experience creating serverless Comfy API endpoint on RunPod ? I keep getting "Max

Does anyone have experience creating serverless Comfy API endpoint on RunPod ?
I keep getting "Max retries reached while waiting for image generation" with fresh installation of https://github.com/blib-la/runpod-worker-comfy

dsienra•10/20/24, 4:44 PM

@Dr. Furkan Gözükara Update https://huggingface.co/nyanko7/flux-dev-de-distill Lora training. 3 people same class in 1 lora with regularization, regularization captions just the class, lr 0.00005 unet and Te, very interesting results, high diversity on the class on inference on regular flux-dev, promoting for the trained subjects high resemblance, one curious thing a . on the prompt produces a total different image, for example same seed "Name class" different image than "Name class." is like the . works like a separation in the prompt, more to test, if you will use regularization use only the class for the captions, on previous tests I used captioned regularization and it produces unpredictable and unconcistent results. more things to test

nyanko7/flux-dev-de-distill · Hugging Face

FFurkan Gözükara SECourses this is next hopefully after i published video to test

dsienra•10/20/24, 5:02 PM

Now training De-distilled same dataset with no regularization for testing new output preservation script

Ssnowmonkey Does anyone have experience creating serverless Comfy API endpoint on RunPod ? ...

Furkan Gözükara SECourses•10/20/24, 5:44 PM

this is totally pro stuff. only runpod team may help you for free

Ddsienra @Dr. Furkan Gözükara Update https://huggingface.co/nyanko7/flux-dev-de-distill L...

Furkan Gözükara SECourses•10/20/24, 5:45 PM

thanks for the info

Ddsienra Now training De-distilled same dataset with no regularization for testing new ou...

Furkan Gözükara SECourses•10/20/24, 5:45 PM

great way to find all

FFurkan Gözükara SECourses 10k images 15 epochs would take 53$

ar.1•10/20/24, 6:09 PM

and 160 hours !!!

ar.1•10/20/24, 6:09 PM

thats the only problem

ar.1•10/20/24, 6:11 PM

I can get A6000 ada or L40S which may be a bit faster, although 3x the price

ar.1•10/20/24, 6:11 PM

I doubt they will have 3x the speed up

deepblue_1337•10/20/24, 6:28 PM

what would be the current way for a face swap? Existing image, existing face.

Ddeepblue_1337 what would be the current way for a face swap? Existing image, existing face.

Furkan Gözükara SECourses•10/20/24, 6:32 PM

rope

Furkan Gözükara SECourses•10/20/24, 6:32 PM

face fusion

deepblue_1337•10/20/24, 6:39 PM

rope seems not really to be in develeopment any more, would face fusion be the safer bet?

ar.1•10/20/24, 6:59 PM

Does anyone have any info about optimal dataset sizes ?

I'm trying to retain to make images look more realistic as if they were taken from a camera. How many images should it need for that ?

FFurkan Gözükara SECourses yes but try to find a class prompt that will match better

Nick44•10/20/24, 7:18 PM

unfortunately that didnt work for me too well

Im trying to make a flux lora with varied yet consistent outputs like this with 20 images ( can use more)

I had the best but not optimal results with 1_ohwx Head and body + joy captioning

when I use only instance + class prompt + no captions its way under trained

not too sure what to try next

Ddeepblue_1337 rope seems not really to be in develeopment any more, would face fusion be the s...

Furkan Gözükara SECourses•10/20/24, 7:19 PM

yep that guy more active

Aar.1 Does anyone have any info about optimal dataset sizes ? I'm trying to retain to...

Furkan Gözükara SECourses•10/20/24, 7:19 PM

no such thing but i have shared 15 vs 256

Furkan Gözükara SECourses•10/20/24, 7:20 PM

also style training dataset as well

NNick44 unfortunately that didnt work for me too well Im trying to make a flux lora wit...

Furkan Gözükara SECourses•10/20/24, 7:20 PM

i can give you private lecture or train for you. other than that doing more experiments :d

tazztone•10/20/24, 7:30 PM

@Dr. Furkan Gözükara try this prompt with your LoRA

:

a OHWX MAN standing inside a small room. the walls are made of hundreds of faces staring at him. each face is displaying different strong emotions

Ttazztone @Dr. Furkan Gözükara try this prompt with your LoRA 🥲 : a OHWX MAN standing i...

Furkan Gözükara SECourses•10/20/24, 7:31 PM

it will be all my face

Furkan Gözükara SECourses•10/20/24, 7:31 PM

flux bleeds currently

tazztone•10/20/24, 7:32 PM

yes it will be very weird

tazztone•10/20/24, 7:38 PM

t4zz-faces.txt1.79KB

Furkan Gözükara SECourses•10/20/24, 7:42 PM

ye out of ordinary prompts

FFurkan Gözükara SECourses i can give you private lecture or train for you. other than that doing more expe...

Nick44•10/20/24, 7:44 PM

is there a channel for pricing

NNick44 is there a channel for pricing

Furkan Gözükara SECourses•10/20/24, 7:46 PM

I will message you now

FFurkan Gözükara SECourses I will message you now

ar.1•10/20/24, 10:30 PM

is the Image that you have on massed compute a docker image ?

Aar.1 is the Image that you have on massed compute a docker image ?

Furkan Gözükara SECourses•10/20/24, 10:41 PM

yep

Furkan Gözükara SECourses•10/20/24, 10:41 PM

@NicB makes it

Furkan Gözükara SECourses•10/20/24, 10:41 PM

man tutorial is 2 hours 49 minutes after editing

Furkan Gözükara SECourses•10/20/24, 10:41 PM

Zuck•10/20/24, 11:10 PM

@Dr. Furkan Gözükara After all your experiments do you have any tips getting the best text outputs? Different learning rate or any techniques?

FFurkan Gözükara SECourses yep

ar.1•10/20/24, 11:13 PM

is that on docker hub ?

FFurkan Gözükara SECourses man tutorial is 2 hours 49 minutes after editing

ar.1•10/20/24, 11:13 PM

almost ready ?

dsienra•10/21/24, 2:57 AM

@Dr. Furkan Gözükara Differential Output Preservation must to be activated manually? is not by default I think, and is not on UI? right?

dsienra•10/21/24, 3:16 AM

I asked @kohya_tech

Maxivi•10/21/24, 3:38 AM

you need to put it in your dataset.toml file

Maxivi•10/21/24, 3:39 AM

Maxivi•10/21/24, 3:40 AM

i've done like 12 trainings already testing, works good for sdxl, not much luck for flux still bleeds

dsienra•10/21/24, 3:41 AM

I'm using the gui, the .tom file is created automatically how can I modify it when I start the traing from the ui?

Maxivi•10/21/24, 3:42 AM

no no you need to make your own one

Maxivi•10/21/24, 3:43 AM

the toml thats created automatically is the config one

Maxivi•10/21/24, 3:45 AM

https://github.com/kohya-ss/sd-scripts/blob/sd3/docs/config_README-en.md

GitHub

sd-scripts/docs/config_README-en.md at sd3 · kohya-ss/sd-scripts

Contribute to kohya-ss/sd-scripts development by creating an account on GitHub.

Maxivi•10/21/24, 3:45 AM

more info here

Maxivi•10/21/24, 3:45 AM

Maxivi•10/21/24, 3:45 AM

put your img and reg folder inside the dataset.toml file

dsienra•10/21/24, 3:49 AM

Ok I think I get it, I can get my last .tom file and add the line to it and then load my custome .tom on "Dataset config file (Optional. Select the toml configuration file to use for the dataset)" on the ui

dsienra•10/21/24, 3:50 AM

I'm right?

Maxivi•10/21/24, 3:53 AM

no bro you need to make a new txt file with just your img and reg images in it

Does anyone have experience creating serverless Comfy API endpoint on RunPod ? I keep getting "Max

Similar Threads

Similar Threads

Similar Threads