Software Engineering Courses (SECourses)•3y ago

could you tell us which models did you use to train your lest set for the costumer you said about ea

could you tell us which models did you use to train your lest set for the costumer you said about ealier? when i tried a few models from there it wasnt very accurate. Also, as discussed here before, why use captions when you use regularlazation images?

DDR.Siva even though 90% of models dont need them , some specify the trigger word as ment...

DR.Siva•10/20/23, 2:12 PM

@Dr. Furkan Gözükara Sir are we talking about this

EElisoli could you tell us which models did you use to train your lest set for the costum...

Furkan Gözükara SECourses•10/20/23, 2:13 PM

i use SDXL 1.0 base

Furkan Gözükara SECourses•10/20/23, 2:13 PM

caption didnt affect much

Furkan Gözükara SECourses•10/20/23, 2:14 PM

i think most important part is training images quality and selection

ElisoliOP•10/20/23, 2:14 PM

oh i meant what captioning model sorry

DDR.Siva @Dr. Furkan Gözükara Sir are we talking about this 🤔

Furkan Gözükara SECourses•10/20/23, 2:14 PM

it is a long thing to explain here

EElisoli oh i meant what captioning model sorry

Furkan Gözükara SECourses•10/20/23, 2:14 PM

i used blip2

Furkan Gözükara SECourses•10/20/23, 2:14 PM

you know i shared on patreon

FFurkan Gözükara SECourses i think most important part is training images quality and selection

ElisoliOP•10/20/23, 2:14 PM

actually the set you helped me in private (eliminated a few images) - turned out my best model ever (out of around 10 made)

DR.Siva•10/20/23, 2:14 PM

ok ok , i will just go through them.. i thought you are speaking about the captioning.. NVM , will look into your tutorials ..

FFurkan Gözükara SECourses i used blip2

ElisoliOP•10/20/23, 2:15 PM

i understand its blip2 but when opening the captioning webui there's a really huge list of models thats the one im talking about

DDR.Siva ok ok , i will just go through them.. i thought you are speaking about the capti...

Furkan Gözükara SECourses•10/20/23, 2:15 PM

as you watch more you will understand more haha

i can also make private session and explain you if you are interested in professionally

ElisoliOP•10/20/23, 2:15 PM

but if you say captioning didnt matter much it doesnt matter

EElisoli i understand its blip2 but when opening the captioning webui there's a really hu...

Furkan Gözükara SECourses•10/20/23, 2:15 PM

blip2 is stand alone

DR.Siva•10/20/23, 2:15 PM

I am now seeing your controlnet tutorial sir. Going ot to take the nexct step into controlnet , and then dreambooth as you suggested

Furkan Gözükara SECourses•10/20/23, 2:15 PM

https://www.youtube.com/watch?v=V8iDW8iprqU

YouTubeSECourses

Best Open Source AI Captioning Scripts For Image Captioning - Use O...

Scripts download link : https://www.patreon.com/posts/sota-very-best-90744385

DR.Siva•10/20/23, 2:16 PM

https://www.youtube.com/watch?v=3E5fhFQUVLo

YouTubeSECourses

Zero to Hero ControlNet Tutorial: Stable Diffusion Web UI Extension...

I show how to install Automatic1111 Web UI & ControlNet extension installation from scratch in this video. Moreover I show how to make amazing QR codes and inpainting and out painting of ControlNet which are very similar to Photoshop generative fill and Midjourney zoom out. Furthermore, I explain and show what are Canny, Depth, Normal, OpenPose...

ElisoliOP•10/20/23, 2:28 PM

I've actually tested so many models but forgot to document (or rather was in a hurry from excitment) - is there a way to take a checkpoint and somehow check it to see if i used captions to train it?

LEGACIES•10/20/23, 3:46 PM

@Dr. Furkan Gözükara training with dreambooth with lion, can I use it without using classification images?

LLEGACIES @Dr. Furkan Gözükara training with dreambooth with lion, can I use it without us...

Furkan Gözükara SECourses•10/20/23, 3:49 PM

yep

LEGACIES•10/20/23, 3:52 PM

for this setting i set it to 0 or leave it to 5?

LLEGACIES for this setting i set it to 0 or leave it to 5?

Furkan Gözükara SECourses•10/20/23, 3:58 PM

dont use it

Furkan Gözükara SECourses•10/20/23, 3:58 PM

use my class images

Furkan Gözükara SECourses•10/20/23, 3:59 PM

or generate from txt2img tab

Furkan Gözükara SECourses•10/20/23, 3:59 PM

manually

LEGACIES•10/20/23, 4:01 PM

for 30 images how many images in my classification dataset and how much "class images per instance images" should i set it to?

mikemenders•10/20/23, 4:04 PM

1/1 - for one classification image one instance image. No more need. Use Furkan's reg images.

mikemenders•10/20/23, 4:06 PM

After the recent Prodigy testing, I have now styled with AdamW. Using the same image set. So far the Prodigy has won.

mikemenders•10/20/23, 4:11 PM

Here is my config for style. I used 10 images with 10 repeats, captions and 10 epoch

Mykee_-_Prodigy_-_SD15_-_style.json2.97KB

Mmikemenders Here is my config for style. I used 10 images with 10 repeats, captions and 10 e...

Furkan Gözükara SECourses•10/20/23, 4:11 PM

nice

LLEGACIES for 30 images how many images in my classification dataset and how much "class ...

Furkan Gözükara SECourses•10/20/23, 4:12 PM

with dreambooth extension

Furkan Gözükara SECourses•10/20/23, 4:12 PM

if class images are good quality

Furkan Gözükara SECourses•10/20/23, 4:12 PM

i suggest 100 per instance

Furkan Gözükara SECourses•10/20/23, 4:12 PM

so you would need 3000

LEGACIES•10/20/23, 4:12 PM

i will use your class images

LLEGACIES i will use your class images

Furkan Gözükara SECourses•10/20/23, 4:12 PM

ok make it 100

LEGACIES•10/20/23, 4:12 PM

ok so i set it to 100 and use 3000 images
thank you

LEGACIES•10/20/23, 5:48 PM

@Dr. Furkan Gözükara my dataset is 512px x 768px and I have trained max resolution 768px. Any particupar reason why the generated images have more ressemblance with my dataset when generate at 768px x 768px compared to those generated 512px x 768px?

Greg•10/20/23, 7:03 PM

Hi all, I'm a brand new Patreon supporter. I'm following the tutorial to make a LoRA training of myself in Kaggle. But I get the following error:

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 14.76 GiB total capacity; 13.17 GiB already allocated; 7.75 MiB free; 13.46 GiB reserved in total by PyTorch

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 14.76 GiB total capacity; 13.17 GiB already allocated; 7.75 MiB free; 13.46 GiB reserved in total by PyTorch

Here is my training command:

accelerate launch --num_cpu_threads_per_process=2 "./sdxl_train_network.py" --pretrained_model_name_or_path="stabilityai/stable-diffusion-xl-base-1.0" --train_data_dir="/kaggle/working/results/img" --reg_data_dir="/kaggle/working/results/reg" --resolution="1024,1024" --output_dir="/kaggle/working/results/model" --logging_dir="/kaggle/working/results/log" --network_alpha="1" --save_model_as=safetensors --network_module=networks.lora --text_encoder_lr=0.0004 --unet_lr=0.0004 --network_dim=32 --output_name="kaggle_glpr123" --lr_scheduler_num_cycles="8" --no_half_vae --learning_rate="0.0004" --lr_scheduler="constant" --train_batch_size="1" --max_train_steps="6400" --save_every_n_epochs="1" --mixed_precision="fp16" --save_precision="fp16" --cache_latents --optimizer_type="Adafactor" --optimizer_args scale_parameter=False relative_step=False warmup_init=False --max_data_loader_n_workers="0" --bucket_reso_steps=64 --full_fp16 --xformers --bucket_no_upscale --noise_offset=0.0 --lowram --max_grad_norm=0.0

accelerate launch --num_cpu_threads_per_process=2 "./sdxl_train_network.py" --pretrained_model_name_or_path="stabilityai/stable-diffusion-xl-base-1.0" --train_data_dir="/kaggle/working/results/img" --reg_data_dir="/kaggle/working/results/reg" --resolution="1024,1024" --output_dir="/kaggle/working/results/model" --logging_dir="/kaggle/working/results/log" --network_alpha="1" --save_model_as=safetensors --network_module=networks.lora --text_encoder_lr=0.0004 --unet_lr=0.0004 --network_dim=32 --output_name="kaggle_glpr123" --lr_scheduler_num_cycles="8" --no_half_vae --learning_rate="0.0004" --lr_scheduler="constant" --train_batch_size="1" --max_train_steps="6400" --save_every_n_epochs="1" --mixed_precision="fp16" --save_precision="fp16" --cache_latents --optimizer_type="Adafactor" --optimizer_args scale_parameter=False relative_step=False warmup_init=False --max_data_loader_n_workers="0" --bucket_reso_steps=64 --full_fp16 --xformers --bucket_no_upscale --noise_offset=0.0 --lowram --max_grad_norm=0.0

mikemenders•10/20/23, 8:11 PM

If I'm tagging with WD14 and it's a drawing style, but I don't want it to be an asian/manga type model, which one should I train on?

mikemenders•10/20/23, 8:35 PM

MeinaMix maybe?

LLEGACIES @Dr. Furkan Gözükara my dataset is 512px x 768px and I have trained max resoluti...

Furkan Gözükara SECourses•10/20/23, 8:44 PM

which model? also did you enable bucketing?

GGreg Hi all, I'm a brand new Patreon supporter. I'm following the tutorial to make a ...

Furkan Gözükara SECourses•10/20/23, 8:44 PM

hello welcome

Furkan Gözükara SECourses•10/20/23, 8:45 PM

gradient checkpointing missing.

Mmikemenders MeinaMix maybe?

Furkan Gözükara SECourses•10/20/23, 8:45 PM

i think you need to test on models

akellio•10/20/23, 10:12 PM

Hi, it is possible to use own lora in rerender?

akellio•10/20/23, 10:22 PM

and is it possible to make the first keyframe separately? (made in the automation interface, for example)

Aakellio Hi, it is possible to use own lora in rerender?

Furkan Gözükara SECourses•10/20/23, 11:15 PM

i think yes

Aakellio and is it possible to make the first keyframe separately? (made in the automatio...

Furkan Gözükara SECourses•10/20/23, 11:15 PM

i am not sure but probably possible

Furkan Gözükara SECourses•10/20/23, 11:15 PM

it is open source

accelerate launch --num_cpu_threads_per_process=2 "./sdxl_train_network.py" --pretrained_model_name_or_path="stabilityai/stable-diffusion-xl-base-1.0" --train_data_dir="/kaggle/working/results/img" --reg_data_dir="/kaggle/working/results/reg" --resolution="1024,1024" --output_dir="/kaggle/working/results/model" --logging_dir="/kaggle/working/results/log" --network_alpha="1" --save_model_as=safetensors --network_module=networks.lora --text_encoder_lr=0.0004 --unet_lr=0.0004 --network_dim=32 --output_name="kaggle_glpr123" --lr_scheduler_num_cycles="8" --no_half_vae --learning_rate="0.0004" --lr_scheduler="constant" --train_batch_size="1" --max_train_steps="6400" --save_every_n_epochs="1" --mixed_precision="fp16" --save_precision="fp16" --cache_latents --optimizer_type="Adafactor" --optimizer_args scale_parameter=False relative_step=False warmup_init=False --max_data_loader_n_workers="0" --bucket_reso_steps=64 --full_fp16 --xformers --bucket_no_upscale --noise_offset=0.0 --lowram --max_grad_norm=0.0

could you tell us which models did you use to train your lest set for the costumer you said about ea

Similar Threads

Similar Threads

Similar Threads