generated them from SD1.5 for women with clothes, that are different from the clothes that i am training

@@AI-HowTo Ok! Thx!

@@AI-HowTo Great tutorial! Do you have an tutorial for training poses?

you are welcome, no i have not created any, but poses are often not trained, they are guided using Open Pose (dw) Control net, in which you present the skeleton and it gives you the pose perfectly including fingers, such as in kzbin.info/www/bejne/Z2TJmHV4fpKNjrc

@@AI-HowTo thank you! I will definitly try that. Thats one of my topics to learn... but i want to train my ai-influencer for specific poses, where I dont want to do every image with control net and open pose... for a more convenient and efficent workflow. So the question for me is how can i achieve that in a trained loRa... or an own checkpoint...

Glad to hear that

I am sorry, not at the time being... you might want to check this video settings which might be of help kzbin.info/www/bejne/rHKVp2V_qsx-bNk and have more recent update on character training ... it even shows how to train 1024x1024 image sizes without running into CUDA errors ... usuaully this error happens if the image size is too large ... i did training on my 8GB RTX 3070 Laptop card without having this issue...but this worked only for SD15, but for 1024x1024 images i get CUDA error for SDXL ... you just need to recheck you image dimenstions once again to make sure that all are proper.

great to know... yes ofcourse, especially for Video generation, the use of a LoRA can increase the consistency of the video generated... and depending on the denoising level, the output would change to become more affected by the LoRA or less.

Yes, absolutely, as both use the same Controlnet models, and the same principles, the application methodology changes... what matters is to understand what these models do then discover how they can be used in the target tool such as A1111/ComfyUI

ti is more powerful than mine (laptop 3070) so you should get better speed than me, if you are training on same image sizes as me with this speed then xformers options might not be turned on, or the drivers require some update i think not, possibly.

sorry, not doing that at the time being, thanks for offering though.

:), ya, i got that note from so many peopel, to be honest at that time i didnt know she was an Artifical character herself, I just wanted a sample data set with so many images that looks real, any way, things still apply.

have not done any recent trainining, and usually the algorithm is fixed so learning rate changes are unlikely to been used differently, not sure, anyway, if you see things overfit quickly then using smaller number of steps could be better, besides, learning rate smaller than 0.0001 doesnt make much sense i think, so we usually consider increasing it not decreasing it to learn faster for instance...not sure if any recent changes in Kohya have made things different.

yes as long as the training data set are of good quality and do not contain deformations in the eyes or fingers, even slightest deformation could be augmented after training if they repeated. as for the best check point, not sure unfortunately at the time being, previosuly for 1.5 i got the best results with majicmix v4, even for wester characters despite that the checkpoint was asian and for SDXL the Juggernaut XL , not sure now....I think in general the principles of training do not change overtime, so the video is still good to rely on for training.

if you disabled and wanted to enable it again, you follow same steps and change the start up type to (Automatic) then click Start from services.

not sure, but i think it is the same file but with .sh extension, and git reset is a command that is independent of the operating system.

I used one image to test the output only, once satisfied with the results, we go to Batch tab and just filled (input directory with source images, and output directory for the output) and batch will generated all the images based on the prompt/details on my test on the single image

sometimes deleting the venv folder inisde A1111 installation could help solver lots of errors, some A1111 are buggier than others.

Glad you find it useful, you are welcome.

Yes it is possible, but it will be slightly slower than you hope for, it can take 20 seconds and more for a 1024x1024 image using Forge UI or using ComfyUI and more ... currently with how AI stuff is heading, if you plan to buy something, you better save and buy 24GB VRAM, it is very expensive, but it is the only option that allows you to run everything such as Animate Diff models without suffocating on memory or suffering slow generation

@@AI-HowTo Then I'll just have to buy a PC good choice.

yes, PC is alot more practical, cheaper, and more powerful, avoid laptops unless you extremely need to move around often, even lower VRMAM RTX 3060 for PC is alot more powerful than its Laptop counterpart and has more VRAM.

not sure, for me it worked without any problems, just downloaded the IP Adapter Face id models into the Controlnet models folder, and the Face ID loras into the LoRA folder, and made sure Controlnet was upto date, and it automatically downloaded necessary extra models related to insightface such as buffalo_l , not sure, why some have troubles with this while others dont.

I have not seen this issue, before, but it could be due to low memory in your PC which forces the A1111 to load/unload models on multiple stages, not sure, I suggest to try github.com/lllyasviel/stable-diffusion-webui-forge which has same UI as A1111, but has better memory management and might run faster and automatically detect best memory settings that could run on your PC.

@@AI-HowTo thanks for reply it was working fine for months and broke out of nowhere will wait a week before fresh installing.

i see, you can try using Git pull (in the command line after set COMMANDLINE_ARGS= line as well, maybe that helps bring up any updates and fix the problem, if auto update is not enable in your installation

great to know, thanks

I think, for perfect consistency, Blender (3D Software) is the right tool, stable diffusion is not the one for perfect consistency of objects/faces/clothes, while one can still achieve good resutls with it.

Thank you. Do you have video on that you can add link yet? Also, would it allow me to create photorealistic portraits like for social media, as that is my goal.

sorry, Blender is a 3D Free software tool, it is for games and realistic 3D stuff and not AI based.... for social media stuff such as creating realistic pictures, then Stable diffusion is really good to create photo realistic images, you can check other videos in this channel hopefully you find something useful, the only drawback for stable diffusion i think is just the Computing power (good graphics card) which makes it difficult for most people to really dive into it and it's capabilities.

kzbin.info/www/bejne/rHKVp2V_qsx-bNk this video shows how one can create a LoRA model that can be used for instance for creating anything related to a specific photo realistic character ... two LoRAs could also be combined to create a new Character with features from both using specific ratio of both ... or one can use the same LoRA with 0.7 weight to get a different variation of the same character... there are no limits on what one can do .... if you were new to Stable diffusion then kzbin.info/www/bejne/iKXNdaatl7h4e7s could be a good starting point ... and you may want to check other creatos content such as www.youtube.com/@sebastiankamph Sebastian has lots of content about creating stuff or www.youtube.com/@OlivioSarikas Olivio start is also fun to watch ... my content contains just limited number of videos and mostly about techniques such as LoRA creation, training, or provides tools and techniques to create certain stuff that other creators may not have covered it properly or in depth.

@@AI-HowTooh excellent, thank you very much 😊

There is no rule... everything is purely experimental, even the creators of Stable diffusion do not know. I think that close up and upper body shots should have the same number to have a balance in training. we would usually have few full body shots such as 10% and most shots are protraits/close up/upper body You might be interested in watching this newer video kzbin.info/www/bejne/rHKVp2V_qsx-bNk which uses smaller number of images... as far as i have seen, full body shots are difficult to reproduce with high quality, this is why we often use After detailer to repaint the face (With prompt that has face with the LoRA inside) some people train only close up in a separate lora and upper body in another lora to get better results as for full body, since they will never function as perectly as one wants, they are best to have small ratio such as less than 10% (For realistic models in which details matter). experimenting is the key eventually, some models may work from first experiemnt, others might take 10s till you get something good, even use different regularization images could afffect the output greatly.

@@AI-HowTo Thanks you. It's helpful.

true, XL is certainly better, but I still dont use it unfortunately on my 8GB video card.

even if the image doesnt have a face, Reactor will output the image (based on my batch image tests), not sure why this is not working out for you, you should do some tests for 5 images for instance to see where the problem happens

not sure, but i think its the same file on mac but named webui-user.sh and you edit it in the same way and put the command there, if you go throught he .sh file it might give you some guidance

I think I did that at some point

I think it is (ip-adapter_sd15.bin) ... all 1.5 models are in huggingface.co/h94/IP-Adapter/tree/main/models

@@AI-HowTo I work with Google Colab, could you tell me which folder to place this file in?

I think it should be the same as local installation folder --- which is the Controlnet model's folder, on my local installation that is stable-diffusion-webui\extensions\sd-webui-controlnet\models ... but i think A1111 also looks inside stable-diffusion-webui\models\ControlNet folder as well

@@AI-HowTo I downloaded that but i don 't have the preprocessor ip_adapter_clip_sd15

it is fine, the developer might have removed the clip version or renamed it, you can use any other sd15 version and try them out.

1) from sample images config section, as in kzbin.info/www/bejne/rXu7ZZV4qamlb7s we choose for instance 1 for epochs to generate 1 image each epoch, and we write in sample prompts the prompt that we want to dispaly, it must be written as shown in the sample text completed including the image size to dispaly 2) yes, as in kzbin.info/www/bejne/iIWVm51jatlrrpo training guide which is a style, this helps train styles/patterns rather than objects -- and yes we eliminate anything related to the pattern in the image description for the training images (leopord or animal pattern ) and keep everything else in the image description

the web ui is built on top of python scripts, so everything in stable diffusion image generation or training is based on scripts, so they can be automated, but i have not used that unfortunately so i dont have enough expertise to guide you on that

this video is from few months ago, back then, dreambooth training was easier on A1111 and i think it appeared on the GUI by default... if that is what you are asking about.

You're welcome!

You're welcome!

not sure, this could indicate a problem with onnxruntime installation possibly, or that you are using frames with large dimensions which results in longer processing time, check the frame dimension first, I have not faced this issue before this is why i cannot advise you more that on the matter, i didnt even had any troubles with onnxruntime installatino before... but when it is slow, then there is an issue.

@@AI-HowTo i really don't know and can't find a solution cuz i've tryed multiple command args but none of them reduced my generation time, imagine 9 seconds took 25 minutes (220 frames) my specs are : rtx 3060 12gb / i7 3th gen / 24gb ram / normal 1TB HDD. my command args : --xformers --no-half-vae

I think this is a good speed for applying ReActor through A1111, 25minutes for 220 frames ... sometimes the size of the image used in ReActor also may affect the output... but 25 minutes it means it is using the GPU and I dont it gets faster than this using A1111.... anything related to AI or face swap can take a while to produce a decent result and creating a long video could take hours or days depending on the quality and length of the clip.

my video card is RTX 3070GB Laptop , it can swap and generate using ReActor at speed of 15 images per minutes, the source video input image sizes are 720x1280, and i used 512x512 face input image in this test.

@@AI-HowTo thank you for your response and assitance, i think it's okay to be patient to get a 2 minutes video in arround 3 or 4 hours. better thank nothing. i was just asking myself how other famous apps give me results within few minutes even if the video is long, maybe they have multiple computers and they split the input for each computer and re-link the parts later as one video. MAYBE

yes, anything that we want to be part of our LoRA must not be captioned... for example, if our subject always have green eyes, then we better not include that in the captions... if she has blonde hair, then we better not include that in the captions either ... we caption only the things that changes, for example: her clothes, the background...etc.... it is a bit confusing... but this is the best way to caption things and improve accuracy for the LoRA model.

@@AI-HowTo hmm that’s interesting. So you saying that anything which is not consistent among images should be not be captioned/tagged. For example if I am training some background images as a style whose theme is neon light cyberpunk, so I should not caption cyberpunk or neon light as it will be consistent among the images but I could caption a group of people for example or tall building because they will not be always there

exactly...we can also add a trigger word that refers to your style (which absorbs all things related to your style common features in all the images) for example which may help increase the chance of bring that style out when prompting too.

@@AI-HowTo I think I understand some of it even if it sounds somewhat confusing. Thanks mate😀

you are welcome.

it will be impractical to run on 2GB, with A1111 this might not work propertly, but Forge have better automatic memory management github.com/lllyasviel/stable-diffusion-webui-forge this repositry provides same things as A1111 with same UI but has better memory management and can run on 2GB for SD 1.5.... it might run SDXL too but it will use the CPU then which will be slow.

@@AI-HowTo yes slow these days is unusable.. the gt1030 was the dumbest purchase i have ever made 2gb idk. if it is the drivers but i cant get it to render or transcode anything..i found a quadro p2000 for cheap so i will run with that thank you for fast response

You are welcome, these days, RTX graphics cards are game changers, they are the way to go for AI/Gaming/3D, they are expensive, but they seem like the only option to save time and be able to stay uptodate with the technolgoy...best of luck.