Furkan, if my dataset only contains 5-10 images, can the repeat be set higher than 100?
Furkan, if my dataset only contains 5-10 images, can the repeat be set higher than 100?
F
Vokay let me see,
Vanother question to name the image, ohwx_man1, ohwx_man2 like this?
Fnope
Fit doesnt read image name
Vhow many images in your dataset, you set repeat 100? more than 30 or 20-30
Fbetween 10-20
Fi had 13
Vif less than 10, can 150 repeats be used?
Fsure . make sure to compare checkpoints
Fto see
Vokay thanks
Vi see your trigger word like ohwx man, then i follow your way like ohwx cat, ef
Veg
MI really don't understand why it's better to train dreambooth instead of lora, when
- larger size
- can be trained more slowly
- you are bound by the limitations of the basic model
- requires much more resources
While training the Lora
- fewer pictures are enough
- faster (about 15 minutes for an SD 1.5 model, SDXL 1 hour)
- requires fewer resources, can be trained on local machines
- can be used with multiple base models
- the same quality can be achieved for persons as for dreambooth.
So I used to train dreambooth because I don't need it, I can train what I want on Lora, be it style, object, person or concept. In very special cases I would only train basic model.
So why do you train dreambooth instead of lora?
- larger size
- can be trained more slowly
- you are bound by the limitations of the basic model
- requires much more resources
While training the Lora
- fewer pictures are enough
- faster (about 15 minutes for an SD 1.5 model, SDXL 1 hour)
- requires fewer resources, can be trained on local machines
- can be used with multiple base models
- the same quality can be achieved for persons as for dreambooth.
So I used to train dreambooth because I don't need it, I can train what I want on Lora, be it style, object, person or concept. In very special cases I would only train basic model.
So why do you train dreambooth instead of lora?
Fbecause dreambooth many times better
Chey, please help if anyone knows. Why doesn't the downloaded controlnet appear web ui runpod?
MBetter at what? I ask seriously.
MBecause the Lora is always adapted to the capabilities of the base model, so if the base model is good, the Lora will be beautiful.
MJust because there is no EMA training in lora, but there is in dreambooth?
MWhen I take Lora pictures of people under SD 1.5 that are almost 1:1 with the original image, I wonder why I need to make a dreambooth?
MBack when Lora came out, even I was training dreambooth. Then I took the Lora out of it, and the quality of the resulting Lora didn't deliver what the dreambooth could. However, the loras that have been trained recently can do what the extracted lora could not. The new optimizers give very good results with loras. When you train a model with dreambooth, you only train a small part of the whole model, not the rest. And Lora is just that small part. If I were to train a separate dreambooth model for each person, I would need several, many terrabytes of hard disk. Even the Lora's take up a lot of space.
Fquality
Fyou can use my auto installer
For follow this tutorial
FYouTubeSECourses
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...
FVFurkan, see my messages to you, erros in checkpoint training on kaggle, thanks
AAny idea why rttensor is giving this weird message
Fyou are trying to train refiner
wont work
wont work
Fi think working
Fhit enter and wait
CThanks, I'll try
Vyour default .json will not be changed at all?
AOh cool ok! So now do I wait for the output section to finish saying processing?
A
Fyou can change
Fbut refiner cant be trained
Fno script supporting it
Fit is ready
FAOh wow, so now my images take about15 seconds with rt tensor
Fit almost double speed
ANo onnx file found please check the progress in the terminal. Thaty's what it says when installing on my other pc
Fyou need to generate again
ASays the same thing when hitting the button
Fsame what?
ANo onnx file found. I tried closing out of stable diffusion entirely but now stable diffusion won't open so I deleted VENV File and I will try again