Forum Replies Created
-
Posts
-
This is quite an uncharted area because it is a rare use case. You can start with 7-10 images and see if dreambooth works. Otherwise, you will need to experiment with more images or other techniques.
Hi! Very interesting application. This one is tough because I don’t think SD would have seen this in its training.
If you have a good number of sample images, you can try training a LoRA or Dreambooth model with specific prompts.
Forgot to mention that there’s a new tutorial on this
You can try the following to see which one works better
- Add training images with your face in a similar size (pixels)
- Use inpainting (masked only) at 0.5 or so to redraw the face at a higher resolution.
when you use high CFG values like 20, you usually get a color-saturated image. Dynamic CFG thresholding is a technique to constraint the sampling process to reduce the saturation effect.
Hi, it seems that your trained model is not stable. You can:
- Train from SD 1.5 if possible. The model is in a better state compared to realistic vision.
- Try reducing the number of training steps and/or learning rate. Use the least you can get away with.
Great!
Dreambooth on SDXL consumes too much memory. I can’t find a way around it on Colab. You can try training a LoRA on SDXL.
Realistic Vision is a good starting point for what you want to achieve.
Welcome!
Dreambooth is the most powerful method, so it should work.
You can try
- Get better input pictures with quality similar to my training example images.
-
If your face doesn’t show, train more steps.
I’m not aware of such controlnet model.
The SDXL canny controlnet works pretty well.
You use SD1.5 segmentation controlnet to generate something close and use multiple rounds of img2img with controlnet to transform the image to an SDXL style.
