Hello everyone. I am Dr. Furkan Gözükara. PhD Computer Engineer. SECourses is a dedicated YouTube channel for the following topics : Tech, AI, News, Science, Robotics, Singularity, ComfyUI, SwarmUI, ML, Artificial Intelligence, Humanoid Robots, Wan 2.2, FLUX, Krea, Qwen Image, VLMs, Stable Diffusion
you need lora only when you want to publish on civitai so that people can use on different base models - otherwise lora is no, it is inferior from all aspects. also you would want lora if you are service so you want to host small files
Hello everyone! I trained my character, Conan, through finetuning, and now I want to add a LoRA with a film style. However, when working with LoRA from CIVITAI in the style of a specific film genre (e.g., Dark Fantasy movies from the 80s) or a type of lens (e.g., Anamorphic Lens), I encountered an issue: at high LoRA values, it radically changes the character's clothing and appearance or removes objects from the environment, but adheres to the style better. At lower values (around 0.5 or less), it either doesn't work or leaves some noise.
I understand that the authors of these LoRA models often use 30-50 images for style training, which might be the main limitation in the flexibility of LoRA models.
I could create a dataset of 100-500-1000 images for the Dark Fantasy movies 80s style. But tell me, should I train it on my already finetuned model with my character (will it affect his appearance too much?), or should I create a new LoRA with a larger dataset of images (100-300-500) for higher-quality generations? What do you think will work better?
If I want a lora to impact only my characters body, should I crop to only include body in training images my current solution is inpaintinf my face with segment as shown, but face does change when using the lora