I haven't trained a checkpoint, only LoRAs. From my understanding, the question would largely be around the labeling of the images, minimum resolution depending on the model, and the types/diversity of images (whether its a general checkpoint that has all your images, or those separated based on the types of topics - i.e. wedding, wildlife, fashion, etc.). This article on HuggingFace (huggingface.co/docs/diffusers/en/tutorials/basic_training) demonstrates training a diffusion model based on 1k images from the Smithsonian Butterflies dataset. Sorry this is kind of a non-answer - but thought it may be helpful to share the thoughts above. Best of luck!

@@PromptingPixels Thank You so much, its a first step into teh right direction i think, much appreciated