Open SCUTlihaoyu opened 3 months ago
Hey @SCUTlihaoyu , thank you for your interest in our work. The "prompt" here actually refers to "content_prompt". We update the training code accordingly. Additionally, to facilitate training, we have provided a JSON file named stylegallery.jsonl that contains all the training data in a uniform format: "image_file":"", "content_prompt":"", .... You can download it from here. Note that stylegallery.jsonl does not provide "content_image_file" paths, but you can process contours using our tutorial available here.
Thank you for clarifying my doubts
I read the training-related code and found that the prompt used in the given training deme is the ‘prompt’ field instead of ‘content_prompt’. Is there an error here? According to the paper, it should be the ‘content_prompt’ without the style, right?