IMG to Text to IMG by SD3.5+LoRA

New IMG to IMG simple workflow

Instead of using the latent method to generate the classic IMG to IMG image, the IMG to text processing is carried out through the image analysis node, and then a new image based on the reference text information is generated through the text to IMG method. Yes, this is the right way to use IMG to IMG in certain IP-sensitive jobs.

Using a node from miaoshouai/ComfyUI-Miaoshouai-Tagger and it's retrained modelMiaoshouAI/Florence-2-large-PromptGen-v1.5 · Hugging Face, it can achieve a balance between quality and speed at the same time. Maybe a better choice than Florence2 and faster than joy_caption.

Of course, you can also choose to modify it. Infer models for your own vision.

Enjoy it!