A collection of custom nodes and workflows for ComfyUI
Based off clip_interrogator.
This is a simple CLIP_interrogator node that has a few handy options:
- "keep_model_alive" will not remove the CLIP/BLIP models from the GPU after the node is executed, avoiding the need to reload the entire model every time you run a new pipeline (but will use more GPU memory).
- "prepend_BLIP_caption" can be turned off to only get the matching modifier tags but not use a BLIP-interrogation. Useful if you're using an image with IP_adapter and are mainly looking to copy textures, but not global image contents.
- "save_prompt_to_txt_file" to specify a path where the prompt is saved to disk.
Decodes VAE latents to imgs, but saves them directly to a folder. This allows rendering much longer videos with, for example, AnimateDiff (manual video compilation with ffmpeg required in post).
A basic Image saver with the option to add timestamps and to also save the entire pipeline as a .json file (so you can read prompts and settings directly from that .json file without loading the entire pipe).
NOTE: Some of the included nodes aren't finished yet!