ComfyUI-GPT4V-Image-Captioner
Nodes:GPT4V-Image-Captioner
Nodes (18)
Turn any image into a tag list with one API call
Load an image and its old A1111 settings in one go
A number that steps forward every time the graph runs
Stop hand-wiring one image at a time
Merge two LoRA prompts and pick who wins per attribute
Rescale bounding-box coordinates from a 512 base
Turn a mask into a real alpha channel
Blur only the part the mask covers
Type a character's Chinese name, get ready-to-use anime tags
Wildcards that also tune CFG, exposure, and skin tone
Turn 01 into AB, the laziest string transformer
Create and rewrite wildcard files from inside the graph
A float in, the same float out
Preview an entire folder in one run
Save anywhere, and save the caption next to it
Shrink an image and letterbox it in black
__pick__ a random line from a text file, that's it
A digital zoom that keeps the canvas size
ComfyUI-GPT4V-Image-Captioner
你可以通过填写KEY和URL快捷地调用GPT视觉模型进行识别和打标。
该项目是我个人对 GPT4V-Image-Captioner repository(https://github.com/jiayev/GPT4V-Image-Captioner) 项目的移植。
感谢GPT4V-Image-Captioner的原作者们。
Input Image Processing:
The project automatically handles image processing, eliminating the need for manual scaling.
GPT4V Integration:
To utilize GPT4V for image annotation, you need to provide the API key and API URL.
Seed and OpenAI Labeling:
The seed value determines the consistency of the labeling process. If the seed and image remain unchanged, there will be no additional calls to OpenAI for labeling. If you are not satisfied with the labeling results, you can modify the seed value to get different outputs.
Prompt Types:
There are two prompt types available: "generic" and "figure". Under the "figure" type, labels will exclude color terms, backgrounds, watermarks, signatures, etc., focusing only on the overall characteristics of the characters.
Weighted Labels:
The option to enable weights for labels is available. When weights are enabled, corresponding weight values will be assigned to the labels.
Excluding Unwanted Words:
If there are specific words in the labels that you want to exclude, you can enter them in the "exclude_words" field. The system will automatically exclude the specified words from the generated labels.