Metadata and Image Description to TXT
This package contains two workflows. One provides a way to obtain descriptions via locally hosted LLM support with Ollama open source free downloadable software and LLMs, and these will obtain text descriptions of any image, plus a structured metadata output, saved in a folder in your ComfyUI output folder, and I went with the first tool I found that could do this, which was Griptape. The workflow also contains a means of copying a URL, pasting it in and using that to obtain a description, but this bypasses the metadata collection side of the workflow.
The second workflow is expressly for collecting metadata and has lower requirements.
The Metadata
Resolution, File Size, Aspect Ratio,
Resources Used (Checkpoint/Model AND LORAS) (yes, even if CivitAI says it can't detect them)
SAMPLING SETTINGS:
Seed, Steps, CFG Scale, Sampler, Scheduler, Denoise,
PROMPTS Positive and Negative
Descriptions:
Descriptions provided are based off of prompts. By default, the VRAM usage of the LLM you host locally shuts down after 240 seconds. If you are running on NVMe, these models load FAST, so you may consider adjusting this time to suit your preferences.
The minimum requirements for generating Descriptions + Metadata:
ComfyUI-Simple_Readable_Metadata-SG
ComfyUI-various
and either Ollama local or an API key setup via Griptape to a provider like OpenAI, Gemini, Grok, or whichever you prefer or have access to. Further instructions are contained within the workflow, including YouTube videos that should demystify the process. These workflows will not be updated and are provided as-is, as a way for users to obtain data from images.
If you just want Metadata, just use the Meta to TXT workflow. If you don't want it to generate a bunch of files, you should be able to figure out what to do (passthrough/delete). here is the workflow:
ComfyUI-Simple_Readable_Metadata-SG
ComfyUI-various - You probably already have this
Simple Readable Metadata requires 2.0 nodes turned off.
They have a node for separating out individual aspects of the metadata to text (positive, negative, loras, etc) , allowing you full control over formatting and what gets included, but how you want to use that is entirely up to you and defined by the intent and purpose. There is also a metadata collector node for video, but if your video's metadata was not saved during generation (as is often the case), then this won't help. For this reason, I recommend saving the prompt to a text file every time you generate so you at least have an archive of prompts to refer to so that you can meet CivitAI's minimums for posting AI generated videos.
First watch the YouTube videos if you're running into problems setting up Ollama (links in the notes of the workflows themselves). I have no advice on doing this via API although yes, I hear they're free ones out there. If you have any other questions, feel free to ask!