FOSS Hashtag Generator
from Joelk111@lemmy.world to selfhosted@lemmy.world on 10 Nov 19:35
https://lemmy.world/post/38619576
from Joelk111@lemmy.world to selfhosted@lemmy.world on 10 Nov 19:35
https://lemmy.world/post/38619576
I recently joined Pixelfed and, considering there’s no algorithm, hashtags are the only way to be discovered.
I hate hashtag optimizing, but I also don’t want to upload my image to someone else’s random server before posting it to pixelfed, just to generate hashtags. Where should I look to find something I can host myself, or even something that runs natively on Android/Linux, that’ll generate hashtags/keywords for an image?
#selfhosted
threaded - newest
#SeenfromLemmy
#WhoNeedsHashtags
That sounds like a job for AI, I think. I’m hopeful that you can find a good solution!
Yeah, GPT or whatever can do it, but I was hoping for something a bit more lightweight than what we’ve come to know as AI, as I don’t want to invest in new hardware.
If you want software that analyses an image and produces tags that describe the image’s content, you want a combination of image recognition, image interpretation, text generation and a promp generator to link these three parts. There is no way that any single of these applications can be “lightweight”, not to speak of the combination.
Just write the hashtags. You know what’s in the photo you took, write like 5-10 and you’ll be fine. Once you do it more often you’ll find the active hashtags.
Some autocorrect bot won’t do that accurately.
This is the way.
Top 10 reddit moment: worthless agreeing meme response to an unhelpful comment that doesn’t even answer the question asked.
This does a great job of not answering the question I guess. I obviously am aware that I can write them myself, but I don’t want to. I like automating things.
I don’t know what’s worth tagging and what isn’t. I don’t know which tags are popular and so have followers vs. which ones aren’t.
The information exists but isn’t presented anywhere convenient.
If I have to do homework before posting a picture I’m massively more inclined not to bother, which isn’t particularly good for the Fediverse.
I would love something that could give me suggestions for relevant and popular hashtags for my content.
Ugh, hashtag optimizing is soul-sucking, but you do not need to hand your photos to some random web service to get tags. Run a tiny image-caption/tagging model locally and convert the results into hashtags. My go-to is clip-interrogator, it runs on your own machine, spits out concise keywords and prompt-like descriptions, and is literally made to extract useful phrases from images. Github: github.com/pharmapsychotic/clip-interrogator
If you want something more generic, use BLIP (Salesforce) image captioning via Hugging Face, e.g. the Salesforce/blip-image-captioning models. Install with pip, run the model to get a caption, then use a simple POS filter or spaCy to pull nouns/adjectives and prefix them with #. That workflow is trivial to script and keeps everything local. Models will run on CPU but are far faster with a GPU.
If you absolutely need Android, use Termux and a lightweight model or run the model on a tiny home server and call it from your phone, do not upload to third-party servers. For anime art, DeepDanbooru is the standard local tagger. For everything else, clip-interrogator + a tiny post-processing script is your best bet.
Honestly, stop treating hashtags like SEO black magic, generate sensible descriptive tags locally, and move on with your life. If you want, I can paste a minimal Python snippet that takes an image, runs BLIP or clip-interrogator, and outputs a ready-to-paste list of hashtags. Which model do you want to try, BLIP or clip-interrogator?
Sounds like a good project!
It is starting to sound like it’d have to be haha
Ollama + Open web-ui could handle this.