Quote from: PZF on October 25, 2018, 08:35
The suggestions it gives are terrible if you are not a native speaker. I put in: Three apples isolated on white. It gave these as GOOD keywords:

fruit, raw, attractive, tasty, brown, macro, nutrition, plant, freshness, vegetarian, fashion, black, pretty, single, ripe, delicious, cute, leaf, hand, lifestyle, group, vegetable, snack, business, face, bright, concept, look, design, vitamin, gourmet, model, blue, summer, close, male, meal, cheerful, juicy, dessert, cut, decoration, colorful, close-up, style, cook, orange, lady, hold, care

'isolated on white' is terrible words in description. I did not remove such kind of words (there are some more like New York and Los Angeles) which are highly used over all trained descriptions. Thus, my engine tried to match 'three apples' and those apples can be found in lot of photos...

Quote from: luiscastor on October 23, 2018, 15:30
I like it, but apparently it only works with photos. I will like to use it with illustrations but I get zero keywords with them

Currently i processed only photos, sorry. May be later i add illustrations processing. Currently i checking is this interesting for somebody to continue this project.

Hi

I've published recently my project for keywording images for microstocks. Its name oiken.space.
May be it can be helpful for somebody.

I developed it for myself, but later decide to make it available for everyone.

One rule: Better description - better keywords.
I.e. you should enter good description, like you describe your image for microstocks. And then you get relevant keywords. If you enter only few words, there will be lot of irrelevant keywords.

Any thoughts and suggestions are welcome.