Spaces:
Running
on
Zero
Enhancing Text-to-Image AI with Turkish Language Support
Text-to-image AIs are constantly being updated and can now generate text in languages like Japanese, English, and even Turkish.
What I mean is, it would be beneficial if you could train this model on some Turkish texts to improve its understanding of the language. This is because it doesn't currently recognize specific Turkish characters such as "ı, ş, ö, ç, ğ, ü."
These characters are essential when creating a text-to-image AI dataset.
Look closely at the example in the picture; it doesn't use the Turkish characters i, ı, ş, ç, ö, ü, ğ in the text.
I think it's safe to assume that you should include all common characters, but i think accuracy might drop because of different fonts used in images. Wouldn't this also require a multilingual model like aya expanse to verify whether the text is correct? Maybe a pass with OCR would also work, which you could feed into the model alongside the visual aspect?
Thank you for the suggestion; I would love to support more languages. However I'm an indie developer so I don't have the resources necessary to do any language other than English justice. I'll try to work more in in future releases as best I can.