Datasets:
Native speakers wanted to check the verdicts (Arabic and Japanese first)
Most of the verdicts in this dataset were made by an AI reviewer, and for non-Latin scripts that is not reliable. When a native Arabic speaker read our 291 Arabic first attempts blind, they disagreed with 23 of our calls and settled 34 of the 41 we could not decide.
The rest of the images are open for native review in batches of about 30 images, roughly 10 minutes each. You see the word the model was asked to write, not the model name or our verdict, and your calls go into the next release under the credit you choose. Arabic and Japanese have the most images waiting; Chinese, Spanish, Portuguese, German and English batches are open too.
If you read one of these languages natively, write to hello@aifont.ai.
Details: https://aifont.ai/blog/native-speakers-wanted
Batches: https://aifont.ai/native-review
If you only want to correct one image, open an issue on GitHub with the record id: https://github.com/Funovate-Technology/aifont-text-rendering