In AI-generated advertising, the watermark feature often struggles to display certain brand names correctly on images. This issue is particularly frustrating when working with unique or less common brand names.
For example, take the fashion brand oddyssey. Even when using one of the most advanced image generation models, nano banana, the output often fails to render the brand name correctly in the watermark. Despite clear instructions requiring the watermark to display “oddyssey” exactly at the bottom-left corner, the model stubbornly generates variations like oddoyseey. The position may be accurate, the clothing and model may look perfect, but the incorrect spelling makes the image unusable for clients.
Example of the Problem
A studio photo of a person wearing a white, short-sleeved crew neck t-shirt.
On the chest of the t-shirt:
- The exact word "oddyssey" in black, stylized, slightly distorted font.
- Behind the text are three light blue star-shaped designs with a dotted texture.
Background: minimalist studio setting.
Watermark requirement:
- Place the exact text "oddyssey" at the **top-left corner** of the image.
- The watermark must be spelled exactly as "oddyssey".
- Do not autocorrect, do not change letters, do not alter capitalization, spacing, or order.
- The watermark is simple plain text, no effects.
However, the final generated image ends up looking like this:
Flowtra AI’s Solution
So, how does Flowtra AI solve this issue? The key lies in advanced text correction technologies.
ByteDance recently introduced Seedream, which provides highly accurate text repair capabilities in AI images. With just a simple instruction—“replace the bottom-left watermark with oddyssey”—the tool can not only correct the spelling but also preserve the model, clothing details, and watermark style. The result is a flawless, client-ready advertisement.
Detecting Watermark Errors Automatically
Humans can easily notice a spelling error, but how can an AI system do the same automatically? The answer is multimodal large models.
Flowtra AI integrates Google’s Gemini 2.0 Flash model, which can analyze images and detect textual inaccuracies within seconds. The model doesn’t just say whether the spelling is wrong—it can also point out where the error occurs. While ultra-precise localization may not always be necessary, this capability ensures that even subtle mistakes are caught.
Example output:
{
"watermark_text": "oddoyseey",
"center_text": "ODDYSSEY",
"is_spelling_error": true,
"differences": [
{
"position": "after 'odd'",
"watermark": "o",
"correct": ""
},
{
"position": "middle section",
"watermark": "se",
"correct": "ss"
}
],
"correct_spelling": "ODDYSSEY"
}
Improved Workflow with Flowtra AI
By solving the twin challenges of error detection and error correction, Flowtra AI integrates watermark validation seamlessly into its Create Professional Video Ads workflow:
This ensures that advertisers can trust the AI not only to generate high-quality visuals but also to maintain brand integrity through accurate text rendering.
Not All Brands Face This Issue
While unusual brand names like oddyssey are prone to AI spelling mistakes, most simpler names—like Safe Pet—are typically generated without error.
Conclusion
Watermark accuracy is a small detail that makes a big difference in professional advertising. With Flowtra AI’s integration of Seedream and Gemini 2.0 Flash, spelling errors in watermarks can be automatically detected and seamlessly corrected—helping brands maintain professionalism and credibility in every ad campaign.