As 2024 draws to a close, getting AI to precisely generate text in images with attractive fonts and layouts remains largely an unsolved challenge.
In our previous reviews of Ideogram and Recraft, we noted their capability to handle some English text generation needs. However, we always had to add a disappointing footnote – “Chinese is not supported.” Readers came with high hopes but left disappointed.
This isn’t the fault of overseas AI models – Chinese characters, with their complex radicals and components, are inherently more challenging than English letters.
You need the right tools for the job. Finally, a domestic AI text-to-image product has filled this void: Bytedance’s Doubao.
A 10-Second Meme Generator That Speaks Chinese
Doubao’s Chinese text generation feature is currently available on the app. Using it is straightforward – simply include the text you want in your prompt. Currently, it doesn’t support adding text to uploaded reference images.
Those brainwashed by the simulation game “Shawarma Legend” surely remember that off-key phrase “Let’s speak Chinese” etched in their minds.
Well, now we can recreate a hand-pulled noodle shop sign, but this time with a difference – we can actually use Chinese. The red signboard displays five clear characters, all completed in about 10 seconds.
Doubao isn’t limited to just Chinese – it handles Chinese-English combinations too. A handcrafted-style coffee shop logo springs to life effortlessly.
A line or two of text might not demonstrate much. Let’s increase the difficulty – more text with proper layout.
Drawing inspiration from “Goodbye My Love 4’s” smoked chicken incident, I gave Doubao several requirements in my prompt.
Though the prompt was long, Doubao understood it perfectly, achieving high overall completion with no text errors.
AI generation has inherent uncertainty. After several attempts, I’m most satisfied with this version. Although the font doesn’t completely match requirements, the chicken isn’t quite photorealistic, and small text is garbled, the overall style is pleasing, with the chicken looking glossy and appetizing.
As the saying goes, “When something is so outdated it becomes trendy” – elderly people’s message stickers are a prime example, typically featuring eye-catching text, auspicious elements, and bling effects.
We, who are used to abstract and ironic expressions, can learn from this straightforward spirit. For instance, I had Doubao generate a “One loving family” image.
Cats, koi fish, roses, and sparkles surround gradient-filled large text, achieving about 40% of that vintage feel.
Doubao’s text-to-image has a convenient feature – style selection. Without writing prompts, we can master various styles like film, cartoon, and film photography with just one click.
Combining special styles with precise Chinese text generation can yield efficient results. Using pixel art style, we can have Doubao create a four-panel comic.
My prompt was simple, only specifying the text and scene content for each panel. While Doubao’s imagery slightly deviated, the text was perfect. Everyone can try this format with different themes.
Some might wonder why generating Chinese text is such an important requirement – couldn’t we just generate the image first and add text later in photo editing software?
That approach works too, but having AI directly generate both graphics and text allows for seamless style integration without any photoshop feel, offering endless possibilities. Though we still need photo editing software to clean up garbled small text.
Useful Everywhere – Everyone Can Design Posters and Wallpapers
Beyond creative fun, Doubao can handle practical design tasks. One-click generation of posters, wallpapers, and holiday greetings is now possible.
What’s practical? For me, chat group backgrounds are essential, reminding me not to say inappropriate things in the wrong places.
So I had AI design one, requesting text placement in yellow warning signs. Doubao seemed to understand this working professional spirit.
Worried about New Year’s firecrackers being too loud, I want to send you New Year wishes now, complete with auspicious clouds and gold ingots. For future red packet covers, why not let AI help with design?
“Wicked,” the Broadway musical adaptation starring Ariana Grande, is currently showing in North America and has confirmed its release in China. However, the poster design doesn’t match its popularity.
Let’s have Doubao redesign it. I reverse-engineered prompts from the original poster, explaining the essential elements to Doubao.
At least Doubao’s version includes font design that better matches the fantasy theme and atmosphere, while the original poster looks like it was made with Word and PowerPoint fancy fonts.
Beyond musicals, Doubao can also handle Eastern aesthetic style posters, presenting grand and imposing brush calligraphy.
Beautiful writing is one thing, but more importantly, Doubao follows prompt instructions – whatever style and color of text, wherever we want to place it, it generally complies with our wishes. If not perfect the first time, we can generate multiple versions.
Using Doubao to design Xiaohongshu covers, text fonts and colors can differ at top and bottom, all according to our desired effect.
Doubao can also create realistic product images, with text tilted according to packaging angles, appearing as if printed on the package.
Designing Bilibili long video covers is no challenge for Doubao either, achieving both professional quality and approachability.
So, while the technology is ready, what matters most is whether our creativity is broad enough and our aesthetic sense good enough.
Design work can’t be completely replaced by AI yet, but we should all develop some design thinking so AI can better realize our thoughts and ideas.
Choose Down-to-Earth and Comprehensive AI Products to Solve Problems
Bytedance’s Doubao being the first domestic product to integrate Chinese text generation capabilities comes as no surprise.
Among domestic large model products, Doubao truly stands out as both practical and comprehensive. Each new feature release feels like exactly what users need.
AI writing, AI search, document and webpage summarization – Doubao has all these standard large model features covered.
For AI agents, Doubao offers users extensive choices, functioning both like GPTs for various vertical applications and as an AI companion for chat.
From photo-based problem-solving to voice calls, image generation, music generation, and video generation – Doubao covers all sorts of multimodal functionalities.
Doubao basically covers all our core AI needs while offering broad channel access across mobile and web platforms, making it available whenever we need to chat, whether for work or entertainment.
Notably, Doubao PC version will beta test video generation features this week. Based on our previous review, Doubao’s video generation model, compared to international products like Luma and Runway, better understands Chinese and Chinese users, better fitting Chinese internet creation workflows.
The same applies to image generation. While it’s great that Ideogram and Recraft can generate English text, Doubao’s solution to the Chinese text generation challenge truly addresses domestic users’ pain points.
Unlike pseudo-intellectuals who love saying “you’re so special, you give me a sense of distance,” “Doubao” feels particularly warm and friendly from its name to logo. This warmth extends into product design – Doubao’s features are very user-friendly, emphasizing accessibility. Everyone who tries it can use it effectively.
Chinese is a profound language that poses certain challenges for precise AI generation. In a world with so many people, everyone should be able to design something with AI, regardless of their device or language. AI products should evolve in this direction, and Doubao is the first to deliver a solution for Chinese users.