Besides GPT 5.2，ChatGPT Finally, they’ve updated the image generation model too. The brand new GPT Image 1.5 was officially released earlier. Its features and… Nano Banana Pro Pretty much, it has precise editing capabilities and maintains character consistency. More importantly, the generation speed is up to 4 times faster than the old version—my tests confirm it’s really much faster. Although OpenAI It mentions that the text generation part has also improved, but mainly in enhancing “dense text”; Chinese characters are still not very good.
However, there is a solution to this. Given that ChatGPT can now edit precisely, you can just ask it to regenerate the incorrect characters. I tested it and the results were pretty good.
ChatGPT welcomes the new GPT Image 1.5 image generation model, featuring precise editing, consistent character rendering, and significantly faster performance.
Earlier, OpenAI announced the official launch of the GPT Image 1.5 image generation model. This model can make highly precise edits while preserving important details, with generation speeds increased by up to 4x. It’s available simultaneously on ChatGPT and the API, and all ChatGPT users can experience the new model now.
GPT Image 1.5’s editing capabilities are outstanding. Compared to the original version, it’s more reliable in understanding and following instructions, can perform more precise modifications, and maintains the relationships between elements in original compositions. In short, it can edit images according to your needs while leaving all other areas unchanged.
To this end, I also tested it. I used Ada’s headshot photo. I first asked ChatGPT to change the clothing to a suit, and it executed perfectly, preserving all other details, though the skin became slightly whiter:
This is the generated image:
Next, I asked to remove Ada’s watch, and ChatGPT successfully executed it; the watch on his wrist disappeared.
I used the same headshot and asked ChatGPT to generate an image based on the following prompts, and the results were quite good as well.
使用我上傳的圖片，製作一張超寫實電影風格的編輯肖像，比例為16：9。採用低角度拍攝——鏡頭靠近地面，向上拍攝一名男子（使用圖片1），他坐著並靠在一個巨大的、毀壞的鋼彈頭部上。
他穿著一套風化的末日風格編輯服裝：多層深色工裝外套，邊緣破損，配有戰術腰帶、褪色的內襯上衣，以及磨損的戰鬥靴。
This is the result generated by Gemini:
Text generation has also advanced further, capable of handling denser and smaller text content. For example: naturally formatting Markdown content into newspaper articles, precisely rendering headings, paragraphs, bold text, and tables, preserving complete formatting and numerical information in images, and so on.
This is a sample image provided by OpenAI:
I also quickly tested the new model’s Chinese text generation capability, here is the prompt:
幫我生成一張「咖啡買一送一」的海報，上面有「咖啡買一送一」、「限時優惠」、「買越多省越多，還可以寄杯」的繁體中文字，真實風格
Clearly, several Chinese characters are not quite right, which shows it still can’t match Gemini in this regard. However, this is the initial generation, and through subsequent editing and modifications, it can also produce images with accurate Chinese characters:
As shown below, I asked ChatGPT to regenerate the characters “省” and “還” in the image, and the original typos were corrected.
I have to say, ChatGPT’s design and color scheme are excellent. Compared to the image Gemini generated below, I prefer ChatGPT—it has more of a poster vibe:
Additionally, the ChatGPT sidebar has also introduced a dedicated image creation space, with the following features:
- Various preset filters and popular prompts (ten preset filters)
- Regularly update to reflect the latest trends.
- A one-time appearance upload feature that lets you reuse the same look in future creations without having to repeatedly pick from your photo album.
Source: KOCPC Chinese