AI Image Generation Model Better Than ChatGPT

V7, MidJourney's model, is said to be able to create details like hands more consistently. Photo: Xeophon .

Midjourney, one of the first AI-powered image generators on the web, just launched its first new model in nearly a year.

The model, called V7, was released in beta at midnight on April 4 (US time). A week earlier, OpenAI had launched a similar feature with ChatGPT and quickly created a buzz on social media with the new trend of creating Ghibli-style photos.

If you’ve never used Midjourney before, you’ll first have to rate the 200 images it suggests based on your preferences. It will then create a personalized profile that will create photos that match your aesthetic. The V7 is Midjourney’s first model to have personalization enabled by default.

You can enable or disable V7 by clicking on the “version” toolbar on the website. Additionally, this model can be used in Midjourney’s Discord server.

Midjourney CEO David Holz described V7 as a “completely different architecture” in a post on X. “V7 is much smarter when it comes to handling text commands,” Holz continued on Discord.

He commented that the model also has the ability to process the original image quite well. The image quality is significantly improved with high perfection. From body parts such as hands, feet, to freckles, ceramic patterns are consistent to every detail.

Some sample images created from the V7 model. Photo: X/MidJourney.

In addition to the two modes: Turbo (which costs more) and Relax, the V7 supports a new tool called Draft Mode, which allows for image creation at 10 times the speed and cost of the standard mode. Images in Draft mode will have lower quality, but users can upgrade and recreate the image with just one click.

According to Holz, some of Midjourney's standard features are not yet available in V7, including upscaling and retexturing. He said these features will be added in the near future, possibly within the next two months.

“This is a completely new model with its own strengths, and perhaps some weaknesses,” Holz wrote on Discord. He noted that V7 will likely require a new style of writing and encouraged people to try out the version to help provide feedback.

Techcrunch reporter Kyle Wiggers noted that the V7 performed well on his basic commands. “Midjourney’s model isn’t specifically optimized for the Ghibli style, at least not officially, but it can still produce aesthetically pleasing work,” he added.

According to AI experts in Vietnam, ChatGPT's image generation feature may be easier to use and more convenient than early image generators such as MidJourney and Stable Diffusion. However, OpenAI's model has not yet created a breakthrough in image quality, which may be worse than previous tools.

Midjourney is a rather unusual company. Founded in 2022 by Holz, who co-founded PC peripheral maker Leap Motion, Midjourney has never received any outside investment.

The company also doesn’t disclose its revenue, but CBInsights estimates Midjourney will have around $200 million in revenue by the end of 2023. The company recently said it’s building a hardware team to work on some undisclosed projects, and it’s continuing to train previously announced models to create videos and 3D objects.

Like ChatGPT, MidJourney is facing several lawsuits, alleging it violated the rights of millions of artists by training AI tools on images scraped from the web without the consent of those images' creators.

Source: https://znews.vn/mo-hinh-tao-anh-ai-tot-hon-chatgpt-post1543472.html