news

---

  New Intelligence Report

Recently, these "TED speakers" have become extremely popular on the Internet, and can be said to be spreading like a virus.

Take a closer look, can you spot any problems?

The answer is revealed - none of these five people is a real person!

The guy who is looking for someone online is about to cry

"None of them are real? It's so creepy!"

Netizens commented: This has surpassed the uncanny valley and reached the "valley of hyperreality".

In just over ten hours, the post sharing this picture had been viewed by over 5 million people on Twitter.

Later, the author was also exposed - he was Leo Kadieff, a former member of the Stable Diffusion team.

He revealed: These TEDx speakers were all made using the latest Flux real version of LoRA.

Previous AI raw pictures would appear inconsistent to the human eye to some extent, but this time the pictures are so realistic, thanks to the improvement of the model using LoRA technology, which greatly increases the sense of reality.

Moreover, the author said that this workflow has another benefit, which is that it greatly simplifies complex prompt words.

This news made those who were not good at writing prompt words ecstatic.

This small 22MB file can save us the trouble of not having to write a bunch of authenticity-related tokens in each prompt word.

Just one sentence, "a RAW hyper-realistic photo, UHD, 8k", is enough. Realism lovers will absolutely love this tool.

The author bluntly asks: Do we still need to fine-tune our realistic models?

- These images are the raw output of Flux+LoRA and have not been enlarged or post-processed in any way

ComfuUI：https://we.tl/t-zrC5tPFG17

Real version of LoRA, outstanding effect

It is not difficult to see from the two pictures below that the comparison between using LoRA and not using LoRA is indeed very obvious.

At the same time, Kyrannio, a "TED speaker", also tried to replicate the trend using Midjourney.

The initial prompt words are as follows:

A woman speaking on stage, from Google, white background, company logo blurred, tech conference -- style raw -- v 6.1

It can be seen that the generated effect is not bad, but there is still a big gap with the pictures generated by Leo Kadieff.

Then, the blogger made some improvements:

A young woman smiling and speaking on stage, from Google, white background, company logo blurred, tech conference - style raw -- v 6.1

After multiple generation attempts, the closest result was found:

At the same time, as Google Imagen 3 became publicly available, netizens also tried it out with this prompt as soon as possible.

For a time, the entire Internet was swept by the craze of AI-generated images.

Yes, as just mentioned, Google’s most powerful text image model Imagen 3 is now officially available.

prompt：Photo of a man holding a sign that says: "Imagen Is Now Almost As Good As Midjourney" in New York City.

Source: Risphere

Netizen chrypnotoad said that he had never seen an AI that could make the Achilles Shield so well!

Imagen 3 can handle such a complex prompt with ease, so it should not be underestimated.

The well-known blogger "Guizang" said after the experience:

The generated content is accurate but the image is not very aesthetically pleasing. As long as it involves people, you have to carefully consider the writing of the prompt words, otherwise there is a high probability that the image will not be produced.

Fortunately, they did a good job in interacting with the prompt words:

LLM will analyze the prompt word type and give you related words that you can switch to directly.

Source: Guizang