Welcome back to the AI For Humans Newsletter!

AI Image generation is something we’ve been tracking here for a VERY long time — some of our first episodes of the pod were about how exciting early Midjourney models were. But as of 2025, we’ve gotten pretty close to parity on photorealism and control in initial AI image generation - it’s mostly been solved.

However, a nagging problem in the space has been photo manipulation. What if you want to remove something easily from the photo? Or swap out something different? Or change the color of the background but keep the main subject the same?

Enter the new multi-modal <takes deep breath> Gemini 2.0 Flash (Image Generation) Experimental.

Google’s new model is a massive step-up (good deep dive here) in being able to ask for changes to existing photos, better comprehend what you’re asking for, and unlock all sorts of photo manipulation that wasn’t possible before.

In context, "multimodal" refers to the capability of a model to process and integrate multiple types of data inputs, such as text, images, audio, and video. This allows the model to understand and generate content that combines these different data forms, leading to more comprehensive and versatile outputs.

The use cases we’ve seen are broad and fascinating: Change clothes on your subject with ease, change of subject but keep the same style, and maybe our favorite one… create start and end frames of a kids movie keeping a consistent character throughout.

Is it perfect? Far from it. Kevin had a pretty terrifying experience attempting to make a new headshot and many users have pointed out that it’s VERY good at removing watermarks from images.

But this is a new world of democratizing the tools around image manipulation and giving users a LOT more control over how AI image gen works. Will it kill photoshop? Eh, prob not but it does give AI imaging tools a lot more power.

More on Thursday!

-Gavin (and Kevin)

We 💛 This - Midjourney Image Retexture

Last week we talked through Runway’s new method to restyle any video clip. In short you simply provide it with the first frame of the video, remade in a different style, with the same structure.

But how do you restyle a single frame from a video? A few ways to do this:

• Magnific - Powerful, but heckin spendy

• Adobe Firefly structure reference - Kinda works but the output images aren’t that impressive

• Gemini Flash image generator - The new kid on the block, works pretty well but if you’re going for specific aesthetics it may struggle

• 👑 Midjourney image retexture - Probably the best tool for this on the market at the moment, if you want very high quality output images

It’s part of their “Full Editor”, available to anyone either (1) on an annual plan, (2) who has been sub’d for a year or more, or (3) has generated more than 10k Midjourney jobs.

In Midjourney simply upload your original image, and prompt to your heart’s delight to “retexture” the image. Then plug it into Runway’s first frame feature to restyle the video clip:

1. The original

2. Two intricately detailed steampunk robots

3. Two babies, shot on a smartphone camera

4. Two bright green cartoon anthropomorphic slime people

5. Two victorian characters from an old fashioned painting