😺Gemini's image editing is UNREAL

PLUS: A 3 step hack to turn AI into an expert...

Welcome, humans.

Today, in “Weirdest use of ChatGPT we’ve seen (this week)”, someone is apparently using ChatGPT to communicate with crows.

The line “the crows KNOW who I am, and are interested” really HIT with Redditors, inspiring short stories and dubbing the guy the new “Crow King.”

Fun fact: I've spent hours researching how to befriend squirrels in my front yard, but never thought to ask an AI for help. Guess ChatGPT can turn me into Snow White after all! GPT, GPT, give me a ring—who is officially the new Squirrel King?

Here’s what you need to know about AI today:

Meta tested AI-generated Instagram comments.
GitLab faced a third lawsuit over AI customer adoption claims.
OpenAI may cut ChatGPT subscription prices in India.
1X Technologies will test humanoid robots in homes by late 2025.

Advertise in The Neuron here.

Gemini 2.0’s image editing tool is absolutely WILD—when it works, AND when it doesn’t…

In case you missed it, Google recently released Gemini 2.0 Flash with image generation that lets you create and edit images through a chat interface (via Google AI Studio). It's one of those "Holy you know what" AI moments—at least, when it works.

Here’s one awesome example of how someone used it to create multiple angles of a product from a single base image. And then there’s more, ahem, creative uses…

That said, it doesn’t always respond as intended. We honestly couldn't stop laughing at some of the image editing fails that happened as we tested it out.

Here’s what happened:

First, we tried to get Gemini 2.0 to move Disney’s BDX robot to turn and face NVIDIA CEO Jensen Huang (featured below). Let’s just say things got a little weird…

Then, we gave it what we thought was some pretty specific and helpful feedback, calling back to an original reference photo we shared. And uh… it got weirder.

Call it a robot Folie à deux…

In all fairness to Google, this is an “experimental research preview” (as are most models that you can try in Google AI Studio), but it’s quite fun to mess around with.

Speaking of fun: Since that initial release, Google Creative Lab also demo’d a new co-drawing tool. This lets you doodle something and have Gemini 2.0 draw along with you.

Here are some examples:

Step-by-step creations: Draw a basic shape and watch Gemini develop it.
Battle sequences: Your character fights an AI-created opponent.
Colorization magic: Black and white sketches become vibrant artwork.
Create new characters: Basic sketches transform into polished cartoons.

P.S: if you’re technical, you can also clone the code to play with your own version.

Here’s why this matters: Unlike most previous image generation models, Gemini 2.0 Flash can edit existing images rather than regenerate an entire image from scratch.

This incremental editing is something the AI industry has been waiting for, allowing you to make very specific modifications while preserving the rest of the image (more like Photoshop than Midjourney).

As Reddit reminded us, this model can do a lot of things that weren’t possible before—and it only gets better from here. Which y’know, is kinda scary given how good it can be.

Here’s how to try it yourself: Click here, add an image with the plus button, and then prompt it with very specific instructions for how it should adjust said image. This video helps explain how to do it.

After failing to get what we wanted in our tests, we looked up some actual prompt tips for these kind of image editing. Here’s what we found:

Be ultra-specific—detail exactly what to change, where, and how.
Use reference points—“make the left eye bigger” instead of just “enlarge the eye.”
Iterate gradually—make small changes in multiple steps rather than one complex request.
Keep it simple—avoid overly technical terms; basic language gets better results.
Try different angles—if stuck, rephrase your request.

Here are six prompts we found for working with Gemini 2.0 (that actually work)—and here’s the Perplexity chat we used to find it (with some extra tips) and the original source. Oh, and here’s an ~11 min video with some additional demos and tips as well!

FROM OUR PARTNERS

100+ ChatGPT Prompts to Revolutionize Your Day

Supercharge your productivity with HubSpot's comprehensive guide. This free resource is your fast track to AI mastery:

Industry-Specific Use Cases: 15+ real-world applications across various sectors.
Productivity Guide: 21 best practices to 10x your efficiency with AI.
Prompt Powerhouse: 100+ ready-to-use prompts for immediate implementation.
Challenge Buster: Overcome common AI hurdles with expert strategies.

Plus, in-depth sections on email composition, content creation, customer support, and data analysis.

Find Your Guide Here

Prompt Tip of the Day

Want ChatGPT to become an instant expert? Try this 3-step hack from Reddit:

Ask for “20 words describing a [specific specialist].”
Request a 4-sentence prompt using those words to “summon this specialist.”
Paste that prompt into a new chat.

The result? Instead of generic bullet points, you'll get a conversational expert who guides you through complex topics with natural paragraphs and deeper insights.

Our favorite insight: It's like the difference between saying “know about taxes” versus summoning someone who'll “walk me through each step of my tax return.”

Treats To Try.

*The Dell + NVIDIA partnership is reshaping AI workflows across industries. Their new podcast shows exactly how it works. Listen here
LHM from Alibaba turns your photos into a detailed, animatable 3D avatar in one second, preserving the same details from the photo—demo, try it here.
Morphic is a creative canvas that turns your anime concepts into videos with character training, pose control, and seamless editing tools—here’s how it works.
A0.dev is a vibe-coding tool for mobile apps specifically—five free prompts per day.
AgentOps helps you trace, debug, and deploy reliable AI agents with visual replays, error tracking, and cost monitoring.
Midjourney now lets you organize your images with folders, describe web images to generate prompts, and create personalized moodboards that blend with style references (example).
Strella lets you conduct AI-moderated customer interviews, and turns them into shareable clips, searchable insights, and summaries to easily review.
Buildship lets you build backend tasks in minutes (e.g. processing payments, storing data in databases, calling external tools like OpenAI) without code.
Alta's virtual agents qualify your leads, make calls, and analyze revenue data 24/7, tripling your qualified meetings while your team focuses on closing deals.

See our top 51 AI Tools for Business here!

*This is sponsored content. Advertise in The Neuron here.

Around the Horn.

Meta has begun to test the ability to write AI generated comments on Instagram (example).
GitLab was hit with a third lawsuit in five months for allegedly misleading investors about customer adoption of their AI capabilities.
OpenAI and Meta want a deal with Reliance Industries (India’s top wireless provider + retailer), and OpenAI could slash the price of its subscription in India by 75-85%.
Pika is working on a new secret feature that lets you manipulate any character object in your video while keeping the rest in tact (6 examples).
1X Technologies will begin testing its Neo Gamma humanoid robots in hundreds (if not thousands) of homes by the end of 2025—using human teleoperators who can take over if the robots go off the rails, which creates some privacy issues.
Dair.AI and The AI Timeline recapped the top AI papers from last week—check them out!

FROM OUR PARTNERS

Mr. Wonderful Lost Out on $400 Million… Will You?

The Sharks all passed on Ring before its $1.2B buyout. Now, RYSE is making waves in smart home tech with $10M+ in revenue, 200% growth, and 127 Best Buy stores. Invest at $1.90/share before their next big move.

Find Out More