18 March 2025

Google Just Changed Image Editing FOREVER [Gemini 2.0 Experimental Demo]

1Frameworks

6Insights

Frameworks in this episode

Productivity6 steps

Single-Step Gemini Image Editing Loop

Make one precise visual change at a time, inspect it, and iterate.

Read →

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Explainer· 2

Explainer00:00

Gemini Brings Text-Based Image Editing to Free Users

Google's experimental Gemini release combines text and image capabilities so users can modify images through natural-language requests. The host positions this as a major accessibility shift because basic edits previously required manual work in tools such as Photoshop or Canva.

Gemini can edit existing images from text instructions.
The demonstrated capability is available without paid design software.
Examples include replacing objects and removing landmarks.
The model combines text and image processing.

“You can now edit images with just text.”
Host · 01:00

“And now I can do it with just typing in a request, and I can do it for free.”
Host · 01:30

#gemini#image-editing#multimodal-ai#design

Explainer10:30

Use One Image's Style to Generate a Different Subject

The model can use an uploaded image as stylistic inspiration while generating a new subject. This expands the use case beyond editing existing pixels and gives marketers a way to create visually related assets without manually reproducing a reference style.

Users can upload a visual reference.
The reference style can guide a newly generated image.
The new image does not need to retain the original subject.
The capability can help maintain visual consistency across campaign assets.

“Use the same style, but generate an image of a dog”
Host · 11:00

“you can upload something and use that as an inspiration to have aI generate a net new image.”
Host · 11:00

#style-transfer#image-generation#brand-assets#creative-ai

Story· 1

Story02:00

A Casual Photo Becomes a Passport-Style Portrait

A user asked Gemini to isolate a blonde woman, give her a neutral expression and white background, and make the result square. The example illustrates how AI can quickly reformat an almost-suitable photo for a rigid visual specification, although official acceptance requirements would still need independent verification.

The source was an ordinary photograph rather than a studio portrait.
The request specified the person, expression, background, and shape.
Gemini produced a square, passport-style result.
The example shows the value of editing an image that is already close.

“take the blonde haired woman in this picture on the right with a neutral face expression on a white background, and make it square”
Host · 02:00

“You can now take an image that might be close and do some basic editing with the new Google Gemini, 2.0 experimental”
Host · 02:30

#passport-photo#portrait#image-formatting#gemini

Tool· 1

Tool09:30

Generate an Illustrated Story Scene by Scene

Gemini's multimodal output can produce both a story and an image for each scene from one narrative premise. The host highlights personalized children's books as an immediate application and suggests that narration could turn the output into an automated media product.

A single prompt can define a character, adventure, setting, and visual style.
The model can generate written scenes and matching images.
Personalized children's stories are a natural consumer use case.
Narration and a simple interface could turn the capability into a subscription product.

“For each scene, generate an image.”
Host · 09:30

“You can make bespoke children's book for your children every night.”
Host · 10:00

#visual-storytelling#childrens-books#multimodal-ai#product-idea

Takeaway· 2

Takeaway04:30

Product Photos Can Become New Merchandising Scenes in Seconds

The croissant demonstration shows that Gemini can preserve a product composition while changing toppings or adding human interaction. For businesses, this suggests a faster way to test product-image variations and lifestyle compositions before investing in a new shoot or manual composite.

The model changed chocolate drizzle to strawberry icing.
A hand was added grabbing a croissant.
The original product composition remained recognizable.
Marketers can iterate through product-image concepts rapidly.

“Notice it did the same croissants, because I said, replace.”
Host · 05:00

“You could take your product and juxtapose it in different ways, really, really quickly and iterate through those images in a really awesome way.”
Host · 05:30

#product-images#merchandising#ecommerce#marketing

Takeaway11:30

Prototype a Presentation's Visual Flow Before Polishing It

The host recommends feeding a presentation script into the multimodal model to generate an initial sequence of slide images and story beats. The purpose is not necessarily to produce the final deck, but to test narrative flow and visual direction quickly before investing in detailed production.

Start with a rough script or presentation concept.
Generate a visual first version of the slide sequence.
Evaluate the story flow before polishing individual slides.
Use AI output as a prototype rather than assuming it is final.

“I would do a v1 like to try to get the flow right.”
Host · 11:30

“I'd have it generate the slot, the the like slide, image and story, and go through it in a really visual story way.”
Host · 11:30

#presentations#prototyping#visual-storytelling#workflow