In Episode 10 (August 29, 2025), Jordan showed a chatbot-style image editor he built in a day, right after a wave of new editing models launched. You upload a photo, describe the change in plain words, and each result becomes the starting point for the next edit. On air we gave him a cowboy hat, a western shirt and a cartoon makeover, and the product-placement test half worked.
Jordan's take at 01:57: everyone is trying to replace Photoshop with AI, and the blocker has been continuity, meaning the edited image still looks like the same person.
So if I ask something like, you know, put a cowboy hat on Sam, I get a cowboy hat on Sam, but I still get Sam.
— Jordan Metzner, Episode 10
New models, including Google's Nano Banana, made that much easier. Jordan saw a use for small businesses and e-commerce brands that constantly need product shots and headshots. Sam added the internal reason: we like to give our teams access to new tools fast, even if they aren't perfect.
Nano Banana came up as the reason for the hype, but Jordan said the model in his app might behave differently from the one in Gemini.
The hat and ranch worked, and Sam noticed the lighting: the hat cast a natural-looking shadow across Jordan's shoulder. The shirt edits and the cartoon version kept him recognizable. Holding the beer worked. "Have me drinking the beer" didn't fully land; Jordan's first reaction at 07:17 was "Not exactly." Sam called the demo "a bit of a flop," then pointed out it was built from one day to the next.
Now, you have to understand these are this is like an image of an image of an image. So, you know, it continues to make changes, and we'll probably see, like, some kind of continuity degradation over time.
— Jordan Metzner, Episode 10
Our later image work used Nano Banana for product photos for a 3D-printed charm store. For how the models stack up, see ChatGPT vs Nano Banana.
Jordan put a chat interface over an image-editing model (FLUX.1 Kontext Pro), let the user upload a photo and type edits in plain language, and fed each output back in as the root image for the next edit. He built it in a day.
That property is what Jordan calls continuity, and it's what the newer editing models made possible. Our tool chained edits by using each output as the next input, which kept Jordan recognizable through a hat, a shirt change and a cartoon version.
Partly. Putting a green glass beer bottle in Jordan's hand worked, but asking for him to actually drink it did not fully come out right on air.
Jordan expected it to. Each result is 'an image of an image of an image,' so he warned that continuity would probably degrade over many edits.
