Lore

AI image generation

40+ Things ChatGPT Images Can Actually Do

ChatGPT Images 2.0 is a substantial upgrade over prior image models because it goes beyond pretty pictures to produce genuinely usable, information-accurate visual output — thumbnails, carousels, mood boards, infographics, mockups, and planners — and, per the video's own tests, it now beats Nano Banana specifically at embedding accurate text and information into images, even though Nano Banana still edges it out on overall photorealism.

Matt Wolfe · 2026-04-28 · English

Key ideas

  1. Generates YouTube thumbnail concept boards in a grid (e.g. 2x3), initially as squares, correctable to 16:9 with a follow-up prompt.

  2. Reads URLs and analyzes website content to pull real logos/branding into generated images (e.g. Facebook ads, contextual visuals).

  3. Builds multi-image carousels for Instagram and LinkedIn, tailored differently per platform via prompt specificity.

  4. Plans and renders a full 30-day social media content calendar as a single visual.

  5. Produces brand mood boards including exact hex color codes for a full visual identity system.

  6. Generates logo exploration sheets with multiple (8+) labeled concepts plus written rationale for each.

  7. Creates product launch one-sheets combining app mockups, copy, and branding.

  8. Makes promotional flyers for local businesses, restaurant menus with plausible food/pricing, and event posters/schedules.

  9. Pulls actual listing photos from Zillow and incorporates them into a real-estate flyer mockup.

  10. Generates multi-day travel itineraries with neighborhood/food recommendations, but is unreliable at accurately placing real geographic landmarks.

  11. Produces packing checklists, weekly chore charts, 30-day habit trackers, and a five-phase garage organization project planner.

  12. Generates infographics directly from a supplied website URL, judged more accurate (if less visually polished) than Nano Banana for this purpose.

  13. Defaults to a blue/white visual aesthetic unless the prompt explicitly steers color otherwise.

  14. Creates timelines, mind maps, and process/workflow diagrams (e.g. an AI customer-inquiry flow).

  15. Builds a comparison chart of AI tools (ChatGPT vs Claude vs Gemini vs Perplexity) and self-flagged its own accuracy caveat within the chart.

  16. Produces an AI image-prompt cheat sheet (formula, ingredients, common mistakes, example prompts, revision tips, pre-generation checklist).

  17. Generates multi-slide decks and one-page sales pitches from a single prompt.

  18. Creates product packaging concepts, six-image e-commerce product sets, and merchandise mockups (shirt, hoodie, sticker, mug, hat) with one phrase across items.

  19. Produces app store screenshot mockups and website hero-section mockups.

  20. Generates course curriculum graphics, client onboarding checklists, and converts messy notes into a polished visual action plan.

  21. Makes printable recipe cards and multi-panel one-page comics.

  22. Content-policy rejections appear non-deterministic: an identical prompt was rejected once, then succeeded on retry without any changes.

Insights

The model's strength has flipped relative to Nano Banana: Nano Banana still wins on photorealism, but ChatGPT Images 2.0 is now described as the leader specifically at getting accurate information and text into an image.

The model doesn't just synthesize decoration — it incorporates real external images (Zillow listing photos, a company's actual logo scraped from its website) directly into generated mockups.

Content-policy rejections can be inconsistent: resubmitting the exact same brand mood board prompt with no edits succeeded after an initial rejection, implying the filtering isn't fully deterministic.

Left unsteered, the model has a strong stylistic prior toward a blue/white color scheme, visible enough across outputs that it became a recurring audience joke in the comments.

Accuracy is domain-dependent, not a single global capability: it can pull a real Zillow photo into a flyer correctly, yet still misplace real-world landmarks in a travel graphic — pulling in real data is more reliable than depicting real geography.

The creator started with a fixed scope ("33 things") but abandoned the count partway through testing because so many more usable prompts surfaced, ultimately not tracking the final number.

«pretty much all I can say about this model is it is just actually way more useful.»

— 00:10

«You can make slide decks. You can make full carousels for places like Instagram. You can get it to explain stuff to you in images and have it be really accurate.»

— 00:14

«I promise you that if you're watching this and you think you know everything this model can do, I'm probably going to show you some stuff that you had no idea that this model was capable of.»

— 00:31

«It actually pulled in the Future Tools logo directly from the website, and it knows exactly what the site's about, and even has like a semi-accurate screenshot to the website.»

— 04:56

«Getting AI to create logos for you used to be such a nightmare, and now it's just so simple.»

— 11:24

«Like, this was one of the things that probably blew me away the most when I tested it. I didn't know if it was going to pull in actual images from the flyer and then incorporate them into this image, but it did.»

— 14:46

«These are not quite very accurately located landmarks.»

— 16:41

«the ChatGPT images are way more accurate. Like they're just getting a lot of the details more right than what Nano Banana was»

— 20:12

«You can actually give it a website and tell it to make an infographic based on what's on that website»

— 20:54

«Chatbots talk, agents do.»

— 25:41

«realistic images but this one is definitely the new leader in actually getting information and text into the image»

— 32:10

Reception

Predominantly positive reception with viewers appreciating the creative demonstrations and practical AI applications, though some criticism about video length and lack of novel techniques moderates the overall enthusiasm.

The video works as a breadth-first demo reel of concrete, copy-pasteable prompts rather than a technique tutorial — its real payload is the sheer catalog of business, planning, and creative use cases, with the standout finding being ChatGPT's newly claimed edge over Nano Banana at embedding accurate text/information into images.

33:32

↳ Matt Wolfe · YouTube

Watch original