AI for Image Generation Webinar April 23rd
A live side-by-side of Google's Nano Banana Pro/2 and ChatGPT's GPT Image Gen 2, showing how image models that reason can produce accurate infographics, before/after repair plans, DIY guides and furnished-room mockups for real businesses.
Pour qui: Business owners and operators (especially in renovation, landscaping, real estate, education or marketing) who want to use AI images for more than decoration.
À retenir
Image models now reason, not just draw
I ignored image generation until Nano Banana Pro and Nano Banana 2, because earlier models mostly made pretty pictures. These models do visual reasoning: asked for a step-by-step infographic of gold extraction, they understand the process and use the correct industry terms. I even fed Nano Banana Pro first-year master's physics exam problems and it wrote correct worked answers on the sheet itself.
Change the audience, keep the meaning
Once you have an infographic, you can ask the model to rewrite its wording for a specific audience. I turned the gold refining infographic (electrowinning, carbon absorption, electrolytic refining) into a version for kids aged five or under: very different words, same meaning. A friend built a business on this idea, generating consistent characters and kid-appropriate stories from photos of a child's plushie.
Use the Thinking cards in ChatGPT
In ChatGPT, clicking Create an Image shows idea cards. Cards with the Thinking label make the model reason for a long time and fetch information from the web before drawing; the others are a quick 'make a cool image' mode. In my view, the real power of GPT Image Gen 2 comes from thinking plus image generation, which is how my detailed, web-sourced infographic was made from a simple prompt.
Before/after plans with budget tiers
Give ChatGPT a photo (real or generated) of a cracked wall and ask for three ways to fix it at three budget levels, with before and after results in one image. If your company has a pricing guideline, for example a price per square meter, add that document and photos of the area so the output follows your own rates. I see this landing first with landscaping, housing and renovation businesses.
One prompt, one visual DIY guide
Ask the model to include the step-by-step how-to inside the image and it produces a visual tutorial: clean the area, apply spackle, hand smooth, prime, paint, or for bigger cracks, mesh tape, first and second coats and a bonding agent. I point out that this gives you a new level of lead magnet, since step-by-step guides that used to be manual now take a single prompt.
Show the same space for different buyers
Generate (or photograph) an empty 20 square meter apartment, then ask for three interior designs for three personas side by side: an 18-year-old student, a 25-year-old couple, a 60-year-old woman wanting calm. Room details like the radiator and AC stayed in place. You can set the resolution manually instead of Auto (for example widescreen). Asking for a new camera angle did not hold the furniture in my test.
Heavy use: Higgsfield or an API
If you generate a lot of images and videos, I recommend one dedicated tool, Higgsfield, which gives access to Nano Banana, Kling and GPT image models with more settings and no Gemini watermark. For marketing at scale, the chat window is inefficient: I would give an OpenAI API key to Claude Code or Codex to produce ten or twenty visual assets at once.
Outils présentés
- ChatGPT (GPT Image Gen 2)Main demo: web-researched infographics with Thinking mode, cracked-wall before/after with budget tiers, DIY repair guides, furnished-apartment personas and a hair transplant timeline, all on the $20 Plus plan.
- Google Gemini (Nano Banana Pro / Nano Banana 2)Visual reasoning demos: a comic cat image, the gold extraction infographic, rewording it for young kids, solving physics exam problems and history timelines.
- HiggsfieldRecommended single subscription for heavy image and video generation, with access to Nano Banana, Kling and GPT image models, extra settings and no Gemini watermark.
- Wispr FlowSpeech-to-text I use to dictate prompts, including snippets that expand saved text such as links or an email address.
- Claude DesignShown as a design-side aside: setting up a design system, generating a landing page mockup and a code-based lyric video, with export to Claude Code.
- CodexRecommended as ChatGPT's equivalent of Claude Code, included in the $20 plan, and as a way to connect an image API and generate many assets at once.
Questions du public
Which speech-to-text tool are you using to dictate prompts?
I use Wispr Flow (my referral link was shared in the chat). I find it useful mainly because of snippets: when I say 'my permanent webinar links', the saved links appear in the text. In the same way, saying 'my email' inserts your email, and saying the name of a referral code inserts the code itself, which is how I dropped my link into the chat during the session.
Is it really possible to generate this many images on a regular paid subscription?
I was on ChatGPT's $20 Plus plan and had generated about 20 images during the session, which surprised me because Gemini used to limit image generation much more. Later I thought I had hit a limit when the generate option disappeared, but it turned out I simply had not selected Create an Image. My conclusion: there is a limit, but I do not know what it is.
Do you think Anthropic will build image or design generation to keep up with OpenAI?
I do not think so, and I do not see it as a good strategic move for Anthropic, though I could be wrong. Image generation is a very different kind of model from text generation, and Claude Code can already connect to the best image models via API. Anthropic built a model that is very good at using tools; other players build the tools. For visual assets at scale, I would give an OpenAI API key to Claude Code or Codex.
If I only have $20 a month for AI, should I pick ChatGPT or Claude?
With $20 to spend, I think you are better off with ChatGPT: you get the new image generation plus Codex, which I call the Claude Code equivalent. Claude Code starts at $20, but in my view you get little out of it below about $100 a month, while Codex at $20 takes you quite far. At around $100 the two become quite similar, and learning one well means you can use the other.
À regarder ensuite
7:52Get Your Money's Worth from ChatGPT Images 2.0 (2026)Make mockups, infographics and before-and-after visuals yourself with GPT Image Gen 2.
8:50Nano Banana Pro Changed Everything About AI Images (2026)Use Nano Banana Pro in Gemini to solve a problem from a photo and diagram it.
10:05Claude + Higgsfield: AI Marketing Assets in MinutesGenerate images at volume by connecting Higgsfield to Claude as a connector.
