ChatGPT ahora crea IMÁGENES PERFECTAS 🤯 Nuevo GPT Image 2

The video opens with a striking demonstration of OpenAI's GPT Image 2.0 model generating a completely fake but convincing YouTube channel screenshot — including readable interface text, thumbnails, tabs, and descriptions — all from a simple text prompt. The presenter uses this to argue that we have crossed a threshold where AI-generated images are indistinguishable from reality for the average viewer.

The presenter then walks through a structured head-to-head comparison between GPT Image 2.0 and Google's Gemini image model (nicknamed 'Nano Banana'). The first test involves generating personalized images of the presenter himself using two reference photos. GPT Image 2.0 produces more accurate likeness without any specialized model training, outperforming Gemini. The second test involves recreating interface screenshots, such as a macOS desktop with a browser open. GPT Image 2.0 again wins, producing results that look nearly indistinguishable from real screenshots, while Gemini's version has subtle inconsistencies in the Mac aesthetic.

However, the comparison is not one-sided. When asked to convert a 2D floor plan into a 3D rendered view, Gemini outperforms GPT Image 2.0 by more faithfully preserving the spatial layout and individual elements of the plan. GPT Image 2.0 reinterpreted and moved elements, introducing inaccuracies. Similarly, in a room organization task — where both models were asked to analyze a cluttered room and generate a tidied version — Gemini produced a more coherent and complete result.

For infographic generation, GPT Image 2.0 generally wins on visual design quality and text rendering, producing a Stonehenge infographic that the presenter says could appear in National Geographic. However, in a more nuanced prompt asking for a 'children's handmade model of the water cycle,' Gemini better followed the spirit of the instruction by generating something that looked authentically homemade, while GPT Image 2.0 produced a more polished but less contextually accurate result.

In a creative comic strip test — asking for a four-panel, 80s European comic style strip about AI's impact on the audiovisual industry — both models perform impressively. GPT Image 2.0's version features dense, biting sarcastic dialogue woven across four panels, while Gemini's version has a more conceptually coherent narrative arc. The presenter calls this a tie.

The video also highlights that to access GPT Image 2.0 at maximum quality, users should use platforms like Freepik rather than ChatGPT directly, as ChatGPT limits the output quality. Freepik, which sponsors the video, offers access to 42 models including GPT Image 2.0 and several video generation models.

The presenter concludes that GPT Image 2.0 is a landmark achievement — a publicly available model capable of generating images indistinguishable from reality without complex fine-tuning. While Gemini holds its own in spatial reasoning and instruction fidelity tasks, GPT Image 2.0 leads in most visual quality benchmarks. The presenter also flags this as both a major creative opportunity and a significant threat to information integrity.