The new ChatGPT Images is a stunning improvement, and enormously fun
SMRTR summary
OpenAI just dropped a major upgrade to ChatGPT's image generation capabilities, dubbed "ChatGPT Images," and it's surprisingly good at something AI has long struggled with: handling text accurately. The new tool can now edit and recontextualize images with remarkable precision, transforming a simple photo into complex scenarios while maintaining realistic shadows and proportions.
A tech journalist put the system through its paces, starting with a basic shirt change request and ending up creating an elaborate Star Trek-themed holiday party invitation complete with accurate text that read "Lights! Eggnog! Punching!" The AI successfully composited the writer into various scenarios at Vasquez Rocks, added a classic Gorn monster, dressed characters in Star Trek uniforms, and even created a snow-dusted Christmas scene.
While the tool still makes some unasked-for changes, like altering facial expressions or camera angles, it avoids the unsettling "uncanny valley" effect that plagued earlier AI image generators. The upgrade represents a significant leap forward for ChatGPT, which had fallen behind competitors like Google's image tools, and is now available across all ChatGPT tiers, including the free version.
SMRTR provides this summary for quick context. The original article belongs to ZDNet.
Read the original article