The Last Invention is AI

Episode: OpenAI Gen: Floral Fantasia
Date: December 24, 2025
Host: The Last Invention is AI

Episode Overview

This episode explores OpenAI’s latest release: a new image generation model referred to as “image 1.5.” The host evaluates its technical improvements, user experience, and its place in the rapidly evolving competitive landscape of AI image generation. With direct hands-on testing, the discussion covers both the model’s capabilities and its current limitations, while also considering broader impacts for OpenAI’s market strategy as it faces strong competition from Google’s Gemini/Nano Banana models.

Key Discussion Points and Insights

1. OpenAI’s Competitive Urgency and Context

• OpenAI released image 1.5 earlier than planned, reportedly accelerating its timeline in reaction to losing ground to competitors, particularly Google’s Nano Banana image model.
• The host remarks:

  “I do think this is a really impressive model. … perhaps it is because prior to them releasing this model, their last image model update I was begging them to make for over a year. The old version of DALL-E… was absolute garbage. They’re getting smoked by literally everybody, including Midjourney and everyone.” (00:40)

• Code Red: Internal urgency within OpenAI, described as “Code Red,” fueled rapid development and release to avoid further market loss.
• Leaderboard pressure:

  “The newest version of Google's rival image generator, Nano Banana, topped the LM arena leaderboard across a bunch of different benchmarks. And I do not think OpenAI appreciated that.” (04:36)

2. Technical and Functional Improvements in Image 1.5

• Instruction Following and Speed:

  “Apparently it’s a lot better at following instructions. I have found that it is more precise at editing and it’s four times faster at generating images, which, let’s be honest, is the biggest thing that would drive me crazy with OpenAI.” (02:13)

• Granularity and Iteration:

  “They have a really cool feature now where if you click on an AI image, you have this feature called select area and you can select a part of the image and have it regenerate that bit of the image only…” (09:01)

• Editing Experience: The model now supports partial regeneration, reducing the frustration of having to re-render entire images for minor tweaks, though the host notes some integration issues with very granular edits (e.g., only updating a head can cause mismatched backgrounds).
• Input Flexibility: By uploading reference images (e.g., Sam Altman’s head, OpenAI logo), users can achieve high-fidelity outputs that previous iterations struggled to create.
• 4K Output: Model is capable of generating 4K images.
• Creativity and Interface Upgrades:
  • New UI features on ChatGPT’s Images tab simplify the workflow for generating and managing images.
  • Trending prompts, preset filters, and ideas for creative templates (holiday cards, album covers, etc.).

3. Hands-On Testing & Use Case: Creating a Complex YouTube Thumbnail

• The host outlines a real example:
  • Prompt: “Generate a YouTube thumbnail of me looking shocked and staring at a giant cloud with letters in the sky written by an airplane that say ‘new AI image.’ The airplane has an OpenAI logo and is being flown by Sam Altman.” (07:02)
  • Findings: The initial image impresses, especially compared to older models, but with minor errors (logo inaccuracy and less-convincing Sam Altman likeness).
• Notable workflow:
  • Used new “select area” tool to regenerate the airplane pilot’s head but encountered blending issues with the background.
  • Solution: Upload explicit reference images (desired logo, actual Sam Altman photo), which yielded much more accurate outputs after re-generation.
  • Quote:

    “Once I did that, it got the correct OpenAI logo and Sam Altman's head and actually everything looked great. … The image looks a hundred times better than its last model. So I’m really, really impressed.” (11:56)

4. Comparative Market Analysis & Implications

• OpenAI is aiming to close the gap—or even overtake—Google’s Nano Banana, especially since the latter leads on major performance benchmarks.
• There’s a dynamic, ongoing “arms race” in AI image models, with companies pushing to outdo each other on both speed and quality.
• The host speculates that improvements in image models will soon be incorporated into video generators (like Sora), as the two are technologically connected.

5. User Experience Enhancements

• ChatGPT’s new “Images” tab streamlines creation and management:
  • Saves historical creations
  • Allows immediate access to trending prompts and templates
  • Intuitive toggling between image and text tools
• Quote:

  “You can discover like holiday cards or… what would I look like if I was a K pop star? … I think they're trying to like create some trends or something. But I do think it's nice—it saves you a couple seconds...” (15:53)

Notable Quotes & Memorable Moments

• On OpenAI’s need to catch up:

  “Every week, every month that they're behind in the benchmarks, a bad sign for them, they lose market share, so they're trying to be faster.” (05:09)

• On improvements in user control:

  “This update that they've added, you can tell it to make small updates like that and it will make the small update across the entire image. So… more like a creative studio.” (13:34)

• On partially updating images:

  “When it regenerated his head, it put like a better looking head on, but all of the space around his head didn’t match the sky beside it. … It looked like I was in Photoshop and I like cut and pasted a little piece of an image on top, so it kind of looked bad.” (10:17)

Timestamps for Key Segments

• 00:00–02:00 — Introduction, initial impressions of the new image model
• 02:01–05:40 — Urgency, market competition, and background on OpenAI’s “code red” state
• 05:41–07:20 — Technical performance, speed and benchmarks
• 07:21–11:00 — Hands-on testing: YouTube thumbnail use case, pros and cons in practice
• 11:01–13:30 — Manual solutions and improvements with reference image uploads
• 13:31–16:30 — Interface upgrades, iteration improvements, creative potential
• 16:31–end — Conclusion, summary thoughts, calls to action

Episode Takeaways

• OpenAI’s image 1.5 is a significant leap in capability, especially in speed, instruction following, and iteration/editing details.
• Despite some minor flaws in very granular edits, the model can achieve highly accurate and creative outputs with well-crafted prompts and reference uploads.
• The competitive environment is driving rapid, user-focused innovation, with OpenAI determined not to fall behind industry rivals.
• End users benefit from improved UX, faster workflows, and a more robust image generation toolkit built into ChatGPT.