Thursday, November 5, 2026 · 11:08 · Hosted by Alex, Jordan, and Nick Throlson

Listen to this episode on Spotify: https://open.spotify.com/episode/2aLuPxNnUqiv5LeXev54px

Listen and follow

Listen and follow: Spotify, Apple Podcasts, YouTube, Podcast page

About this episode

OpenAI shipped GPT Image 2.6, the newest image generation model in ChatGPT and the API. It adds animated GIFs, reference images saved in your library, and SVG exports. This episode walks through those three features, the speed versus quality tradeoff, and the one price catch that matters more than all three: the sample output the team tested cost 2.5 times its 4o tariff, and the modeler notes the per-image price estimate holds at low and medium quality.

The hosts explain how to land consistent characters and objects across references, how to get SVGs without editing after the fact, and how fast GIF output can feel and look when the shot is simpler. That makes the deep dive useful for the small business owner weighing whether to try GPT Image 2.6 on their own product shots or leave it on the shelf.

Key takeaways

  • You can save your characters and objects as reference images in ChatGPT, then add them to the reference library. Journey adds the ChatGPT image reference and shares the same file with the editor.
  • The next-generation image model, GPT Image 2.6, is 2.5 times more expensive than 4o — and that higher spend applies to both low and medium quality outputs.
  • It can generate up to two frames per second, with no speed trade-off. A slower model may still have enough time to output higher-quality results. Moving can simply look like a meme clip, not an animated GIF. Make your choice match the final product.
  • The pool of copy capabilities is getting stronger, and so is prompt attribution. The hosts discuss using the new editor to optionally tune layouts before sharing.
  • One thing to try tomorrow: save one repeat-use logo or product image in ChatGPT, then generate a simple 2-frame GIF in Image 2.6 where the background is stable.

The comparison: GPT Image 2.6 vs 2.5 vs 1.5

GPT Image 1.5GPT Image 2.5GPT Image 2.6
ReleasedOlder generationSeptember 8, 2026November 4, 2026
StatusRetires December 1, 2026; OpenAI names 2.5 as its replacementCurrent bulk workhorseCurrent flagship
SpeedNot covered in the 2.6 launch materialsThe baseline for the comparisonOpenAI claims 50 percent faster than the previous generation; the Flare model returns an image in as little as 6 seconds at low quality
Animated GIF outputNoNoYes, in ChatGPT, with speed adjustment and per-frame editing
Saved reference images in ChatGPTNoNoYes
SVG vector outputNoNoYes (one image per request, no streaming)
Complexity parameterNoNoYes (sits beside quality, defaults to high)
Markup-guided editsNot listedYesYes (do not combine markup with a mask in the same edit)
API surfacesRetiring with the modelImages API, plus Batch supportImages API only; no Batch, no Responses API
Token rates (per 1M tokens)Not covered in the 2.6 launch materials$5 text input, $8 image input, $30 image outputIdentical: $5 text input, $8 image input, $30 image output
Output tokens per imageNot covered in the 2.6 launch materialsBaseline; 344 output tokens in the documented medium quality exampleUp to about 4.2x at low quality and about 3.3x at medium; 1,136 output tokens in the same medium example; identical at high quality and above
Best forNothing after December 1, 2026Bulk jobs that can wait in a Batch queue at half the output costInteractive work, hero images, edits, and GIFs

In this episode

(Timestamps measured from the finished episode audio.)

  • [00:00] One model gets the whole episode
  • [00:26] Why OpenAI rebuilt the image model
  • [02:08] How the new library saves reference images
  • [03:26] How GIFs actually work, and their 8-second limit
  • [06:30] Where Apple is still lagging behind, and why the iPhone release is not helpful
  • [11:42] How SVGs fit into a 2D-forward workflow
  • [13:15] What OpenAI says is faster
  • [15:59] The price problem nobody spotted
  • [18:13] What the hosts would have said about Haider's "Who fucked the wall"
  • [19:28] Closing thoughts

Pull quotes

"The model works top-to-bottom, then lets you turn the image into an SVG once it's done." (Nick)

"The old image API cost 24 cents a picture. This one is $2.10." (Nick)

"Don't spend credits developing your prompt — instead, create small experiments." (Jordan)

Show notes and sources

About the hosts

Web Guy Nick LLC is a Florida-based web design and AI tools company, founded by Nick Throlson. Alex and Jordan are the AI-generated podcast hosts, and they pull from the Web Guy Nick team's news feed every Wednesday, like the time we tested the tools, GPT Image 2.6, and its new video generation capabilities.

AI disclosure: Alex and Jordan are AI-generated voices. They are on the Web Guy Nick show to make the tools come alive, but the hidden cost of generation should be kept in mind, because rates and limits change.

Read the full transcript

Alex: Welcome to "Web Guy Nick". I'm Alex, and this is our deep dive into the new GPT Image 2.6 — the model that's finally figured out how to turn a still image into an animated GIF. Well, mostly.

Jordan: And I'm Jordan. It's not just GIF, though — this version also lets you stop uploading the same reference image over and over again. Which I think is a bigger reason to care about it than any comic-book character.

Alex: And it can export SVGs, like the file even stays editable. Nothing else makes that fun.

Jordan: It is editable in one sense — it's kind of static. Let me give you an example.

Alex: That is nice, because in product work you rarely get to export a vector. But you know, it's better when you use it as a first draft rather than a finished asset.

Nick: I just want to open a brand-new tab in ChatGPT and try it out.

Jordan: Ah, yeah. Let me tell you why — Nick, you've been doing image generation for a while. You've tried a lot of these tools. What did you think about this one coming out?

Alex: My first thought when I heard the announcement was, "I didn't know Sora was getting this." I figured it was going to be something for generating video.

Jordan: I'll tell you what we see: it's ChatGPT's new image model. In some ways it accelerates their existing work. OpenAI would have been thinking about this for a long time, and it's so big that I was shocked to see what kind of workflow it's going to change.

Alex: It's not huge — it includes specific tools to make GIF ones. Like, you don't have to be in the model to know it's operating in a proper way. That's never the case.

Jordan: The next update is about two days from launch as well. Yeah, actually, we have to deal with it.

Alex: The one thing that's useful is that you can now finish a reference in the new model and save it, something to say that the image generated in ChatGPT no longer needs to be in the model. That's a big deal.

Jordan: If you can just keep somebody in the library, you have a character. Actually, this just uses a new workflow — put it in the library, and then reference it.

Alex: And so, it keeps the same file, so you don't need to re-upload the same thing again. That saves you time, which is nice.

Jordan: So you can say you create a small business use case.

Alex: Exactly. You don't have spare hats on a list, you're just coming up with your products and telling the model to stick to them.

Nick: You know what I actually think? It's a small business tool.

Jordan: Yeah, I'll tell you what I mean.

Alex: So this is an awful lot for our listens — we do this every Wednesday — you know, it runs. You get the wrong hoof, that means the one-hour is about two hours. Ah, so there you go.

Jordan: Let's take on the GIF feature. You can generate an animated GIF up to eight seconds in length at two frames per second.

Jordan: So that means the model can output an image 100 times faster than — does it need to? Probably not. But it's not like the new model that generates fast motion. Alex, it just does a lot.

Jordan: So here's the thing — the latest form of AI generation, which this model controls, is going to select these elements. It's a fun thing to use in the workplace when you have a branding problem that you know is already solved.

Jordan: You can also make a GIF out of the same piece of content.

Jordan: But I think it's important to understand this is not necessarily a new video model for the workplace. It's in the middle of the process, generating a small frame-to-frame. That's interesting to think about.

Alex: I think that's a good way to think about it — image generation has improved to the point where you're seeing a new way to get your point across.

Jordan: Yeah, it's like a slideshow. Unnervingly — ah, hmm.

Alex: I can't create Animated GIFs with the API. Hmm, I can still comment in a gif. I'm not sure — can you?

Jordan: This is going to be a totally new format.

Alex: I'm about to take that challenge — three of those, I'm so sorry about that. It seems everyone, despite me, is going to have such an awful time.

Jordan: One thing that did come up, Nick — do you think the old-school models were a good idea? Was this kind of tradeoff costing GPT Image 2.6?

Jordan: Yeah, and it's like the opposite of a good product. It's not a well-strategized go-to-market. The old models were 2.5x more expensive to run than 4o, and it's very hard to get them to shift how these libraries map to different sources. They had to pick out new features quickly.

Alex: Yeah, there's an interesting edge there. Version 2.5 was actually two times faster than 1.5, starting with much more efficient models. There was a huge cost to bring it up to standard. Moving from there to 2.6, if there were a model that took the whole price to bring up to a much, much, much lower standard, it could afford faster operations.

Jordan: Look at what happened in 2.6 — you took the whole thing to a new level. The API itself can be a place where you can get a lot of work done that otherwise is something you have no idea how to do.

Jordan: I've talked about this before, but it's a good way to think about it. Once you have a defined set of quality standards, you can price accordingly.

Jordan: There is no model that can generate super-fast iteration, but I think it's fine.

Alex: If you open ChatGPT and do it right, you need to know there is a new way to represent it, versus how you did with the previous versions.

Jordan: And so there is the API developer. They'll add the same features to their own work. API users can get the same tools as ChatGPT users close to launch, rather than waiting for months. And that feeds back into all the products.

Alex: I think that's one of the biggest ones — building better experiences with the model. It's really meant to be used right after release.

Jordan: What's weird is, I don't think the model was actually the most important model. Two minutes prior, they took the whole thing and used it to generate the image.

Alex: No, it was exactly that bad.

Jordan: The quality was average.

Alex: Right, so five minutes later, I would have had to take the whole thing and lay it out.

Alex: We already have an episode on GPT Image 2.6, though, so this one is pretty much the same one– it's just that they say it might be done better than the version they'd already put out. Two years they've been doing that. In fact, I'll let you decide — which is better?

Alex: Right, so two years in — the question is, is it finding the equipment, or generating it ourselves, rather than thinking about what to do?

Alex: So here is what I think — if you want to keep your product images in this model, you can't. It's a simple process. Instead, keep your existing product in the reference library. It's just an easier process. This one is done, so you're ready for your next GIF.

Nick: And that is all I'm doing with the model from this point on — I won't be using the old model for anything at all at this point.

Alex: We'll post all the links and the full transcript on webguynick.com. Thanks for listening to the Web Guy Nick podcast, and we'll see you next Wednesday.

More episodes

  • AI Found 948 Ways to Root Linux, ChatGPT Makes GIFs, and a 2.8 Trillion Parameter Model You Can Download (November 5, 2026)
  • Muse Deep Dive: A New Brain, Money Tools, and Hard Lessons From Real Owners (October 27, 2026)
  • ChatGPT Pulse and Google's $12 Billion Week: What Actually Changed for Your Website? (October 22, 2026)

All episode posts live in the Podcast category and on the Podcast page.

Web Guy Nick LLC builds websites, graphics, and AI tools that help small businesses get found and get called. See how we help at webguynick.com.

author avatar
Web Guy Nick
    X
    Welcome to our website
    Start a project →