Fable 5.1: What Actually Changed and What to Use It For | EP 13
Is Fable 5.1 actually better, and what should a marketer use it for?
Fable 5.1 is a net improvement on Fable 5 for marketers, mainly because it is more concise and burns fewer tokens, which makes it cheaper to run for the same work. Its highest-value use is not writing copy. It is auditing the skills, agents, and markdown files you already built, finding the redundancies that were making them skip steps, and then running the cleaned-up prompts on a cheaper, faster model. Video work and translation are dramatically better than they were a few months ago. Image generation is not, and dedicated tools still win there.
In this episode, Andrew Foxwell, Thomas Moen, and Will Sartorius break down what actually
changed in Fable 5.1 and what it is genuinely good for. You'll learn why it is cheaper to
run than Fable 5, why the best use is auditing the skills you already built rather than
writing new copy, how to route tasks across different models instead of paying for the
expensive one every time, and why video translation went from unusable to production ready.
We discuss:
- What changed between Fable 5 and Fable 5.1, and why it costs less to run
- Why the highest-value use is auditing your existing skills, agents, and markdown files
- The prompt audit that found the redundancies making skills skip steps
- Why you should run cleaned-up prompts on a cheaper model instead of the flagship
- How to route specific tasks to specific models: planning, coding, research, execution
- Why less instruction now produces better output, and the Meta ads parallel that explains it
- The claim that Fable 5.1 copy carries a detectable watermark, and what that means for blogs and newsletters
- Why the writing finally stopped reaching for bold headers and three-item lists
- The rise of the AI-forward individual contributor, and how to become one
- Why one agency won't let a creative strategist touch an AI tool until they can do it analog
- Why Thomas's agency stopped hiring junior people entirely
- Localizing an entire instructional video library by dropping MP4 files in
- Why image generation still belongs in HiggsField and not here
- Tagging a folder of B-roll with Gemini Flash and cutting mashups in Claude Code
To connect with Andrew Foxwell reach him here Andrew@FoxwellDigital.com
To connect with Will Sartorious DM him here https://x.com/will_sartorius
To connect with Thomas Moen DM him here https://x.com/thomasmoen
To learn more about Foxwell Founders and conversations like this one, go here: www.foxwellfounders.com
Learn more about Foxwell Digital: https://www.foxwelldigital.com
All episodes: https://www.ai-dtc-wtf.com
Join The AI DTC WTF Newsletter: https://www.ai-dtc-wtf.com/aidtcwtf-newsletter
Join the Foxwell Digital email list: https://www.foxwelldigital.com/email
This episode is brought to you by Runneth by Motion: https://runneth.com/
AI DTC WTF is the podcast for people who work in DTC, use AI, and want to know what is
actually working right now. Every week: tactical tips, strategies, and ideas you can
implement in your AI workflows to make your brand or agency more money.
Chapters
00:53 Why Fable 5 was too greedy with tokens
01:30 Fable 5.1 is more concise and cheaper to run
01:45 Which model to use for planning versus doing the work
02:12 Why Anthropic keeps botching the marketing
02:26 The limits announcement that said 50% and meant 17%
02:42 Why skills kept skipping steps
03:23 The prompt audit that found the redundancies
03:53 You don't need the expensive model for everything
04:42 How to stay agnostic and route tasks across platforms
05:20 What Grok is genuinely faster at
06:20 Using Grok for small coding and research tasks
06:43 How to actually tell a model which model to use
07:14 Why stripping down your markdown files gets better output
07:31 The Meta ads parallel: less micromanaging, better results
08:19 Auditing a brief flow that was writing copy before finding personas
09:01 The watermark on Fable 5.1 copy, and what it means for blog posts
09:19 Why the writing finally stopped sounding like AI
10:53 The rise of the AI-forward individual contributor
11:47 Why creative strategists here can't touch the tools yet
12:18 AI as a compass, not a destination
12:39 How to learn creative strategy the old fashioned way
13:41 Amplifying good people versus replacing them
14:54 Why this agency stopped hiring junior people
15:49 Why Europe is further behind than the feed makes it look
16:56 Video and image generation on Fable 5.1
17:41 Localizing a whole Norwegian video library in a week
18:15 Why better reasoning makes project orchestration better
18:56 Dropping MP4 files in and getting a finished English version
20:03 Why images still belong in HiggsField
20:51 Tagging B-roll with Gemini Flash and cutting mashups in Claude Code
Full Transcript:
(00:04) If you work in DDC and you use AI and you're wondering what the F is going on every week, this is your podcast. The AI to see F podcast your home for tactical tips, strategies, and ideas that you can implement right now in your AI workflows to make your brand or agency more money. Welcome to another episode of AI, DDC, and this is a really an emergency episode.
(00:27) Sound the sirens. Get everybody together. Fable 5.1. What is going on with this? Let's talk about it. That's what we're here to get into. Today, I am joined by my illustrious co-host Thomas Moen and Will Sartorius. Guys, let's talk about it. What's what are your thoughts on it? What does it do? Well, is everybody speaking Claude language? What do we think about this? Thomas, what are your thoughts right away? Well, I have a lot of feelings, and I've actually I've been kind of upset about anthropic and club for the last 2 or 3 months.
(01:03) I've felt frustrated that it's getting dumber and it's just eating up on the tokens and all that kind of stuff. But I honestly, I felt a little bit in love again with fable 5.1, they will find was very greedy in terms of the token spend. And, you know, it would tell you 10,000 of words to answer simple questions.
(01:27) And my impression of fable 5.1 so far is that it is more concise and more to the point in the way it is communicating, and it's also not so token hungry, so it's a little bit cheaper to use, but I still use table one mostly and mainly to to plan what I'm going to do. And I'm still using Codex as the kind of the hands of what's actually doing doing the work.
(02:05) But fable 5.1 I'm I'm positive versus favorable. Five yeah, I would tend to agree net positive obviously anthropic. It's just really been botching their marketing. And it's just like obviously being the forerunner in the space is far more challenging than being the underdog. But by I think it was maybe a few days ago, they were saying, oh, we're increasing limits by 50%, but they were actually, you know, decreasing limits by 17%.
(02:34) And so like, you know, they have a marketing issue. So like for the marketers out there, you know, maybe talk to them, I'm sure there's great opportunities. But in any event, like, I think the thing that I found to be the most useful for fable 5.1 is we've built so many skills and, you know, marked on files or agents or whatever you want to call them.
(02:56) And oftentimes when I was running the skills, whether it's like creating a a pitch deck or a series of briefs, it would skip steps. And I always was trying to audit, you know, these skills and be like, why does Claude keep skipping steps like, this is so frustrating. And I'd have to say, hey, you know, to Thomas's point, it felt like it was getting a little dumber.
(03:20) Well, what actually was the most useful thing I found with 5.1 was just using, you know, slash clawed API prompt audit. And I had it go through all of my skills to find redundancies. And then as soon as that happened, I reran the the prompt to generate briefs for one of the creatives. And it happened and I just and I ran it through fable, just regular fable.
(03:45) And it happened much faster and it was actually far better. So what I'm seeing is that, like, you don't necessarily need to use this model for everything. It's an expensive model, but ultimately like using it to audit some of your past work and just like cleaning up your skills, cleaning up your ages, cleaning up your markdown files, I find that to be extremely valuable.
(04:07) And it also has just like cut out a lot of like fluff. Like if I look through like a markdown file of mine or I look through a skill of mine, I just like, I can't really tell, to be honest. Like, you know, is this helping? If I delete this, is this going to be problematic? So yeah, it's it's literally just slash cloud API prompt audit and it will go through you, you know, command shift the skills or the files you want it to audit.
(04:33) And it made it made a pretty big difference. Great use case. So many people going. It's great to audit it because ever see this stuff changes so quickly I think that's an awesome idea. The other thing that I've been hearing a lot of people doing is talking about like, okay, being basically agnostic around which platform is using, but using, you know, saying, hey, this is going to root to this task.
(04:54) This is going to root to this task for cloud. It's going to root to this task for grok. Like what are how have you guys done this, built this like in this 5.1 taking over more now or like how are you building a system where, hey, you're putting it in and saying, okay, I don't care what model you use to accomplish this.
(05:09) Use a combination of them, because that's what a lot of the dialogs around. Yeah, I think, you know, I'm I'm rooting more and more specific tasks to specific things. And I also actually honestly been quite impressed by my 4.6 and kind of the, the workhorse that is and it's it is so fast. It's really, really fast.
(05:30) So I find myself routing more and more tasks to different models. And I've spent a lot of my time just optimizing my whole setup of skills and MDP files and everything to be platform agnostic. So everything will just run smoothly on on all of the different kind of models that is out there. But just ask a question on that.
(05:56) So, so one is, is like grok is like, what are you using it for? And how are you assessing that? It's faster. Like, I just how do you do that? And then like the second thing is like how do you build a system of going through and like having a choose automatically. Like what what what's technically in between there? I mean, I sort of know, I think, but I'm just curious how you have it set up.
(06:17) Yes, absolutely. So, so first of all, I use grok a lot for small, simple coding or research tasks because it, again, is quite quicker than the other models, and I like it better than using the term model for from, from OpenAI, for example, or the opus agents when it's small quick tasks in terms of how you get the model to kind of switch and understand what is going where, it is as easy as just telling the model that, I would like you to use this and this.
(06:54) And now kind of in my muscle memory, I will say using opus five subject for this, use a subject for this, and it will just kind of go out and do it. And I have a few friends who are using their their agent or cloud MD to also set in rules and kind of automate kind of these kind of things. But I've been stripping down all my MDF files a lot, and I think as the models are getting smarter and smarter, the less instruction you as a human have on the models, the better it delivers.
(07:30) Just like back in the days, you know, when you were doing meta ads, you had to really control everything of who you were targeting, etc., etc. remember six, seven, eight years ago it was all about just micromanaging things all the time. But now, because, you know, Meta's so smarter, you don't have to interfere that much.
(07:49) And in my opinion, that is also the stage where we're getting with these models. Now that the more you kind of overcomplicate your your MDM files and the more you send your instructions, the the more confused and the less good results you get. I don't know. Well, what's your experience? Yeah, I tend to completely agree.
(08:08) I think I bloated a lot of my markdown files early on, and I just assume like that they just like this was the best way to go about it. And again, in auditing our brief flow, which you know very quickly what it does is like, if I'm starting from scratch, it will use apathy, it will scrape subreddits, it'll try to find net new personas, you know, having already pulled in the ad account data and tagged all the ads, we have a good sense of like what new personas we can be testing.
(08:41) But, you know, what was happening before is it was writing the copy before it was actually finding, you know, this up, Sonus and read it. And I didn't really realize that. And I was like, this is why our copy feels so, you know, it lacks persona orientation. So ultimately having that audit and what I will say is that, like the caveat here is fable 5.
(09:05) 1, all of the copy it writes is going to have a watermark. So just like be aware of that. Like if you're writing a newsletter or, you know, putting out, you know, blog posts with, with fable 5.1, it'll be very easy to detect that you've done that. But what I will say is 5.1 the copy style. It feels much less like conventional AI.
(09:27) It doesn't like keep reaching for these, like sort of like bold headlines and like, if it's not this, it's that it is far more organic in a way that I hadn't seen before, because ultimately I would have to write copy for ads. And then, you know, and my company still does this, you know, we rewrite a lot of the cloud copy or we're just like sort of starting from scratch, but it's it you're getting far less lists, like, you know, you're not getting those bold headers and then like, you know, the the paragraph. Right.
(10:01) And you're able to tell right away. That's the one thing that I've noticed as well. Yeah. That was the immediate tell for me. And I think that the like authentic look to it is going to change because I think that's, that's like the big one of the big complaints of super users is, hey, like this all looks very similar.
(10:19) You can tell it immediately and you can see people complaining about it. So the more that that gets removed and the more that it can also scan like understand your your native tone and everything. If you attach it, the faster or adapts to that, the better off it's going to be in terms of stickiness, because then people will say like, okay, well, I can just use this over here.
(10:37) People are going to be aren't going to be able to tell that AI is using this. And I think it's interesting. So as it relates to these models are going to keep changing. These models are going to keep updating and getting faster and different, etc.. And you know, Reza from Motion Run has been talking a lot about the rise of an icy.
(11:00) I'm sure you've seen this an individual contributor in terms of somebody who's who's, you know, really AI forward in an organization, but also has the fundamentals of being a great marketer. How do you think, just as people go through this, like how do how can they position someone or themselves in an organization to be ready to continue to test and understand these things and these models and these changes? And what does that look like in marketing organizations and even six months by the end of the year? Yeah, that's a great question.
(11:39) And Thomas and I may differ on this one if you ask me this question, probably 6 to 8 months ago, I'd have a much different answer. You know where I'm at today again, like sort of coming out of the AI psychosis state is, you know, internally, we don't let any of our creative strategists touch any of our internal tools until we can trust they can write creative strategy, analog style.
(12:03) The tooling out there is best for amplification of existing hypotheses or theses. Right. And so like what I tell my team and what these models are very good at is serving as a compass, very good at pointing you in a direction, but you don't know what the end goal is. But if you start walking in that direction, you'll start seeing things that actually might make sense.
(12:30) And so to that end, if you are sort of like looking to get into the field as much as it sucks like strategy the old fashioned way and like, what is the old fashioned way? Well, it's just like figuring out, you know, your ICP, you know your persona, figure out their problems, go onto a subreddit. Maybe your persona is like someone with ADHD.
(12:53) Go on to those subreddits, figure out what those people are struggling with, and then find specific angles for your ADHD persona. And that's your angle or angle or problem. And then write copy based on that angular problem. Right? Oftentimes you can just take the copy verbatim from the subreddit, and that works great.
(13:12) Once you sort of get the hang of knowing how to identify the ICP and the angle or problem and writing the copy, then you have a foundation of what could create a strategy is, and then you can start utilizing these models to amplify that process so you can have it go search for. Net new personas for you that you haven't tested.
(13:29) You can go have it, search for angles that you haven't tested, and then you can have it help write copy for you. And so, you know, to that end like you know asi AGI side, right now we are in a space where you can amplify really, really good people. And if you have the foundation of strategy under your belt, then you can really, really ten x yourself in a way that I haven't seen others do.
(13:59) And so like again, in our organization we teach analog style. And once you sort of pass the OKRs analog style, we introduce the tools to you. And then people are like, oh my God. Like I can be so much more powerful now. I can cover 2XA number of brands and get the same results. And I think that's I mean, I think that's the argument for one of the main arguments for, for, for utilizing and superpower is super powering.
(14:22) I just I think the organizations keep shifting. I think what you're, what you're saying is really smart, which is starting with the basics and looking at this, Thomas, what do you think in terms of like where the marketers going to look like in this and understanding the changes in these models and stuff, because, you know, you're spending a certain percentage of your time looking at this all the, you know, and messing around with it and testing and stuff.
(14:45) So, you know, is it you? Is it the founder? Is it somebody else? Is it, you know, what does that look like? Because I think it's I think it's changing a lot a lot. And I'm sorry to be boring. I think I tend to agree with, with Will most of the things he was saying. So so first of all, I think you really have to be good at your craft and you have to do the reps.
(15:06) And the way I'm looking at it now is in our agency. We stopped hiring junior people and we're just looking for all dogs with ten, 15 years of experience that has that gut feeling, that intuition of what's what's good and how processes worked and how it used to work. Because in my experience, those kind of people is the one who can really 10 or 100 extra their output because they know what they're talking about.
(15:32) Right. So, so I think that having the craft and have done the reps is super important. Standing point. And I think you would grow and position yourself much faster if you understand the craft that you're trying to amplify. But I also I've spent a lot of time in the last few weeks, both in Norway, Denmark and in Finland, in Scandinavia, and also a little bit talking to agencies in Spain.
(16:00) And I realized that we are all we as in the part we are looking at the top tier of things, of the YouTube videos, the expo, the expos, etc. and we feel like we're behind. But but the world is really, really there, much slowly adapting to this AI world than I think that we feel like it's doing. I feel like I'm behind.
(16:28) But when I talk to both big ideas in smaller agency all over Europe, they're they're like mind blown about the things we're doing and how we're thinking about things. But what I'm saying is that if you love your craft and you understand the crafts have done your reps, I don't think it's so hard to position yourself because you just have to be curious and try to figure out and ask and have a conversation with these different tools on how you could amplify the more effective share work or accelerate. Cetera.
(16:57) So I think that loving your craft and having a little bit of that entrepreneurial curious is a good match to, to to get a much better output. It's interesting. So in fable five one you guys feel like the language is more native. You feel like it's it's faster to a degree. You feel like it's a it's able to do complex tasks.
(17:21) I mean, I've certainly seen people talking about image generation and video generation being better, being like world's better on five one. Yes. What are the other main big things that you guys see? See that's really changed in 5.1? Just just to point out the video part, because we have a lot of instructional videos on, on in Norwegian and the last few weeks or the last week we spent time with fable 5.
(17:56) 1 actually translate everything. And Claude have been creating ads, creating voiceover, creating text, etc.. Yeah, so, so video creation is so much better now and the quality is super high. And we'll talked about earlier on that that it's writing is better. But I also think it's reasoning is better. It's project management of orchestrating your project is is much better because again the language and the thinking is better and that makes everything else better.
(18:28) So for me, having it look at your projects and finding patterns or finding out what to focus on is also something I got a lot much more out of in five one. Yeah. I'm curious your take on sort of the video component as well. Like I assume you're not having anthropic models build the videos, right? You're using a model and then you're we used to use Higgs Field and all that stuff for that, but not with 5.1.
(18:59) I'm basically jumping in the video saying, here's an original video, make an English version, make slides for it, sure got voiceover, etc. so so so this is more instructional videos for how to use a platform, how to think about creative work for our clients more than creating ads and those kind of things. And that used to be really sucky, you know, a few months ago, you're not able to do that at all.
(19:25) But the quality is surprisingly good now. So we really ramped up our production of of our Norwegian videos because we're now starting in Finland and in Spain, etc.. So, so localizing our content with basically just dropping the MP for files in and say, hey, make this into an English version, make this into a finished version.
(19:48) It is. It is really good these days. Cool. I that's really smart. I have like consulting clients that are have always been asking me that like translating videos. That's interesting. And so that's wow okay. That's that's incredible. Okay. And have you found the same for images as well. No, I honestly I haven't really tried images because.
(20:09) Yeah, because of just how bad it was. But I expect it to be better. But but when you're talking about creating images and videos for, for ads and those kind of things, then I think Higgs Field and the other usual Suspects are a much better way to do it. Yeah, no, I completely agree. And obviously I've talked about sort of my qualms with Higgs field in the past here.
(20:38) But one thing that I thought fable did a very good job at is like, and we sort of built the system for this, and you have to sort of combine it with Gemini, and maybe I'll talk about this more in the New York meeting. But like if you feed a folder of b roll content, what you can have it do is have it's slice and dice the B roll content, and then have it tag each piece of B-roll content and then have Gemini Flash go and like visualize each of those and add additional tags.
(21:08) And then you can have create mashups very easily. And I would say like with a fable model is probably 80% of the way there. Like you would sort of have to go back and forth in cloud. Could be like, hey, can you just like move this here, move this there? I also think like cloud code, this may be a hot take is like a really good editing software.
(21:25) Just like you could just be like, hey, you know, can you just like, move that this up thing? They also have a really good markup tool that I think is pretty slept on. Yeah, yeah. And then, you know, to that end, I candidly haven't tried this, this workflow with 5.1, but based on what you just said, I think there's probably serious legs to it. Yeah.
(21:43) You should. You'll be impressed. Oh that's amazing. That's a really good tip. Cool. Well, I feel like we've covered a lot of good stuff. Many more things on fable five one to be coming, but I think you for listening to this quick episode, everyone, and have an awesome rest of your day. We're always here for you and we appreciate any.
(22:03) Review your feedback, so take care and we'll talk to you soon. Bye! Thanks all.
Made With AI