Full transcript
0:00Someone needs to tell all of these AI
0:02companies to slow down because we have a
0:03yet another model to start testing out
0:06and seeing if it's good for creative
0:07writing. And that is Gemini 3.1. If you
0:09saw my videos in the past on Gemini 3,
0:12it quickly became one of my favorite
0:14models for writing and I even thought
0:17that it was at the same level as the
0:18Claude models, which traditionally have
0:21been better for writing. But today we
0:24have 3.1 right off hot off the heels of
0:27Claude Opus 4.6 and Claude Sonnet 4.6,
0:30which were recently released. And if you
0:32saw my Sonnet video, which I literally
0:34recorded just like an hour ago,
0:37you'll know that I actually was
0:38pleasantly surprised by that model. I
0:40thought it was really good overall and
0:42was excited to get into this one, Gemini
0:443.1, to see if it's just as good or
0:47better than what we had before. So let's
0:49dive in.
0:55First of all, we have our logline prompt
0:57where I ask it to give us a bunch of
0:58loglines for an epic fantasy. I do keep
1:00these prompts intentionally simple to
1:02see how good do these models do right
1:05out of the gate with no special training
1:07wheels or examples of my prose or
1:09anything like that with one exception.
1:11And so this is the one that it gave us.
1:13And I do think that some of these ideas,
1:16while they are a little bit generic and
1:18relying on some really old tropes like
1:20this chosen one here, I do actually
1:22think that Gemini does just a little bit
1:25better at brainstorming than the Claude
1:27models. Sorry, I'm going to do a lot of
1:28comparisons to the Claude models because
1:30I've recently reviewed those two models.
1:32It's just in my head right now. I do
1:33think it does a better job at being a
1:35little bit more unique. And honestly,
1:37the Gemini models have been better at
1:40things like brainstorming and stuff in
1:41the past. I remember the first Gemini
1:43models at the time were the best ones
1:46for brainstorming compared to other
1:47models, which tend to be a little bit
1:49more flat and a little bit more generic.
1:50And I do think that is the case here,
1:52just looking at some of these. Like when
1:54the prophesied hero is accidentally
1:55killed on the eve of a demon invasion.
1:57His cowardly squire must pose as the
1:59chosen one to rally a fractured kingdom.
2:01He must face down a dark lord who cannot
2:03be defeated by a fraud, risking
2:04everything to become the legend he is
2:06pretending to be. It's not necessarily
2:09a full It is not really an original
2:11plot, but you know, it's different than
2:12what I get out of Claude models. Does
2:14that make sense? One thing I did like
2:16about what it gave me here is it
2:17actually divided it up into sections.
2:19So, here it gave me five log lines about
2:21misfits and outcasts. And then we had
2:24five about that. And then here it says
2:26five log lines about curses and magic
2:28gone wrong. So, the fact that it did
2:30this kind of categorization actually did
2:32result in a little bit more of a variety
2:35of ideas here. So, for instance, we have
2:37banished to the frozen waste for
2:39possessing forbidden fire magic. A
2:40disgraced pyromancer must master the ice
2:43sorcery she despises to defeat a rogue
2:45elemental god. In a cruel twist of fate,
2:46she discovers her hated, uncontrollable
2:48flames are the exact key to saving the
2:50people who exiled her. You know, so
2:52things like that in different, you know,
2:55the different categories here. But, if
2:57I'm being honest, nothing particularly
2:59groundbreaking here. Kind of similar to
3:01a lot of the other ideas that I get from
3:04this prompt from other models. So,
3:06definitely, you can get good
3:08brainstorming out of most AI models
3:10these days, but you do have to supply it
3:12with more of your own uniqueness, your
3:14own ideas. Otherwise, it's going to
3:16default to the the really generic ideas.
3:19Quick pause. If you're an author who
3:20wants to start writing more books, I put
3:22together my complete AI prompt pack that
3:25helped me and many others in my
3:26community write a book in a month. It's
3:28completely free. Hit the link in the
3:29description. Join the free waitlist for
3:31Story Hacker Gold. Then you just answer
3:33a couple quick questions, and I'll send
3:35you the full prompt list plus a few
3:37extra free goodies. I'll also email you
3:40when I open the paid Story Hacker Gold
3:42group, so you can join if it feels
3:43right. There's no commitment to join,
3:45though, and you'll still be able to get
3:46all of the freebies you signed up for
3:48either way. All right, back to the
3:49video. All right, now let's look at the
3:51outline. So, I actually thought the
3:52outline was pretty good in comparison to
3:54the Claude models. The Claude models all
3:56went into a a lot more depth per
3:58chapter, which might seem like a good
4:01thing. For me personally, I like
4:02shorter. And if you look at these, we
4:04have like chapter one, chapter two,
4:06chapter three. This is about the level
4:08of outline that I like just in terms of
4:11how much detail we need per chapter, and
4:12then I sort of figure it out as I go,
4:14but that's me personally. A lot of
4:15people like far more detailed outlines.
4:17And if you are one of those people, the
4:19Claude models are definitely going to be
4:21better for you. This one was fine. It
4:23did tell a cohesive story. I was able to
4:25get through more of it because it was so
4:27much shorter. It did tell a cohesive
4:29story, but I don't think there was
4:31definitely was not giving me a whole
4:32bunch of detail. And if I were somebody
4:34that wanted more detail, I would
4:35probably pick the other ones. I do also
4:37feel like Claude, all of the Claude
4:39models have have traditionally been
4:41fantastic for outlining. It's one of the
4:43things that they seem to be good at. So,
4:45I would give the edge a little bit to
4:47the Claude models in this case. All
4:48right, going to the basic proms. This is
4:50where I ask it to write a a quick scene.
4:53I give it just a brief description of
4:54the scene, and I don't really give it
4:56any explanation about how to write it or
4:58anything like that. Just ask it to
5:00write. Unfortunately, I was a bit
5:02disappointed by the results here, and I
5:03had to actually go back to my original
5:06results for Gemini 3, and I looked
5:08through that and I was like, "Man, did I
5:09like this?"
5:11I mean, it's not bad, and I think at the
5:12time I was making it we didn't have that
5:15many good options yet still. And but I
5:18feel like the Claude models have gotten
5:19a little bit better since then, and
5:21having just done the Claude model the
5:23reaction video to Claude's sonnet 4.6,
5:26and then coming back to this, I think I
5:28was a bit disappointed by this. First of
5:30all, you notice there's some gigantic
5:31paragraphs here, which no one likes to
5:33read through, which is funny because I
5:35often feel like Gemini doesn't do that,
5:37but that's what it seems to have done in
5:39this case. And looking through it, let's
5:42read through the little moment that I I
5:43I tend to read through the most. Let's
5:45see. His knees buckled, catching on the
5:47sharp outcrop of flint and he crashed
5:50hard into the earth. A ragged sob,
5:52there's a AI slopism right there, uh
5:55from his raw throat as he forced himself
5:57up casting a desperate bloodshot gaze
5:59over his shoulder. Like there's a lot of
6:01this sort of overly dramatic adjectives
6:03and things that are kind of indicative
6:05of AI here. There cresting the hazy
6:07horizon where the darkened sky met the
6:09earth was the scattered figures of his
6:11nightmare. They moved with a terrifying
6:12unnatural cadence, the jerky
6:15motion, arms twitching and legs lurching
6:17forward as if their limbs operated on
6:19decaying rusting hinges. Through the
6:21dimming light, Haldor counted at least
6:22200 of the fiends spilling over the
6:24rocky ridge. They wore no armor and
6:26carried no weapons for the hungry needed
6:28neither. Uh their gaunt stretched skin
6:30was the color of wet ash blending so
6:32perfectly with the gray
6:32twilight-drenched wasteland that it
6:34seemed as though the very stones had
6:35sprouted limbs and began to pursue him.
6:37So definitely like way more description
6:39than we needed. Uh it's it's certainly a
6:41kind of an important dramatic moment,
6:44but I feel like the amount of drama here
6:46is like over the top and what I got out
6:48out of Sonnet 4.6 was a little bit
6:50better and more natural, I feel. Uh
6:52going to the advanced prose prompt when
6:54I where I give it the same scene but I
6:56do give it examples of my prose and
6:57other things for it to follow, you can
6:59see first of all they did a little
7:01better on the paragraphs, which it
7:02probably got from my the samples of my
7:05work that it was looking at. But we can
7:07if we read that same moment here, Haldor
7:09spun his head and searched the horizon.
7:10Figures crested the distant ridge line.
7:12They did not march or run, they lurched
7:13forward with spasmodic joint-locking
7:15spasms, flinging their limbs forward to
7:17drag their bodies across the uneven
7:19terrain. He counted two dozen of them
7:20spread out across the ridge. The hungry,
7:22their molted ash-gray skin mirroring the
7:24barren rocks behind them, rendering them
7:26near invisible save for their twitching
7:28silhouettes against this dying sun. They
7:30closed the distance, they tracked the
7:31scent of his exhaustion. I do think this
7:33is better. It's definitely better than
7:34the previous one that we read. So and I
7:36actually think that this is often true
7:39of the Gemini models in my experience so
7:42far, I think Gemini is actually better
7:44at molding to the style that you ask it
7:47to mold to. I I think it responds better
7:51to your prompts in terms of style than
7:54Claude models do. That's just my like
7:56intuition there based on my experience
7:58so far, but I think Claude is out of the
8:01box better. And you can still manipulate
8:04the Claude models and get it to to write
8:06in your specific style and stuff, too.
8:08I'm not saying you can't. I'm just to
8:10say I think Gemini does that a little
8:12bit better, but the vanilla version of
8:14both, Gemini is not as good, I think.
8:17That that's my new assessment there. I
8:19then we move to the dialogue prose
8:21template where it's a different scene,
8:23same story, but different scene with a
8:25little bit of dialogue added in there.
8:27Let's see. The prince speaks with
8:29admirable compassion, your majesty. Alby
8:31with the passionate foolishness of
8:33youth. A silken voice slithered from the
8:35shadows behind the throne. Argol stepped
8:37into the light, his dark velvet robe
8:38sweeping silently over the polished
8:40floor. The royal advisor offered Lathan
8:42a condescending smile that didn't reach
8:43his dead calculating eyes. My
8:45hot-blooded prince, the statutes you so
8:47eagerly wish to abandon are the very
8:49bedrock of Varlden survival. If we crack
8:51the gates for every pursued soul, we
8:53risk letting the monsters inside. Surely
8:55even you can comprehend that our
8:56rigidity is for the safety of Varlden as
8:58a whole. A tragedy outside the gate,
9:00yes, but a necessary one. Yeah, I don't
9:02like this. It's a little over the top.
9:03It's a little windy wind baggy and not
9:06something that I would probably use
9:08without a lot of editing. It is what it
9:10is. Then I ask it to take the same scene
9:12but written by a really poor model from
9:14like back in the day and see if it can
9:16improve on it. I did not see anything
9:18particularly better about this.
9:20It keeps a lot of the same dialogue and
9:23the description is just not much better
9:26if I'm being honest. Although all of the
9:27models that I've tested recently do make
9:30improvements on the original scene that
9:32I give it, very few of them do so in
9:35such a way that I wouldn't just rewrite
9:37the scene with a newer model instead.
9:40That's That's easier to do than to try
9:42and improve on the scene
9:44on the original scene. All right, so now
9:46let's go to add headline to see how well
9:48it does with this. And I actually
9:50thought it did a pretty good job here
9:52overall. I If you remember my Claude
9:54Sonnet 4.6 video, I mentioned that a lot
9:57of the ads were really kind of too long
10:00and I agree here. These are still too
10:02long. Like we want it about half this
10:04size to get good ad headlines cuz
10:06brevity is king when it comes to ads.
10:08They exiled the prince to die in the
10:09waste. They didn't realize what he would
10:11find out there. He thought the mountain
10:12city was built to keep the monsters out.
10:14He was completely wrong. Everything the
10:15mountain king told him was a lie. Out in
10:17the waste the exiled prince just found
10:19out why. So some of these are good. I
10:21wouldn't say they're that much better
10:23than anything else, but like they're
10:25they're not bad. They're certainly not
10:26bad. They're all kind of running off of
10:28the same sort of thing. Once again
10:30though, it did actually give me
10:32different categories of hooks. So we
10:34have like the monster and survival
10:36hooks. How does a betrayed prince
10:37survive a wasteland of energy vampires
10:39and the ravenous dead? And then down
10:40here we have the rebel and discovery
10:42hooks. The exiled prince thought
10:44humanity was dead outside the mountain
10:46then he met the wasteland rebels. So it
10:48did divide them up which I like, but
10:50overall nothing particularly
10:52groundbreaking here. All right, let me
10:53get the email newsletter prompt. I was
10:55not too happy with the Claude models for
10:56this. So let's see how Gemini 3.1 does.
10:59I am absolutely thrilled to share some
11:01exciting news with you today. I've just
11:03finished writing my newest epic fantasy
11:04novel The Crown of Hunger and it will be
11:06officially released on December 21st. If
11:08you are a fan of The Count of Monte
11:09Cristo, but always wished it featured
11:11zombie-like hordes, shadowy magic, and a
11:13brutal fantasy landscape, this is the
11:14book you've been waiting for. Here's a
11:15quick taste of the story. Um so I
11:17actually I think this is better than
11:18what Claude was giving me for sure. So
11:20maybe Gemini is a little bit better at
11:22like natural sounding non-fiction speech
11:24or something like that. Like maybe if
11:26you want to use it in an email or
11:28something or in an article or some kind
11:30of non-fiction, maybe Gemini might be a
11:32little bit better cuz I do think this is
11:34it's more succinct, it's a little bit
11:35better
11:36than what I was getting out of the
11:37Claude models. All right, last but not
11:39least we have the SEO article super
11:41prompt that I give it where I give it
11:43this enormous prompt asking it to write
11:454,000 words of an article titled how to
11:47write a fantasy novel. So, let's check
11:50this and just see how many words it
11:52wrote to start with. 1,700, so not very
11:56good here and if I'm looking through
11:58this, yeah. Okay, so one of the issues I
12:01typically see with this prompt is a
12:03prompt that doesn't go through and write
12:06the whole thing out will kind of make
12:07just like a more detailed, more fleshed
12:09out outline, but not actually get into
12:11the point where I would want it to be
12:13the full article. So, I would prefer 3
12:16to 4,000 words here and it only wrote
12:181,700, which having done the research on
12:20this particular article topic, uh my
12:23background's in SEO. Uh like I chose the
12:25number 4,000 for a reason. I would
12:26definitely like if I were paying a human
12:28to write this article,
12:30I would not be okay with 1,700. I would
12:32ask them to do 4,000.
12:34So, that's a fail on the SEO article. I
12:37have noticed that the Claude models tend
12:39to be better for writing lengthy things.
12:42Like the outlines, uh the outline that
12:44uh Sonnet 4.6 wrote for me was um
12:48it's like 17, 18,000 words or so.
12:50And and this one was a lot less and then
12:53obviously the article is also a lot
12:54less. So, if you want something to
12:57really go to town and flesh out a whole
12:59big thing, go with the Sonnet models
13:00instead, it'll do better generally. All
13:03right, that pretty much does it. So, my
13:05final thoughts on Gemini 3.1, like I
13:08said, it almost felt worse to me, but
13:10then when I went back and checked my
13:12results for Gemini 3, I don't think it
13:14actually was worse. I I just think I my
13:16standards have risen a little bit. And
13:19now looking at Gemini 3.1, I'm not
13:21really seeing a huge difference from
13:23Gemini 3. So, it appeared worse to me,
13:26but I definitely don't think it is
13:27actually worse than Gemini 3, but I do
13:29think that the Claude models have caught
13:31up a little bit and are like I'm happy
13:34to go back to those for most of my prose
13:37generation. Now, I do think Gemini 3 has
13:39been fantastic for anything regarding
13:41like reasoning, like thinking through a
13:43scene and telling me what should be in
13:45the scene, that sort of thing, and then
13:46I go to a Claude model to actually write
13:48the scene. That's more of a workflow
13:50that I would still use and still trust
13:53Gemini 3.1 to do. So, those are my
13:57finishing thoughts. Let me know what you
13:59think. Go ahead and, you know, see what
14:01your tests show and I'd be happy to see
14:04what those are in the comments below.
14:06So, be sure to let me know and I will
14:07see you in the next video.