How To Find New Styles in Midjourney & Use Genspark to Test Image Prompts

A "Video Guide" and a "Hot Take" walk into a bar...

Happy Thursday, folks!

Huge thanks to everyone who voted in last week’s polls.

“How to” video guides and a new “Hot Takes” section were among the most popular initiatives:

That sure is convincing!

Less convincing but still solid.

You also said you wanted me to tackle them as the first priority:

Well, ask and ye shall receive!

I went into action mode and can now bring you a sneak peek of both initiatives.

---

📹Video guides

By far the most popular of my proposed initiatives was the “How to” video guides.

I’ve already been thinking of posting such guides to YouTube, and you just gave me the final nudge to start doing that.

Here’s my first stab at it and a preview of things to come.

How to discover new styles with Midjourney

With Midjourney images, you will often be trying to nail a specific scene and will have an idea of the style you’re after.¹

But there might also be times when you want to stumble upon new styles and get inspired.

If so, this guide is for you. I’ll cover:

1. Two ways to discover new styles:

   1. Using --sref random

   2. Using --chaos 100

2. Three ways to reuse those styles for future images:

   1. By using a known --sref number

   2. By appending a reference image

   3. By creating a new Moodboard

Enjoy:

If you want to test my moodboard style code from the video, here is the number:

--p m7278031216647864324

Did it work as I’d hoped and apply my style to your images? Let me know in the comments!

---

🔥Hot takes

Hot Takes is a new section where I will occasionally test drive new AI tools, models, or features and share my quick thoughts.

Unlike the regular Thursday and Sunday posts, Hot Takes will only go out sporadically, if and when I find something worthy of attention.

If Hot Takes aren’t your cup of tea, simply go to your account at www.whytryai.com/account and toggle the “Notifications” settings accordingly:

I expect to have the following standard structure for Hot Takes:

1. What is it? - short intro to the tool or feature.

2. How do you use it? - a guide (possibly video) that shows how it works.

3. Why should you care? - my thoughts on its benefits and practical applications.

For a sneak peek at what a Hot Takes entry might be like, here’s my look at Genspark’s "Generate Image” agent.

Genspark lets you test image prompts at scale

Genspark isn’t new, but I recently discovered its Generate Image agent and wanted to use it as my Hot Takes guinea pig.

1. What is it?

At its core, Genspark is an AI search engine similar to Perplexity.

But it also offers a collection of agents for research, chatting, image creation, and translation:

I want to focus on the Generate Image agent, which—among other things—lets you test the same image prompt in multiple image models.

2. How do you use it?

First, click the “Agents” tab in Genspark’s left-hand menu, then select the Generate Image agent.

Right now, it includes four popular image models:

• DALL-E 3

• FLUX (.1 and 1.1)

• Ideogram 2.0

• Recraft V3

(Notably, a few big names are missing, like Imagen, Midjourney, and Stable Diffusion.)

To test how your prompt performs in different models, use the following settings:

• Mixture-of Agents: Leave this on instead of selecting a specific image model.

• Auto Style: Leave this on instead of selecting a specific style.

• Aspect Ratio: Set this to whatever ratio you need (I’m using 3:2).

• Auto Prompt: Leave this unticked. You want to compare your “pure” prompt rather than letting the chatbot modify it for different models.

Now, simply enter your prompt.

For my example, I’ll use this:

Photo of a man hiking in a forest

Let’s look at the results:

Genspark’s agent picks four models that it finds appropriate for a given prompt.

At this point, you can zero in on the most promising model and work further on the prompt itself to get the result you need.

But that’s not all!

You can also use a reference image to guide the models.

Simply select the paperclip icon on the left:

Now upload an image that represents what you’re after to “nudge” the models in the right direction. I’ll use this one from Midjourney:

Like so:

I get the following results:

The results are decidedly not photos, which tells me I should use reference images in my intended format.

But doing this gives me a good idea of which image model best captures the reference character.

3. Why should you care?

As you know, I always encourage people to try image prompts in multiple models, because they might treat the same exact prompt quite differently.

Genspark makes it really easy to compare four models side-by-side with one click. It’s really convenient to be able to do this without having to switch between different sites.

In addition to that, you get the option to use reference images with models that might otherwise not allow it.²

So if you’re trying to create an image but aren’t sure which model is the best for it, give this Genspark agent a try.

Here’s my invite link that grants you a free month of Genspark Plus.

---

🫵 Over to you…

As you can see, I take your feedback to heart.

So let me know what you think of these upcoming initiatives and tell me how I can make them more useful and relevant to you.

Leave a comment or drop me a line at whytryai@substack.com.

1. If that’s the case, you might find some of my Midjourney posts helpful.

2. For instance, uploading a reference image is a paid feature on ideogram.ai.