🦄 TUTORIAL: Mastering Midjourney's Seed Numbers
Many people don't realize they can retrieve the SEED NUMBER for each image generated on Midjourney + everything that can be done with it. Trust me, it's a game-changer for diving into AI explorations.
Let's go! 🧵👇🏼
#midjourney
Many people don't realize they can retrieve the SEED NUMBER for each image generated on Midjourney + everything that can be done with it. Trust me, it's a game-changer for diving into AI explorations.
Let's go! 🧵👇🏼
#midjourney
🚨 Beware! This is an ADVANCED Midjourney tutorial. Only for experienced argonauts. I don't know of any other MJ-focused creator who has covered this in such depth as you'll see in this tutorial. If you find one, send me the link, I'd love to learn even more!
First things first: what are seed numbers?
Diffusion models (like MJ) use a random seed number for each generation to create a field of visual noise (like TV static) as a starting point to generate the initial image grids.
Diffusion models (like MJ) use a random seed number for each generation to create a field of visual noise (like TV static) as a starting point to generate the initial image grids.
I don't want to bore you with the technical stuff, but what you should remember is that if you use the same seed number and the same prompt, Midjourney will generate two identical images (a few pixels may vary, but the human eye won't notice).
Let's see an example:
💬 "film still, vhs screengrab, I have beautiful cute dog multicolor like an unicorn inside the fridge, 80s VHS film --seed 1126902108"
These are two different Midjourney generations but with the same SEED. They're the same, right?
💬 "film still, vhs screengrab, I have beautiful cute dog multicolor like an unicorn inside the fridge, 80s VHS film --seed 1126902108"
These are two different Midjourney generations but with the same SEED. They're the same, right?
Actually, no. Due to the non-deterministic 🤓 nature of diffusion models, they are NOT exactly the same. Some pixels have changed. But it's hard for the human eye to notice!
It's like the "Spot the 7 differences" game on nightmare mode 😅
It's like the "Spot the 7 differences" game on nightmare mode 😅
In short, what we achieve by using the same SEED is the ability to ANCHOR the generation. That is, the same prompt will generate the "same" image over and over again.
Fantastic, you might be thinking: what's the point? 😄
Bear with me! It's about to get interesting.
Fantastic, you might be thinking: what's the point? 😄
Bear with me! It's about to get interesting.
But before diving into exploration, how can you get the seed number in Midjourney? Easy:
➡️ Option 1 (the simplest)
React with the ✉️ (envelope) emoji on the generated image. Yes, I'm serious! I show you in this video:
➡️ Option 1 (the simplest)
React with the ✉️ (envelope) emoji on the generated image. Yes, I'm serious! I show you in this video:
➡️ Option 2 (requires more steps)
You can use the command /show <Job ID #>
You can get the "Job ID" in several ways: docs.midjourney.com
You can use the command /show <Job ID #>
You can get the "Job ID" in several ways: docs.midjourney.com
Great! Now we can obtain the SEED for any image. Now, here comes the interesting part!
Have you ever wondered what would happen if you change just a couple of commas in the prompt?
Well, now you can! I removed all the commas from the prompt on the right...
Have you ever wondered what would happen if you change just a couple of commas in the prompt?
Well, now you can! I removed all the commas from the prompt on the right...
And just that, such a minor change in the prompt, made two little dogs appear on the TV 🤣 Not only that, the dog generated WITHOUT commas is more detailed.
Is there a way to anticipate which parts of the prompt will cause the most change in the image?
Yes! Use the /short command on the prompt and then click on "Show Details". We see that "dog" and "fridge" have a huge weight. Funny: because it's not generating the fridge 😆
Yes! Use the /short command on the prompt and then click on "Show Details". We see that "dog" and "fridge" have a huge weight. Funny: because it's not generating the fridge 😆
What if we change the lighting type in the prompt? How will the image react? Let's see!
🎴 Golden hour 🌃 At night ☀️ Sunny day 🔥 Fireplace
Prompts in ALT.
🎴 Golden hour 🌃 At night ☀️ Sunny day 🔥 Fireplace
Prompts in ALT.
What if we anchor the seed and change only one parameter like "--stylize" to see EXACTLY how it impacts our favorite image?
Well, now you can!
Well, now you can!
And how much impact does "--style raw" REALLY have? Let's see:
← Left: without
→ Right: with
Indeed, as we already knew, "--style raw" achieves higher levels of realism.
Prompts in ALT
← Left: without
→ Right: with
Indeed, as we already knew, "--style raw" achieves higher levels of realism.
Prompts in ALT
What if we anchor the seed and change the color of a character or object?
🔵 Blue 🔴 Red
🔵 Blue 🔴 Red
👉 What if we change the style referencing different artists/studios but anchoring the seed?
👉 What if we change the Midjourney version?
👉 What if we change the aspect ratio?
👉 What if we change the framing?
Etc, etc, etc.
👉 What if we change the Midjourney version?
👉 What if we change the aspect ratio?
👉 What if we change the framing?
Etc, etc, etc.
Do you start to see the point of ANCHORING the SEED?
Think about it...
Exactly! To iteratively TWEAK the final generation by seeing EXACTLY how the prompt reacts.
I've been doing this for months with brutal results.
Think about it...
Exactly! To iteratively TWEAK the final generation by seeing EXACTLY how the prompt reacts.
I've been doing this for months with brutal results.
In fact, that's how I came up with most of the generations in my prompt guide 🤖 BestAIPrompts. From iterative searching to find truly striking prompts. I highly recommend it to you, by the way! 😉
bestaiprompts.art
bestaiprompts.art
Ah! Important! Anchoring the seed works in any software based on diffusion models! I use this technique daily in Stable Diffusion, Leonardo, etc. to generate brutal images like these:
In conclusion, anchoring with a seed number is a great way to iteratively find the best prompts and see exactly how much impact they have on the final generation.
You're eager to try it, aren't you? 😄
But before you rush off to set seeds...
You're eager to try it, aren't you? 😄
But before you rush off to set seeds...
If you learned something from this tutorial and would like me to continue writing similar threads, an RT/REPLY on the first tweet of the thread will encourage me to keep doing it. Thanks! 😉👇
جاري تحميل الاقتراحات...