#+title: S02|L12 - midjourney pt3 - make ai image with reference prompts #+HTML_HEAD: #+OPTIONS: H:6 * Links #+attr_html: :class links - [[../../toc.org][TOC - ecomm clubhouse]] - [[https://jellyfin.ronnyabraham.com/web/index.html#/details?id=44219ee1aad759c73033d0af72f346b4&serverId=99488804638e465693eed17924c57b60][video]] * Notes - you want to create a lot of images to create a lot of takes ** midjourney discord *** import - upload image into the chat - click on it - right click on popup - get image address *** using - paste the address into prompt /imagine - then add the rest of the prompt - midjourney will use the image as a reference guide ** image weight - ~-iw [0-3]~ - specifies that the reference is to be maximized - default is 1 - maximum is 3 ** other purpose - you can use the reference as way to determine the pose * Review - Using Reference Images in Midjourney ** Summary This lesson explains how to use reference images in Midjourney to get closer to the image you want faster. Reference images help Midjourney understand: - pose - composition - structure - style direction - what should or should not appear ** Core Idea Reference images reduce wasted generations. Instead of relying only on text prompts, you give Midjourney a visual example. This helps it skip many wrong outputs. ** Why Reference Images Help Text prompts can be vague. Example: #+begin_example minimalist black line art of a sunflower top down on a white background #+end_example Problem: Midjourney may still add: - stems - leaves - extra angles - details you did not want A reference image can show: - only petals - no stem - no leaves - top-down composition ** Basic Reference Image Workflow [0/5] - [ ] upload reference image into Midjourney / Discord - [ ] open the uploaded image - [ ] right-click and copy image address - [ ] start a new =/imagine= prompt - [ ] paste image URL before the text prompt ** Example Prompt #+begin_example [image URL] minimalist black line art of a sunflower top down on a white background #+end_example ** Important Point Midjourney does not simply copy the reference image. It uses the image as guidance while also following your text prompt. So the result is influenced by: - the uploaded image - the written prompt - Midjourney’s own generation process ** Flower Example Without reference image: - many outputs had stems - many outputs had leaves - several images were not close enough With reference image: - outputs stayed closer to the petal-only flower - no unwanted stems - no unwanted leaves - fewer wasted generations ** Using Internet Images as References A reference image can be: - something you generated earlier - a public domain image - a product/design reference - an image found online Warning: Use reference images carefully. Do not copy protected artwork directly. Use them for structure, style direction, or composition, not pixel-copying. ** Corgi Example The tutorial uses a corgi image as a pose/style reference. Goal: Create a new corgi image in a specific artist/style direction while keeping the pose closer to the reference. ** Image Weight Image weight controls how strongly Midjourney follows the reference image. Format: #+begin_example --iw 2 #+end_example In the tutorial: - =--iw 2= was used - this was the maximum image weight in that Midjourney version ** Image Weight Effect *** With High Image Weight Midjourney prioritizes the reference image. Result: - closer pose - closer structure - closer overall composition - less reinterpretation *** Without Image Weight Midjourney uses the reference more loosely. Result: - more influence from the written style prompt - more freedom - less resemblance to the reference image - more reinterpretation ** Example With Image Weight #+begin_example [image URL] corgi character by [artist name] --iw 2 #+end_example ** Example Without Image Weight #+begin_example [image URL] corgi character by [artist name] #+end_example ** Practical Difference With =--iw 2=: - “make something close to this image, but in this style” Without image weight: - “use this image as loose inspiration, but interpret it more freely” ** When To Use High Image Weight Use high image weight when you care about: - pose - shape - layout - composition - specific object arrangement ** When To Avoid High Image Weight Avoid or lower image weight when you want: - more creative interpretation - stronger style transformation - more variation - less dependence on the original image ** Reference Image Checklist [0/6] - [ ] choose image close to desired structure - [ ] upload image - [ ] copy image address - [ ] paste image URL into prompt - [ ] add clear text instruction - [ ] decide whether to use image weight ** Key Insight Reference images are especially useful when words are not enough. They help Midjourney understand what you mean visually. ** Bottom Line Use reference images when you already have: - a pose you like - a composition you like - a style direction you like - a previous generation that is close but not perfect Then use text prompts and image weight to guide Midjourney toward the final result. * Transcript #+begin_example 00:00 Okay. In this video, we are gonna talk about how 00:03 you can use reference images 00:06 to help you 00:08 quickly generate the images 00:10 that you want. 00:12 So reference images allow MidJourney 00:15 to skip a lot of, 00:17 the process 00:18 of, 00:19 creating images that don't match what you want because it 00:22 can use a reference 00:23 to help better create, 00:25 what you are looking for. So let me show you 00:27 an example of this. So I've got 2 reference images 00:31 here that I'm gonna be using as part of this 00:32 tutorial video. 00:34 Let's start out with this flower. So, 00:38 if we come here remember how earlier 00:42 we generated a bunch of minimalist 00:45 flowers. 00:47 And, 00:48 these were fine, 00:49 but they weren't quite, you know, what these ones here 00:53 looked like. Well, I've actually, 00:56 prior to this already done a lot 00:58 of generations of this. And I was able to create 01:02 a flower which was much more closer through doing through 01:05 repeating this a lot of times because that is really 01:07 what you wanna use my journey to do. You wanna 01:09 create images a lot of times. For these tutorial videos, 01:12 I only do it a couple of times. 01:15 But really, I should be doing it a lot of 01:16 different times to get a lot of different takes. 01:19 And so, 01:20 I did that for this and I got something much 01:23 closer to what I wanted. 01:25 And so, 01:26 so I got I made this through Midjourney but let's 01:28 say for example that you found this on the Internet, 01:30 you could save this. 01:31 And then what we do 01:34 is we're gonna upload it 01:37 to MidJourney. 01:39 Come and drag it in. 01:41 Click enter. 01:45 Now it has been uploaded. 01:47 So what we can do is we can click on 01:49 it 01:51 and then, 01:52 we can right click it and we can copy image 01:55 address. 01:56 And now we can go imagine 02:00 and we can paste in 02:03 the code for it. 02:05 And then we can come in and type in what 02:07 we want. 02:08 So minimalist 02:11 black line art of 02:14 a sunflower 02:15 top down on a white 02:18 background. 02:22 So now what it's doing 02:24 is it's using this 02:26 image that I uploaded 02:28 as a reference 02:30 while also 02:32 applying 02:34 my prompt request. 02:36 So, 02:38 it's not just generating different versions of this. It is 02:41 taking, 02:43 the prompt text that I gave it, but it's using 02:45 this image as a reference guide. 02:49 And we're gonna see 02:51 why that is very helpful in a minute once it 02:56 has finished, 02:57 using it 02:58 and once it has generated for us the images. 03:03 And you'll see why this can save you 03:06 a lot of time. 03:19 Almost there. 03:22 Okay. 03:23 Here is why it's so helpful. 03:25 So 03:27 you look at this 03:29 because the reference image that I gave for it through 03:32 that I had generated by doing a lot of different 03:34 generations on the journey, but again, you know, if I 03:37 found this on the internet. So this here, 03:41 it's a sun you come here. 03:44 There are no there is no stem. 03:46 There are no leaves. There's just, you know, the petals. 03:50 I managed to generate something similar to that here with 03:53 Midjourney. 03:54 Now 03:55 and I saved it. 03:57 And so you come back here. When I did it 04:00 before, when I was generating these before, a lot of 04:02 these had, 04:03 you know, had the stems, it had the leaves. 04:06 And so we were essentially kind of, you could say, 04:09 wasting time by not wasting time because it's still good 04:11 to do it, but you have to generate lots of 04:13 images. But it was frustrating because a lot of the 04:16 images we were getting were not what we wanted because 04:18 even though it was on a top down, it was 04:19 including the leads and it was including the stem. But 04:22 by adding in a reference image that doesn't have those 04:26 for my journey 04:28 to use as a reference, then it didn't generate those 04:31 for me. And instead it generated 04:34 versions of it that matched, 04:37 what I wanted so much more closely. 04:40 And so that 04:41 is how you can use reference images 04:44 to save you a lot of time. 04:46 You can also use reference images in another very fun 04:50 way too. 04:52 So 04:53 let's come and take our next image. 04:55 So this is another image that I generated earlier using 04:59 Midjourney. 05:01 A cookie. What would you expect? 05:03 Anything less from me. Absolutely 05:06 love dogs. Adore cookies. 05:10 So we now have this cookie. 05:12 We can use this as a reference 05:15 to help us get 05:17 the 05:19 type of cookie that we want, but in a particular 05:21 style. 05:22 So I've uploaded this version of a Corgi and it's 05:24 in this particular pose. 05:26 Now remember earlier, 05:27 we generated Corgis, 05:29 using Elsa, 05:31 or 05:32 I don't know if I pronounced her name right. I'm 05:34 sorry if I didn't. 05:36 And we got 05:37 what a cute little 1 with a little hat. 05:40 We got them in different poses. 05:42 We didn't specify to be fair what pose we wanted 05:44 it in. So we just got a bunch of different 05:47 poses. Well, 05:49 what we're gonna do, 05:51 click this, right click it, 05:53 copy image address, 05:57 imagine 06:01 character 06:02 by 06:13 and 06:14 we're gonna add in something else. 06:18 Image, weight, 06:19 and we're gonna do 06:21 2. 06:25 So 06:27 what we have done 06:28 is we have asked 06:30 Midjourney 06:32 to, 06:34 draw us, 06:36 a picture 06:41 using this cookie as a reference guide, 06:45 And we've asked it to do it in 06:48 the style 06:50 of, 06:51 this incredible artist. 06:54 And we've also given it an image weight of 2, 06:57 which is the maximum that you can give it right 06:59 now on the journey version 5. It could change in 07:02 different versions. 07:03 But the image weight is 07:06 us saying that we want it to look as closer 07:09 to this 07:10 as possible. 07:11 If we'd given it a lower image weight of 1 07:13 you don't have to give it an image weight, by 07:15 the way, and we'll we'll do it again without giving 07:16 it an image weight. But by giving it an image 07:19 weight, 07:20 of 2, it means then that, 07:23 Midjourney is prioritizing 07:25 making it 07:26 as close to the reference image as possible. 07:29 And so this is what we've been able to achieve. 07:32 Now, let's 07:34 come 07:35 and we'll do this again. We'll copy this. 07:38 This, by the way, is just 07:40 MidJourney, 07:41 has 07:42 URL shortened it, 07:44 which is fine. We can use the URL shortened version 07:47 of it. 07:51 And we're not gonna add an image weight. 07:56 So it's using our reference image again. 08:03 And, 08:05 this time it's got free rein 08:07 to, 08:09 use this image, but create its own, 08:12 pictures 08:13 based upon 08:15 her art style. 08:17 And you'll see that we'll get, 08:19 different results again. 08:25 Loading. 08:35 And they're gonna be quite different 08:37 and I'll explain why they're quite different. 08:41 You can already see that they're very different. 08:47 And we're almost there. 08:55 Okay. So very, very different. 09:00 This is what we got by not giving it an 09:02 image weight, 09:03 And 09:04 this is what we got by doing it. 09:07 As you can see, this is much, much closer 09:10 to our original image than this 1 here. This 1 09:14 here 09:15 takes our original image but redoes it quite substantially. 09:20 And it adds a lot more of the artist 09:23 and how she would have interpreted it, 09:28 than, this 1 here, where this 1 here takes the 09:31 image and 09:32 makes a version of it that's inspired 09:35 by her, whereas this 1 here takes more of her 09:38 art and then uses the image. 09:41 But, 09:42 but rather than you know, that's what I mean. Like, 09:44 this is focused more on the artist herself. This 1 09:47 here is focused more on the image 09:49 itself. 09:51 And so, 09:53 if you want to just see what the artist themselves 09:57 would create, then you don't have to add in an 09:59 image weight and you can do this. 10:01 Or you can do this here and you can add 10:04 in the image weight and get a very different image. 10:08 As you can see, it can save you a lot 10:10 of time because you can imagine 10:12 that if you found a pose that you like 10:16 that by adding in a reference image, then you're gonna 10:18 save a lot of time 10:20 by skipping the generation process 10:23 of, 10:25 of having to, like, find 10:27 to get images that match that pose. 10:30 You can just tell MidJourney the pose that you already 10:34 like. 10:35 So, yes, 10:37 in this video series, we're also gonna discuss how to 10:39 save these images. That's not something that I've discussed, but 10:42 it's very, very important because once you've got your image, 10:45 you then need to save it. So, 10:48 that 10:50 is the next video that I'm going to create. Whether 10:52 that's the next video you watch in the series or 10:54 not, I'm not sure. But either way, I hope that 10:57 you enjoy it and I will see you on the 10:59 next video. #+end_example