Transcription
Welcome to another MidJourney video tutorial. Today I'll show you how I create looping videos and explain how the starting and ending frame can be useful.
If you check the MidJourney explore page, you'll see there are already a lot of examples you can get inspiration from. Let's go to the create page. If we click this button, we can add images and then you have this upload option that lets you select an image. It's just a simple color background. As you can see, it added that image as the start frame. Then I add a second image which is the logo and that gets automatically added to the ending frame. You can click the X to remove a frame and you can drag other images you uploaded if you want.
Now keep in mind it works best if the frames are somewhat similar. If there's a big difference, it will be hard for the AI to make a smooth transition. You can also enable a loop here and you'll see that infinity symbol, but I'll show you later how to use that. Let me deactivate the loop and use the logo as the end frame so we can create a logo reveal. For the prompt, I'll add something like a logo appears from purple smoke. Then we can run this. In around two minutes, we get four different results. As you can see, because I didn't mention what kind of smoke, some versions look more 2D and others more realistic. So, it's important to be more descriptive if you want something accurate and to use general terms and shorter prompts if you want more variety.
Let's go to the add images button again. Since I already have those uploaded, I'll just drag the first image to the starting frame and the second image to the ending frame. Now, I want the smoke to look a bit more realistic, so I'll write realistic purple smoke in the prompt. You can also use ChatGPT to improve your prompts if you want. Let's test it. Now, we've got a more interesting logo reveal with smoke that looks closer to what I had in mind. So, basically, you can start from nothing and reveal your logo. You can try using fire or smoke or maybe water. Be creative and see what works and what doesn't. Try different seeds and different prompts. If the result is close to what you want, but you'd like to try again, use this rerun button. In about 2 minutes, you'll get a new set of videos like I did here. You never know when you'll get lucky and get exactly what you're looking for. This one looks really cool.
So, let's also test the loop option to see how it works. We add one image as the starting frame, and for the second frame, we don't add anything. We just enable the loop from here. Then, we need a prompt. Maybe something like the robot looks curious at the camera. And then, we run it. I got a few different versions and in all of them the robot is moving. It didn't have eyes, so it added some cartoon eyes in a few versions, but it came out pretty interesting. You get decent results with graphics. For some, you can't really tell where the video ends and starts, and for others, especially with photos, there can be a tiny glitch, just a fraction of a second. And if you look closely, you can spot it.
Let's test it with a more realistic image. Maybe I'll do a portrait of a sheep. By the way, I'll add the prompts and resources on Discord for free. If you want to get them, check the video description. For the settings, maybe I'll go with a portrait ratio. Then, let's generate. Since I didn't have any image uploaded, it will generate one for me. I got four different sheep. And I kind of like this one. Now that the image is generated, I can animate it easily. And now we have the loop version here. If you don't want it to animate automatically, you can use the manual animation option. But for this example, I'll use low motion loop. Let's see what loop videos we get from that. You can see in the thumbnail that I used the loop option because of the infinity symbol. They came out pretty good. In one version, there's something like a flying bumblebee. The loop is almost perfect. You can spot it if you look carefully, but this is the first AI that actually does loops, so I imagine it will get even better with future updates. I kind of like this version.
Let's do some more experiments with starting and ending frames. First, I'll generate an image for the starting frame. Let's try a portrait of a warrior man. Looks interesting. Now, if I click here, I can reuse the same prompt, maybe tweak it a bit, and make the man wear bare fur. Let's test that. Okay, this one looks interesting. I think I can work with it. So, I'll go ahead and save it. Next, I want another image of a bear in a similar style. I'll click here to use the style from the previous image. You can see it was added as the style reference. Now, I can write the prompt for what I want in that style. I'll go with a warrior bear portrait with red eyes. Let's generate. I got four results. Out of those, I think this bear looks more similar to the man image. So maybe I can get a better transition between the two frames. I'll save this bear image as well.
I want the warrior to transform into a bear. So I go to add images and click upload. Then I upload the man for the starting frame. After that, I go to upload again and upload the bear for the ending frame. Now that we have both frames, all we need is a prompt. Let's see if it works with a simple one, like the man is transforming into a warrior bear. I got the first results, but the transition is not how I imagined. It's too sudden, like it jumps from one image to the other. If I look closely, it feels more like a fade from man to bear. So, we need a better prompt for that. I'm not that good at writing long prompts, so I'll use ChatGPT to help me. I'll go and add both of those images into ChatGPT so it understands better what I'm trying to do. Then I'll explain what I want to create and maybe it can give me a better prompt. I'm not sure what we'll get, but usually it's pretty good at prompts. It also gave me prompts for the first and last frames, which I didn't really need, but included a transition prompt. That's the part I needed. You can see how it described the transition more visually, like how the body is changing. So, I'll copy that transition prompt and see if it works. I'll just drag the first and last images in here really quick, then paste the long prompt. And let's see if MidJourney understands better what I want. Okay, the result is better now. Still for transformations like this, probably other AI tools might do it better. Let's check a few of these transformations. Maybe one will come out better than the others. It's an improvement compared to before when it was just a simple fade. Now it looks more like a transformer style transformation. So with better prompts, you can still get some decent results.
I have another idea I want to try. This time, let's do a landscape image. I want a cartoon 3D illustration of an old man and a woman sitting on a bench with some distance between them. You'll see why in a minute. Everything looks quite nice. I think I'll go with this one. It reminds me of the movie Up. So, I'm downloading this image. Then, I'll go to the edit tab. I usually prefer to use Photoshop for this, but since the option is here, why not give it a try? I click to upload an image and use the illustration I just made. Then I adjust the brush. And with the eraser brush, I make a selection a bit larger than the woman since I want to remove her from the scene. When I try to submit the edit, it asks for a prompt. I'll just write what I want to be there, a bench in this case, and submit the edit. In a minute, I get four different variations to choose from. I like this version, so I'll download that image.
Now go back to the create menu and add the images. For the first frame, I'll use the one where the man is sitting alone on the bench. For the ending frame, I'll add the one with both of them on the bench. For the prompt, I'll say the woman enters the scene and sits on the bench next to the old man. Let's generate. Here are the results. Some of them have artifacts like flying bricks or something, but this one looks pretty good until the last second of the video. Not sure what's happening there. Maybe something in my prompt triggered it, or it's just an issue with the AI. Let's rerun it and see if we get a better version. Again, all of them seem to have some defects, like random objects flying around. Maybe the AI is trying to animate something in the scene and it doesn't work well. This one looks pretty good until the end when it seems like an alien teleports her into another dimension. Still, it has potential. Even if it's not perfect, it gives you more control than what you'd normally get with just a single frame. This one also looks pretty interesting until the end part.
Let me show you something interesting. I have here a photo of a man that I opened in Photoshop. It could be a photo of you or a friend. Then I want to introduce something into the scene for the end frame. If I just use a prompt for it, it will be random each time, but I want control over it. And that is why I use an end frame. For example, let's say I want a red parrot to fly in and land on the man's shoulder. So what should the last frame look like? The man with a parrot on his shoulder, right? I'm using a selection tool to make a big selection over the shoulder area. Then with generative fill or in many AI tools, this is called inpainting. You can prompt for what you want to be there. I'll type the prompt and click generate. Now I have three variations to choose from. You can generate again if you want more, but I'll go with this version. I'm saving that image as a JPG file.
Now back in MidJourney, I upload the image of the man alone as the starting frame and the image of the man with the parrot as the ending frame. For the prompt, I say a red parrot flies in and lands on the man's shoulder. Let's test it. The result looks better than I expected. There's still some random motion in the background, so the AI seems to have a problem with that. But aside from that, I like how the parrot flies in and lands on the shoulder. It looks pretty cool. Try with different animals, dragons, or maybe introduce elements or change the outfit or something. Be creative.
Let's try a few more videos with the loop function. I'll use a portrait ratio with this long prompt to generate a wizard hand with magic fire. Then I'll animate it using low motion loop. The result is pretty interesting and the looping part looks quite smooth. You need to think about images that can be looped naturally since some images might be harder to animate like ones with strange angles or something like a car chase which is difficult to make look natural in a loop. Here I generated an image of a zen bamboo water fountain. This image looks quite nice so I'll use low motion loop to animate it. Some of the results have strange water drops and don't look continuous. In some of them the water flow stops and then starts again when it's on automatic. In one version, the water drops look like they defy gravity and just fly around. So, I'll go back to the original image and instead of low motion loop, I'll choose animate manually. Then I can enable loop where it says end frame and adjust the prompt to try something like continuous water pouring. Maybe that helps. Some of the results seem more continuous, but a few still have water drops flying around. So, the AI still needs to learn the laws of physics, but I think this last version works, though. It has that continuous look and the looping part is pretty believable. Could be nice for something like an Instagram post. Definitely draws more attention than a single image. Also, you can use the heart icon to add it to favorites so it's easier to find when you have a lot of generations.
Let's try a more fantasy scene like a man statue with a waterfall coming from the statue's mouth. I like this version, so I'll try to get some variations of that image using variation strong. Now, I got four different variations that have potential. So maybe I'll go with this version. I'll click low motion loop since it's faster. And if that doesn't work, I'll try manual animation. In this case, everything seems to work fine. I'll go with this version since the loop looks continuous and has a nice motion. The cool part is that if you generate multiple loop variations from the same image with different prompts, you can combine all those videos to create a longer video that does different things, like I did here with this cartoon.
That's all for today. If you found something useful, leave a like and a comment to help with the algorithm. Thank you, AI Titans. And to everyone who subscribed to the membership and supports this channel so I can create more tutorials for you. Have a nice day and I'll see you on Discord.