Midjourney InPainting - a new feature in the AI graphics generator (video, 6 minutes)
Olivio Sarikas shares intriguing insights about the new inpainting function in Midjourney. This new tool allows users to edit images directly within Discord, significantly enhancing the creative potential for artists and content creators. In his video, Olivio showcases various uses of this feature, including placing a new head onto the image of an older man. He employs the lasso and rectangle selection tools to achieve satisfactory results, though he sometimes comments that the new head or background may feel slightly too high or disproportionate. As Olivio experiments with backgrounds, we see how beautiful the final results can be, even when making changes to large areas of an image.
When exploring different images of astronauts, Olivio notes that while the results are aesthetically pleasing, they don't always match the descriptions provided in the prompts. Interestingly, not only does the appearance of the individual change, but so does their gender, which highlights the challenges in achieving consistent results. Olivio also points out the difficulties in attaining uniform images across different facial expressions, a concern for creators involved in game development and visual storytelling. In some cases, the beautiful lighting and details are sufficient to satisfy artists, even if the images are not exactly what they expected.
The new 'vary region' feature in Midjourney allows for more advanced shaping and editing of created images. After generating an image, users can select the appropriate selection tools to easily edit chosen fragments. Olivio encourages selecting larger areas to achieve better results rather than just small parts. Creating more coherent and visually appealing works requires practice, but the new tool offers a myriad of opportunities.
Despite this, Olivio notes certain limitations, such as the inability to make inverted selections, which can be frustrating in some scenarios. He encourages testing to get the best outcomes while working with the system. Olivio concludes that this new feature opens up powerful potential for anyone looking to experiment and refine their artistic skills in a more interactive manner.
At the time of writing, Olivio Sarikas's video on this topic has already garnered 20,915 views and 521 likes, demonstrating the keen interest in this new approach to editing images within Midjourney. He encourages viewers to engage and leave a like if they enjoyed the material, which is a great way to foster a creative community. Be sure to check out the video for insights into this fascinating new feature!
Toggle timeline summary
-
Introduction to Midtourney's new inpainting function.
-
Demonstrating head replacement on an older man.
-
Using a lasso tool for the selection around the head.
-
Showing another example with a rectangle selection tool.
-
Discussion on head size in the examples.
-
Background replacement demonstration.
-
Results are generally satisfactory despite minor issues.
-
Introducing an astronaut example and its visuals.
-
Observations on face consistency and gender changes.
-
Highlighting a female astronaut image with cinematic quality.
-
Results didn't align with prompts; unexpected outcomes.
-
Navigating the function within Discord for image generation.
-
Explaining selection tools available for modification.
-
Recommended usage for larger areas instead of small selections.
-
Limitations on inverted selections discussed.
-
Concluding thoughts on the new tool's capabilities.
-
Encouragement for creative experimentation with the tool.
Transcription
Midtourney has a new function for inpainting and in this video I'm going to give you a quick intro on how it's working and how it's supposed to be used. Let's get started. I want to show you several examples where I try to put a head on the head of the older man. And that worked pretty well. I selected this with the lasso tool with a close area around his head that is roughly the size of the head you're seeing in these images. Personally I would say that the head looks a bit small but also a bit high up on the head of the man. Here's a different example where I selected it with the rectangle tool with a bigger part of the selection. You can see the part of the background changing that I have selected. So it really replaces everything you are selecting. Now surprisingly the head is fitting very well in size compared to the head size. But still I have the feeling that in this example at least the head is sitting a little bit high for my taste. This one is a good one but in a lot of the other images it feels like it should be a little bit lower. Now here you can also see examples where I replaced the complete background. So I wanted to have a city in the background. I feel like the results are really beautiful. The haircut is changing a little bit because I selected a little bit of the hair also to have the person really fused with the background. It is a little bit surprising that on the left side the tree is changing so much. And in this case we even have a little bit of text in here. But for most of the results I'm very happy with what I got from that. Here I have another example where I have a male astronaut inside of his suit with the helmet over the head. And it looks really beautiful. I really love everything about this. And now here we have the changes of the smiling astronaut. Now first of all it is interesting to see that the astronaut does not look like the original astronaut. So this is not about the consistency of the face. And it's also interesting to see that this also changes the gender of the astronaut. So even though the results are beautiful this doesn't seem to be a tool where you have a consistent face with different face expressions. And I know that a lot of you would like that for example for different stories or video game assets and things like that. Here I have another example for you. Very nice image of a female astronaut. Beautiful cinematic light in here. The prompt for this image is simply beautiful female astronaut. Now here I selected the face and at the end of the prompt I added bright face smiling. And it gave me some horrible images. The images look consistent from the lighting. They look very nice. But the image description does include beautiful female astronaut bright face smiling. So I'm not quite sure why these results came out like that. Here we even have a male astronaut who is rather screaming than smiling. And here we have another male astronaut who is smiling. The results in all cases are very beautiful. It's just not exactly what I asked for. Let's talk about where you can find that function. When you generate a new image inside of Mid Journey in Discord of course, you want to click on the U button for upscaling to get this image individually. Once this has been generated you can see down here one button that says vary region. When you click on this it will open up an extra window for you. And then on the lower left side you have two buttons here. One is a rectangle selection tool and the other one is a lasso selection tool. Of course with the rectangle selection tool you can draw out a rectangle over the area you want to change. And when you right click on that area it is going to be removed again. And with the lasso tool of course you can select areas in the image you want to change. And again with right click you can remove that. Then below that you have an area where you can enter text in here. I did not really find a difference between only describing what you want to change and adding what you want to change to the prompt. But of course if you have a very long prompt you might reduce it only to what you want to change so that the AI exactly knows what kind of thing you want to have changed in there. Now there is something that is pointed out by the Mid Journey team. This is supposed to be used with 20% of the image or even 50% of the image. So larger areas not just small areas. So for example just selecting the hand here can work but it is a very small area. But on the other hand what you can do is for example select a bigger area. For example the complete background here and replace it with something else and get a very nice result from that. Now one thing as far as I know that you can't do at the moment is to do inverted selections. For example you could not just select the man and then have replaced everything but the man. That doesn't work. Also the Mid Journey team says that the changes should be consistent with the image. So for example putting a dolphin sitting next to the man on the bench is probably not going to give you good results. But overall this is the introduction of a new tool. I'm really amazed to see that. I'm really mind-blown how they integrated that into Discord and the results seem to be fantastic. So this opens up a huge new dimension of experimentation with Mid Journey results and refining your artistic output. Leave a like if you enjoyed this episode. Thanks for watching and see you soon. Bye. Oh you're still here. So this is the end screen. There's other stuff you can watch like this or that. Really cool. And yeah I hope I see you soon. Leave a like if you haven't yet. Yeah.