Lesson 4. The Magic glass elevator. Reels breakdown.
Brief description of the process of creating this Reels:
1. Take a real photo in the lift at 0.5x according to my reference from the post.
2. Colour correction of the photo (Download Lightroom preset).
3. Replace the walls in the lift with transparent glass, replace the background (Prompt).
4. Prepare 7-10 different location backgrounds, 3:4 format (Prompt)
5. Generate a video of all locations with the lift moving downwards 3:4 (Prompt)
6. Remove the person from all frames and all locations in the video Kling O1 Edit, leaving only the transparent lift with the background (Prompt)
7. Upscale all video frames without the person Topaz
8. Cut out the person and place them on a green background, generate a video of the person's micro-movements on a green background separately and upscale the video
9. Edit Reels in Adobe After Effects, Roto Brush Tool 3.0, cut out the person and place them on top as an overlay, add frames of locations from the lift to the music
1. We take a photo in the lift at 0.5x according to my reference from the post.
We find a super beautiful large lift with good soft lighting and take a photo like this with a reflection in the mirror 👇 Iphone camera mode 0.5x, exposure -1.3 or -1.7
2. We make colour corrections to photos
It is important to edit the photo well so that it looks perfect. We still have some work to do on it. Check out the post on Boosty at the link - Buy Lightroom Preset #003
3. Replacement of elevator walls with transparent glass and replacement of background locations
We will be working with Nano Banana Pro. You can choose whichever aggregator you find more convenient: SYNTX AI or Higgsfield AI. In this case study, I did everything in Higgsfield AI.
Prompt to make the elevator walls transparent in the TXT file:
txt
Prompt - Glass wall elevator.txt1.80 Kb
Study the prompt in detail and make any changes you want. If you are a girl, change everything to ‘woman’. If you want, change the description of the background location in the prompt.
Since we will have different lifts, manual adjustments to the prompt may be necessary. But the essence remains the same. Generate as many times as you need until you get the perfect result. It took me about 15 generations and different variations of the prompt to get the desired result. Be prepared to do several generations for one action; do not use the first thing you get from neural networks. Neural networks are completely random, but it is a controlled randomness.
We carry out the entire production process in a ratio of 3:4 for photos and videos.
Once you have obtained the transparent lift with you, it will now serve as your primary base photo, which will be used to obtain other locations.
4. Preparation of 7-10 different location backgrounds, 3:4 format
We search the internet or generate interesting panoramic locations ourselves, such as a view from above.
Using one location as an example, I will show you how to replace the background in a glass lift with any other. Take a picture of the location and your photo with the transparent lift from the previous step, and upload both images to Nano Banana Pro. Always generate in 4K quality and a 3:4 aspect ratio.
Prompt Nano Banana Pro to replace the background behind the lift glass TXT file:
txt
Replacing the background behind the elevator glass_Promt.txt354 Byte
After generating the result, make sure that the parts of the lift and you in the frame have remained in the same place, nothing has shifted, you are standing in the same pose, and only the location behind the lift glass has changed — this is important for the next steps. Make as many attempts at generating as necessary to achieve the best result.
Repeat the same process for all locations to obtain 7-10 frames for further animation steps.
5. Generation of video frames of all locations, elevator moving downwards
I got 11 frames. I'll show you how to make the elevator go down with you using one example.
Load the first frame into Kling. In Higgsfield, this is the Start Frame window. Select Kling 2.5 Turbo, 1080p, 5 seconds, and always set the Kling generation mode to professional.
Prompt:
txt
Elevator movement down _ Kling Prompt.txt320 Byte
In the prompt, be sure to change the text describing your current location behind the lift glass. Describe the location in simple language. Then start generating the video.
Just like with photos, make sure that when the video is generated, the parts of the lift and you remain in the same place in the frame, nothing has shifted, the lift does not sway from side to side in the video, you are standing in the same position (a slight swaying of the person is acceptable), but the lift cabin itself must remain firmly in the frame — this is important for the next steps. Make as many attempts at generating the video as necessary to achieve the best result.
Now we have all the video frames where you are eating in the lift going down against the backdrop of different locations. Let's move on to the next step.
6. Remove the person from all frames and locations in the Kling O1 Edit video, leaving only the transparent lift with the background.
Now our task is to remove the person from all the video frames we have obtained. You may ask, why can't we just remove the person in the lift from the image in Nano Banana Pro and immediately generate the lift's downward movement without the person? - To avoid neural network errors. The fact is that if we animate a transparent lift without a person, the neural network gets very confused and produces a lot of errors in the generation results.
The task is to do this without any jumps between location switches while the lift is moving down. Removing the person from the transparent lift will be done using the new Kling O1 Edit model on the same Higgsfield platform.
I will show you how to remove a person from a video using one frame as an example. Select the Kling o1 Edit neural network. Upload your video with the first location where you are eating downwards. In the first window shown on the screen (Higgsfield platform), describe briefly what needs to be done. You don't need to upload anything else.
Before uploading the video, write the first sentence in the prompt field - ‘Remove the man from the video’ and upload the video. You will see that after your sentence, a variable with your video appears in the prompt window. After the variable, add the following text to the prompt window briefly describing your location: ‘In the background behind the elevator glass, a large-scale panoramic view of a villa on a cliff, golden sunset.’
Do this with all your videos. Remove the person from the video in all locations. Move on to the next steps.
7. Upscale all video frames without people Topaz
In my case, I ended up with 11 videos where a glass lift descends without a person in the frame against the backdrop of different locations. And these are still videos in 3:4 format. We upload all these frames to Topaz and set the quality to 4K / 60 frames.
In Higgsfield AI, this is located along the path:
That's it, we have 11 videos in 4K quality, aspect ratio 3:4.
Let's move on to the next step.
8. Cut out the person from the photo and place them on a green background, generate a video with micro-movement of the person on the green background separately, and upsample the video to 4K.
We've reached the main character in the frame — it's a person. To make your final video look professional, the person will be a separate layer in the final Reels video.
Cut the person out of the photo processed in Lightroom, making sure to do it manually and with high quality along the outline on your phone using the Magic Eraser app.
Go to Higgsfield AI > Edit > Relight.
We need to change the lighting of the person, namely add a control light - just like in real life. Since the lift has become transparent and there is additional light from the locations, the person should become slightly darker in the light and the outline should be slightly illuminated. Set the Relight tool settings as shown in my screenshot. And upload the photo on a green background.
The result is that we cut out the person again and place them on a green background with a blue circle behind the person in the head area. The blue circle will be a reference point for the person's gaze for video generation. You can set a green background in any editor. I use the Figma or Adobe Photoshop programmes on my PC. On your phone, you can use the Picsart app.
We create an animation of a person moving against a green background. We upload the photo to Kling 2.1; version 2.1 worked well for me in this case. And we insert the prompt.
txt
A person moving against a green background _ Kling Promt.txt531 Byte
Upload the video of the person on a green background to Topaz AI and improve the quality to 4K / 60 frames. Next, when editing reels in Adobe After Effects, you will need to cut out the person in the video on a green background using the Roto Brush 3.0 tool.
____
Next
Generate the initial frame as a person enters the lift. To do this, load your processed original photo in the lift into Nano Banana Pro and remove the person from the frame, then write the following prompt:
txt
Remove a person from the elevator_Prompt.txt222 Byte
Now we have two frames with an empty lift and an open door, and the original frame with you in the lift. We load these two frames into Kling 2.5 Turbo as the start and end frames. And we get a video of a person entering the lift.
Prompt:
txt
A man enters an elevator_ Promt Kling.txt122 Byte
We also upscale the video in Topaz - 4K / 60 frames per second.
From the resulting video, we take the last 0.5-0.7 seconds. And we use it at the very beginning of Reels during editing.
9. Edit Reels in Adobe After Effects, use the Roto Brush Tool to cut out the person and place them on top as an overlay, add frames of locations from the elevator to the music.
We edit in any convenient programme in 1080x1920 format, 30 fps, and glue all the frames together. I made each frame 0.7 seconds long, but you can decide for yourself so that the transition matches the beat of the music.
I recommend downloading Adobe After Effects to your computer and learning how to edit in it. This lesson is not about how to edit in Adobe After Effects, but there are tutorial videos for beginners on YouTube. It has the perfect tool for cutting out objects and people in photos. Learn in detail on YouTube how to cut out a person on a green background using the Roto Brush Tool 3.0 in Adobe After Effects.
The process is as follows: we upload all your videos: the starting video in the original lift, with a transparent background and no people in the frame, into any editing programme. We upload the video of the person on a green background and place it on top of all the frames using the Roto Brush 3.0 tool in Adobe After Effects to cut the person out of the green background.
Slightly speed up the video frames where the lift is going down against the backdrop of different locations to make the speed of movement realistic.
If you decide to edit in CapCut on your phone, which I strongly advise against, you will not be able to remove the green background from the person properly, and there will be many errors around the edges. But if you still decide to use CapCut, after applying the Chroma Key tool, you will be left with green edges, and you will need to remove the saturation of the green and blue colours completely.
Done!
ai
reels