Your First Afternoon With Video-to-Video: A Step-by-Step Walkthrough for Beginners

If you have never tried motion transfer, the idea can sound more complicated than it is. This walkthrough takes you through a first session in a sensible order, with a small budget and no pressure. It is written around Genjutsu AI, which does motion transfer only: one source video supplies the movement and camera, and up to 20 reference images supply the look.

You do not need any special equipment. A phone, a few pictures, and about an afternoon are enough.

Step 1: Understand the two inputs

Before you touch anything, fix this in your mind:

The source clip decides how things move and what the camera does.
The reference images decide how things look.

Whenever you are unsure what to change, ask which of the two is responsible. Movement wrong? Change the clip. Look wrong? Change the pictures.

Step 2: Film a simple source clip

The source clip must be between 4 and 30 seconds long and in MP4 or MOV format. Phones usually record MP4 or MOV already, so this is easy.

For a first clip, keep it simple:

One person or one object doing one clear thing: a wave, a slow turn, a short walk.
A plain, uncluttered background.
A steady camera, unless you specifically want the camera to move.
Between 5 and 10 seconds. You can always go longer later.

Make sure the person in the clip is happy to be in it. Remember that videos made before a first purchase are public in Community, so film something you would be happy to share.

Step 3: Gather a few reference images

Choose pictures that show the look you want: a drawn character, a costume, a toy, a style. You can use up to 20, but for a first try, five or six is plenty. Make them consistent: same subject, a few angles, similar style and lighting.

Only use pictures you have the right to use, such as your own drawings or photographs. If in doubt, leave it out.

Step 4: Decide your resolution

There are three resolutions, and the cost is per second of output:

480p: 3 credits per second.
720p: 7 credits per second.
1080p: 11 credits per second.

For a first run, choose 480p. A 6-second draft costs 18 credits. It is the cheapest way to see what you will get.

Step 5: Run a draft and look at it twice

Submit your source clip and reference images and wait for the result. Then watch it twice.

On the first watch, look only at the movement. Does it do what the source clip did? Is the camera behaviour similar?

On the second watch, look only at the look. Does the character, costume, or style resemble your references?

Separating the two views makes it much easier to decide what to change.

Step 6: Change one thing

If something is not right, change only one input and run another draft:

If the movement was off, choose or film a different source clip.
If the look was off, add a picture that shows the missing detail, or remove the picture that seems to cause trouble.

Changing both at once makes it impossible to tell what helped.

If a generation fails, it is automatically refunded, so you do not need to worry about losing credits to a technical problem.

Step 7: Make a final

When both the movement and the look are right, run the same inputs at a higher resolution. A 6-second clip at 720p costs 42 credits and at 1080p costs 66. Because you have already tested the inputs at 480p, you are spending these credits with confidence.

Step 8: Keep notes

Write down what you used: the source clip, the reference set, the resolution, and what you learned. A few lines is enough. Next time you will be glad to have them.

How credits work

Credits are sold as one-time packs. There is no subscription, and credits never expire. So if you buy a pack and then do not use it for a while, nothing is lost. You can work in short bursts whenever it suits you. If you want to see the packs, they are at https://genjutsu.app/pricing.

What about code?

If you later want to process many clips at once, Genjutsu AI offers a REST API. For a first afternoon, you will not need it.

A realistic expectation

Your first result may not be perfect. That is normal. The goal of the first session is to learn how the two inputs behave, not to produce a masterpiece. By the end of the afternoon, you should know what your source clip contributed, what your reference images contributed, and what you would change next time.

When you are ready to try, the tool is at https://genjutsu.app.

— Genjutsu AI