Reuse a training video with your brand mascot: three routes and what each costs

Most small teams end up with one training video that does its job and looks dated. Re-shooting means a day of someone's time, so the idea of keeping the old footage and swapping in a brand character is attractive. One tool that does this kind of swap is Higgsfield, a web platform that puts many AI image and video models in one subscription. It is not the only route, and for many training videos it is not the best one. This page puts the options in order of effort and cost, then shows how to test and approve the result.

This page may contain affiliate links from which Blur Studio earns commissions.

Is your video a performance or a script?

This one question decides most of the plan.

• A script video is someone talking to camera or over slides. The value is the words. You can rebuild it with a new voice and face from the same text.
• A performance video shows a person doing something: a hand gesture, a walk through a store, a way of holding a tool. The value is the movement and timing.

Motion transfer copies movement. If your video is mostly words, you are paying for a feature you do not need. If the movement is the lesson, it may be worth a test.

There is also a third case: the brand has no character yet. Motion transfer needs a consistent set of images of the character first (front, side, a few expressions). If you do not have those, your real project is character design, which is a separate cost. Settle that before you look at any video tool.

The cheapest option first: re-film it on a phone

For many teams the cheapest honest answer is to record it again. A phone on a stand, a script you already have, a branded title card and a lower-third (the name bar at the bottom of the screen) can match your brand in an afternoon. You keep real people, real hands and real products, and there is nothing to disclose or explain.

Re-filming is also the right answer for anything where accuracy matters: safety steps, medical procedures, equipment handling, regulated compliance training. An AI rebuilt hand position or glove can look fine and be wrong. If a wrong detail could hurt someone, film the real procedure.

Cost: your time and the time of the person on camera. No credits, no watermark, no review of AI artifacts.

Script-led training: an AI presenter

If the video is mainly words, an AI presenter tool is the next step up. HeyGen is a tool that turns a written script into a video of a talking avatar, with your choice of face and voice. From HeyGen's pricing page, read on 3 November 2026 (check the current page before you buy):

HeyGen plan	Price	What it includes
Free	$0	3 watermarked videos a month, up to 1 minute each
Creator	$29 a month, or $24 a month billed yearly	600 credits, and a Custom Video Avatar with your own voice and likeness
Pro	$49 a month, or $41 a month billed yearly	1,000 credits
Business	$149 a month plus $20 a seat a month (per HeyGen's pricing page)	For teams

An avatar tool is better than motion transfer when the training is script-led. You edit the text, regenerate, and the lesson changes with it. You do not need a source performance at all. SHRM, a US human resources body, reported in 2022 on one company that used avatars as moderators in an e-learning course to avoid hiring actors or paying for a studio day. If you want a brand mascot rather than a human avatar, check what the tool allows before you commit, because avatar libraries are built around human faces.

Movement-led training: motion transfer with your own character

Motion transfer takes a reference video and a picture of your character, and produces a new clip where your character moves the way the person did. Two places to get it:

• Higgsfield Genjutsu. According to Higgsfield's help center (updated October 2026), you give it a reference clip of 4 to 30 seconds and up to 30 images. It has two modes. Motion transfer rebuilds the scene around your character. Objects swap changes one element and tries to keep the rest of the shot. You choose 480p, 720p or 1080p. A cheaper version, Genjutsu Lite, takes up to 10 images, has no mode choice, and offers 720p or 1080p.
• Kling directly. Kling is an AI video model that Higgsfield and other platforms build on. Its own Motion Control guide says the reference must be one continuous shot of 3 to 30 seconds, and that fast or complex motion may give a shorter clip with no refund of credits. If your team already pays Kling or a platform that hosts it, that is a fair route.

Higgsfield's value here is having Genjutsu inside one subscription with other tools, plus the object swap and a style block. Per Higgsfield, Genjutsu on the website uses no credits only when the model sits on your plan's Unlimited list (the models your plan runs without a credit charge, on the website only). Use through the API, MCP (a connector that lets an AI assistant run tools), the command line or plugins always uses credits. Higgsfield does not publish a credit price for Genjutsu, so the cost shows on the Generate button before you confirm. For plan prices see the pricing explainer, and check the pricing page for today's figures.

Be realistic about results. The one independent write-up I know of, from the AI newsletter The Rundown, describes an object swap that did not remove the original golf ball and did not place the replacement where the ball was going, and says its sports-character attempts were blocked as protected content. Treat the vendor's examples as best cases. Read more in my guide to Motion Control and Genjutsu.

One more point on accounts. Higgsfield says every generation on a free account carries a visible watermark, and paid plans do not. A free test is fine for judging the look. It is not something to hand to a manager as the finished training.

Build the character kit before any video

Studios that make branded training characters and an AI video vendor (Hailuo AI, which sells video generation) agree on one thing: treat the character as an asset you reuse across many videos, not as a one-off picture. A kit that works across tools has:

• One approved master image: a neutral pose and expression on a plain background, high resolution, signed off by the brand owner. Make every later clip from this image so the character does not drift.
• A few extra views and expressions: front, side, and two or three moods.
• Rules for movement and voice: how the character gestures, how it sounds, and what it never does.
• Brand usage rules and ownership: who owns the design, where it may appear, and a written copy of any freelancer's assignment of rights.
• A simple register: a sheet listing each clip, its source file, who approved it and when.

If a design studio builds the character, ask for the expression set and the rights in the contract. A studio that builds characters for a whole course can hold the look steadier than repeated AI generations, so it can be the right choice for a large library.

What the source clip must look like

Motion transfer is fussy about the input. Before you spend anything, check your original against this list, drawn from the Kling and Higgsfield guides:

• One clear person, full body or the part you need, with limbs and face visible.
• A steady camera and no cuts. A clip with a cut may be shortened.
• Moderate speed. Fast moves are where errors appear.
• Length inside the tool's range (4 to 30 seconds on Genjutsu, 3 to 30 seconds on Kling).
• Your character image framed the way the source person is framed. A face-only picture does not carry clothing, hair or props in Kling's face binding, and works best when the character faces the same way as the person in the video.

If your training video is 12 minutes with cuts, do not feed it in whole. Break it into beats of 5 to 10 seconds, generate each beat, then stitch them in an editor and keep the original audio track. That also makes a failed attempt cheap.

Steps to reuse a video with a brand character

1. Choose the video to reuse. Pick one that is movement-led and short, not your most important course.
2. Check you own it. You need the right to change the footage and the written consent of every person in it (see the consent section below).
3. Prepare your brand character. Front, side and one or two expressions, on a plain background.
4. Make a test version. Cut one 5 second beat. Use the lowest resolution and open Genjutsu in your Higgsfield account or your chosen tool. Pick the mode on purpose: motion transfer if the room can change, objects swap if the room must stay.
5. Get a manager to approve it. Someone who knows the subject checks that nothing wrong is shown, before you make the rest.

A test ladder with pass and fail marks

Use the same character image for all three steps. This is a plan you can run, with cut-offs I suggest, not a result.

Step	Clip	Score each item pass, fixable or fail
1	A 4 to 6 second hand gesture	Timing matches, hands have five fingers, no extra objects
2	A body turn	Face stays the same character through the turn, clothing does not change
3	A full-body sequence	Feet touch the ground, background is acceptable, motion matches the lesson

Pass: all items are pass or fixable on steps 1 and 2 within three attempts each. Fail: any "fail" on hands or timing after three attempts, or any wrong demonstration shown. Stop there. You have learned for the price of a few short clips that this video should be re-filmed.

Cost per usable minute: the formula

Do not judge cost by the price of one generation. Judge it by the cost of one clip you can actually use.

Cost per usable clip = (seconds per attempt x credits per second) / success rate.

Here is a made-up example using Kling's own published rate, so the numbers are not Higgsfield's. Kling's guide lists Motion Control on its 3.0 model at 12 credits per second for Professional quality. A 20 second clip is 240 credits an attempt. If only 1 attempt in 3 is usable, you pay for about 720 credits to get one good 20 seconds. A 10 minute course in 20 second beats is 30 beats, so roughly 21,600 credits on those assumptions. Run the formula with your own success rate from the test ladder and the cost shown on the Generate button, then compare it with an afternoon of re-filming. For many courses, filming wins.

Rights, consent and review for training content

The person in your old video is a real person whose movement and image you are about to reuse. Treat this as a consent job first. This is information, not legal advice: check local law and each platform's current rules.

• You own the source video, or hold a licence that allows changing it.
• The person in it agreed in writing to reuse in a new form, and you told them their movement will be applied to another character.
• You decided whether their voice is also reused, and got consent for that too.
• You keep the signed consent with the project file.
• Your character is your own design. Higgsfield's help center says its filters block protected content such as franchise characters and logos, and its Genjutsu help article does not mention consent, so the consent step here is a general caution rather than a Higgsfield rule.
• Higgsfield says outputs are not guaranteed to be unique. Do not describe the result to anyone as one of a kind.
• A subject expert reviews every beat for wrong or unsafe actions.

Also decide up front how you will tell staff the video uses an AI character. Telling people is simple and avoids doubt later. Read what to check before you pay Higgsfield, including how it may use uploaded content to improve its models unless you are on an Enterprise agreement.

Who should skip this and film it instead

• Anyone making safety, medical, equipment or compliance training.
• Teams whose video is mainly a script. Use an AI presenter or re-record.
• Courses with dozens of videos that must all show the same character on model. A freelance animator can hold a character consistent across a whole course better than repeated generations.
• Anyone who cannot get written consent from the people in the original footage.
• People who only want a free test for a client deliverable. A free account adds a watermark.

What to ask before paying

• Is the video a performance or a script?
• Can the tool take my clip length, and what happens if it shortens the result?
• What does one attempt cost on the Generate button, and what is my success rate after the test ladder?
• Does the plan I need remove the watermark?
• Who signs off on accuracy, and who holds the consent forms?

How I'd approach this

I would sort the library first: scripts go to re-recording or an AI presenter, safety content goes back in front of a real camera, and only the movement-led clips get a motion transfer test. I would run the test ladder on one 5 second beat, cap attempts at three, and let the success rate and the subject expert decide, not the demo reel. If the numbers work and you want to try it, you can start with a Higgsfield account and keep the first month small.

If you reuse video often, the weak point is rarely the model. It is the folder of consent forms, character images and approved beats that nobody can find. Blur Studio builds the repeatable workflow, approval portal or shared asset library around tools like these. Compare it with the cost of an AI video tool against hiring an editor before you decide.