Skip to content

Learning AI · From my channel

From image to video: testing Midjourney, Runway and HeyGen

In this class, I show visual-content experiments with images, animated characters and avatars. The interesting part is selection: some variants resemble what we want, while others need changes before moving to the next tool.

By Eric Muriel3 min read

English article adapted with AI assistance from the transcript of my Spanish video. About this adaptation · Explore my channel

Eric Muriel · Own content · original Spanish video · Published on August 27, 2025. YouTube loads when you press play.

01Decide which piece you want to create

The session distinguishes generating an image, animating a character and preparing an avatar presentation. These tasks are related but produce different results. Before starting, define the piece's role in your video or publication.

A short asset accompanying an explanation needs a clear visual idea. You do not need a short film to check whether the asset works. Define the framing, approximate duration and message it should reinforce.

Reference [1]: Eric Muriel · YouTube

02The reference affects the likeness

During the image test, we compare several character versions. Some retain the features better and others drift from the intended result. We also discuss how a face that is too small in the reference can make the test harder.

Use your own or authorised materials, selecting references that clearly show what matters. Keep alternatives and compare the face, proportions and consistency with your intention. A striking image is not necessarily the best one to animate.

03Animation can introduce defects too

In the demonstration, we animate an image using a performance reference. Reviewing the gestures reveals a problem: overly strong transfer can distort the image. We therefore test variants and observe the differences.

Review the entire clip, not only its first frame. Check edges, hands, the face and abrupt shape changes. If something fails, reduce movement or change the reference before trying to fix everything in the final edit.

04An avatar represents a person

I also show examples created from my own image and voice. This opens up formats but requires review of the script and representation. Visual resemblance does not guarantee natural voice, pronunciation or messaging.

When publishing, explain that content is generated or modified if it could be mistaken for a real recording. Work with the represented person's permission and check that the result does not attribute unapproved messages to them.

05Evaluate the test before expanding the project

The class mentions tool plans and costs at recording time. Those amounts are not current references: features, limits and usage depend on the product and plan available when you work.

As an editorial exercise, prepare a piece lasting a few seconds and record how many attempts you need for a usable version. That record helps assess time, review and resources. The goal is a use that fits your content, not effects for their own sake.

Sources and further reading

These references expand on the concepts indicated. The examples and exercises are original editorial material.

  1. [1] Eric Muriel · YouTube

    From image to video: testing Midjourney, Runway and HeyGen ↗

    Primary source: the author’s Spanish video and supplied transcript. Personal experiences retain their original context and date.

    Back to the related section
How we use sources, quotes and images

Companion video · YouTube

From image to video: testing Midjourney, Runway and HeyGen ↗

Eric Muriel | IA ·

English translation of an AI-assisted adaptation of the transcript supplied by Eric Muriel. The original video is in Spanish. Verbal fillers and transcription errors have been edited, and examples distinguished from promises of results. Suggested exercises are editorial additions. Tools and interfaces reflect the recording date, except for updates identified with sources.

More videos from my channel →