How to make a 3D animated short that keeps the same character throughout

Last updated 2026-08-23 · 6 min read

What is the short version?

The hard part of an animated short is not any single shot, it is the same face appearing in all of them. Holding that means generating every scene against one reference and reviewing the whole set together before committing to animation.

1. Go to Vizard Agent and attach a reference image of the character.
2. Describe the scenes and what happens in each.
3. Say the language for the narration and the on-screen text.

What do you need before you start?

A reference image and a sequence. Vizard Agent generates and animates the scenes itself, so you do not need any 3D tools, and the reference is doing most of the work because everything downstream is checked against it.

• The reference image. One clear picture of the character.
• The scenes. What happens, in order.
• The tone. Cinematic, mysterious, warm.
• The language. For both the voice and the titles.
• The length. Thirty to sixty seconds for a single short.

What do you type into Vizard Agent?

Describe the character in words as well as showing it. Vizard Agent uses the image to generate against and the description to judge what it produced, and the two together hold a look across scenes far better than the image alone.

Prompt

Variants worth knowing:

• A recurring series. The same reference carries into later episodes.
• Sound effects on the beats. A tension hit, a transition, a book closing.
• A closing card. A title and a source at the end.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real animated short. Two habits stand out: it reviewed every generated scene three times before animating anything, and it tested two fonts on a real frame before writing a title.

1. Opens the reference image and studies the character closely.
2. Checks the generation and voice tools and reads the film-style guide.
3. Searches for a voice matching the character and records the narration.
4. Extracts the exact word timings so the scenes can be cut to the read.
5. Generates the scenes in parallel with the music search, then reviews every generated frame against the reference for consistency.
6. Animates the opening, the narrator scene and the key story beats, then samples frames from every clip to check the motion.
7. Generates the sound effects the beats need and measures every audio stem in LUFS.
8. Downloads two typefaces for the language and tests each on a frame, checking that the letters join correctly.
9. Assembles the timeline with transitions, renders, then re-renders three times to correct the title sizes, the audio balance and the placement of the closing card.

Step 8 is easy to skip and expensive to skip. A font that does not shape the language properly produces text that is technically present and wrong to any reader of it, and the only way to know is to render a frame and look.

What does the result look like?

From the run this page is written from, probed on the delivered file: 1280x720, H.264, 30fps, 40.5 seconds, AAC audio. Forty seconds, animated scenes holding one character, narrated with subtitles in the requested language, sound effects on the beats and a closing card.

Forty seconds is short for a story and correct for this build. Every additional scene is another chance for the character to drift, so a tight short that holds is worth more than a longer one that does not.

When does this not work well?

Character consistency is the whole difficulty of this format, and image generation is probabilistic by nature. Vizard Agent reviews every scene against the reference before anything is animated, and the constraint is a real one that gets managed rather than solved.

• The character will drift. Reviewing every scene against the reference is what limits it.
• Real people are off limits. Do not use a photograph of someone as the reference.
• Longer shorts drift more. Keep episodes tight and make a series instead.
• Motion can misbehave. Vizard Agent samples frames from every animated clip to catch it.
• Complex scripts need typeface testing. Vizard Agent tests them, and unusual scripts stay the risk area.

How do you fix a result that came back wrong?

Name the scene that is wrong. Vizard Agent keeps the reference image, every generated still, all the animated clips, the narration word timings and the tested fonts, so regenerating one scene is a targeted rebuild rather than starting the whole short again.

• "The face is wrong in scene four." Regenerated against the same reference.
• "The motion is too fast." Re-animated from the same still.
• "The title is too small." Re-rendered, the way this run did.

How does Vizard Agent compare to doing it yourself?

By hand this is slow and frustrating work. By hand it means generating scene after scene and watching the character quietly change between them, animating each one separately, and then discovering that your titles are rendering in a font that breaks the language.

Common questions

Does this work for other animation styles? Yes. The method is the same whether the character is 3D, illustrated or claymation-style: one reference image, every scene generated against it, and the whole set reviewed together before anything is animated.

How many reference images do I need? One good one. Vizard Agent generates and reviews everything against it.

Can I use a photo of a real person? No. Use an illustrated or generated character.

Will the character stay consistent? Vizard Agent reviews every scene against the reference before animating, which is what holds it.

Can it narrate in my language? Yes, and it will test the typeface for that script on a frame before building titles.

Can I make a series? Yes. The same reference image carries into later episodes, which is what keeps a recurring character recognisable across them.

How many scenes can Vizard Agent hold? As many as the short needs, and fewer is safer. Vizard Agent reviewed every scene here in three passes before animating one of them.

How long should one short be? Thirty to sixty seconds. Longer means more chances for the look to drift.

Why record the narration first? Because the scenes are cut to it. Vizard Agent extracts exact word timings from its own narration and builds the timeline against them rather than fitting a voice to finished picture.

Does Vizard Agent check the animation? Yes. It samples frames from every animated clip to confirm the motion is smooth before assembling anything.

Does it add sound effects? Yes, on the story beats, measured in LUFS against the music and voice.