Can AI do VFX? Veo3 is a strong bet for yes.
This is a 16 second spec, done entirely with Veo3 - no actors, no plates, no traditional pipeline. Just a director and AI animators. What does this mean for the future of VFX?
#ai#aivideo#vfx#veo3#KlingAI#midjourney#houdini
I build AI animation tools for a living, so this isn't an argument against AI. But AI use should never be lied about, because audiences judge the effort they believe went in.
Audiences aren't dumb. This season Prada and Dior both released hand-drawn holiday films. Dior's film drew criticism from animators online, who said parts of it looked animated with AI.
Part 2: Dior said every frame of its Christmas film was drawn by hand. Within a day people were calling it AI anyway.
Would you rather a studio told you it used AI, or never mentioned it?
Does knowing something was made with AI make you stop caring about it?
It does for me, which is awkward because I build AI animation tools for a living. So I tried to work out why.
Jaron drops out of the sky swinging. Isaac, under the mirror, catches it.
A test fight for The Myth, our original animated sci-fi series in development. Rain the whole way, a puddle used as a weapon, and a blade slide at the end that ends on Isaac's mirror with Jaron reflected in it.
Which of the two swords would you pick up?
#TheMyth #IndieAnimation #SciFi
Prompting a fight as a timed list of strikes gets you a fight. Prompting it as "an exciting sword fight" gets you two people swaying in the rain.
27 seconds, 6 shots, a beat per second. The blade slide at the end only worked once I made it its own shot.
What do you split into its own shot when a model keeps blending moves together?
Left is a scene from The Myth as it looks now, top right is the hand drawn animatic it follows, and bottom right is the version I made with gen AI in 2024. Two years ago the models couldn't follow an edit, and now the fight follows Mauro's animatic shot for shot.
Animatic on the bottom, animation on top, and every cut, pose and hit in this fight comes from the drawings.
The animatic is Mauro's hand drawn pass at the scene. The animation is the finished shot from The Myth, built straight from it with Seedance 2.5 and Seedream, a Blender anime effects pass for the dust and the hits, and the compositing done with Claude Code. Music: Scars of Time, Yasunori Mitsuda.
#TheMyth
We blocked a whole chase in Blender, fed it to the video model, and got a ghost mule back. What broke, what worked, and where Blender still earned its keep. Which part do you want the full breakdown of?
Keep the shot, change what the character does. The lantern goes up, holds, comes down, and the dock, the light and the framing stay put. Keyframes on a timeline instead of regenerating and hoping. Still rough on timing. What one revision would you make to a generated shot?
One drawing in, a puppet out. Prism reads the joints off a single generated character and cuts each body part to its nearest bone, so the elbow bends because the pose model said where the elbow is. One turntable gives eight angles and every one rigs the same way.
Hand-painted on every frame.
Replace the character: he drifts.
Hard mask edge: it crawls.
Feathered alpha, plus a guard where paint and source disagree: it holds.
An invisible effect wrapped in a visible artifact is worse than none.
Still rough — not on Smithery yet, no SDK polish.
If you build with Claude Code and care about animation, reply or DM. Looking for a few people to break it before I list it.
Got Tether — my plugin for directable AI video in After Effects — wired into Claude Code as an MCP.
The interesting bit: it doesn't expose a prompt box. The agent gets a structured motion vocabulary instead.
The bet: an agent that animates needs the same control discipline as a human animator.
"Make the dragon roar" doesn't tell the model where the breath starts, how the chest swells, where the recovery lands.
Those aren't prompts. They're animation choices.
Hand-drawn, AI styled - we've been working on a rendering system for 2D and 3D animation designed to take rough inputs and create finished results with AI. Still a ways to go, but crazy how far AI has come even in the last 6 months #ai#animation2d#sakuga#anime#comfyui