Several people asked how I took my comic and turned it into a 1-minute video based on the first 3 pages of my book using the Abacus AI agent. Here’s the process:
I turned OVERMIND Page 1 into a 60-second animated video using AI. Here’s the prompt I started with, every problem that came up, and exactly how I fixed each one.
The Initial Prompt
To kick it off, I uploaded the comic pages directly to the AI agent. The AI read the panels: it could see the artwork, the layouts, the character designs. I didn’t have to describe every visual detail from scratch because the visuals were already there. What I paired with those pages was a single prompt covering style, story logic, action beats, narration, and score. This was my prompt with that:
Create a 60-second animated motion comic video for OVERMIND Page 1. Style: Batman: The Animated Series meets X-Men ‘97 — dark, cinematic, sci-fi action. The protagonist is Ayla Rin — she wears slim-framed rectangular glasses, bright white armor with gold shoulder pauldrons, a purple sash with a gold W insignia, gold bracers, and a compact gold jetpack on her back. Red/auburn shoulder-length hair. She is fighting two armed mercenaries in a dark space station interior, hits a door control panel to escape, discovers a bomb planted by the Robeni Pirates in an adjacent room, panics, crashes through a window, and the bomb explodes as she soars away. Narration from the comic captions. Score: dark orchestral noir, Shirley Walker Batman TAS style. 60 seconds, 1280x720.
It came out pretty nicely, but it still had a lot of errors which I had to go through and fix one by one.
Step Zero: Generate the Character Sheet First
Once I realized the character was inconsistent, I had the AI produce a character profile and model sheet for Ayla directly from the uploaded comic pages. This gave the AI a single canonical reference image: Ayla’s glasses, her armor colors, her hair, her jetpack, her insignia, all locked and confirmed from the actual artwork before a single frame of animation was attempted. Even though this is out of order from when I discovered the problem, I thought it was worth putting first because after I did this, the output became more consistent.
That model sheet became the anchor for every subsequent generation. Instead of relying purely on text descriptions like “slim-framed rectangular glasses” and hoping they stuck, I could point back to the locked image: “match this exactly in every frame.” Text descriptions drift. A canonical reference image holds.
If you’re adapting a comic, do this before anything else: generate the character sheet first, confirm it matches your actual design, then reference that locked image in every video prompt. It won’t eliminate all drift — the problems below prove that — but it gives you a much harder starting point for the AI to hold.
Problem 1: The Mercs Looked Like Allies
What happened
In the opening standoff, one of the mercenaries was animated walking casually next to Ayla, like a partner, not an enemy. He wasn’t shooting at her or threatening her. He was just there.
Why
The prompt said “fighting two armed mercenaries” but never said they were actively shooting at her right now, or that the scene was a hostile standoff. The AI filled the gap with its best guess.
The fix
Explicitly state the conflict dynamic and what each character is doing in the scene.
Two armed mercenaries are actively firing at Ayla from opposite sides of the room. They are enemies trying to kill her. She is outnumbered and looking for a way out — this is not a fight she’s winning, it’s a tactical retreat.
Once the relationship was stated explicitly, the scene generated correctly.
Problem 2: The Glasses Kept Disappearing (or Becoming Goggles)
What happened
Ayla’s slim-framed rectangular glasses — a defining character detail — kept getting dropped or replaced. Some frames she had them. Some she didn’t. In the restyle passes, they became thick sporty wraparound goggles.
Why
The glasses were described in the middle of a costume paragraph. The AI treated them as one detail among many and occasionally deprioritized or reinterpreted them.
The fix
Pull it completely out of the costume description and put it at the very top of the prompt in all caps, before anything else.
MANDATORY: AYLA WEARS SLIM-FRAMED RECTANGULAR GLASSES IN EVERY SINGLE FRAME. NOT GOGGLES. NOT BARE-EYED. SLIM RECTANGULAR GLASSES ALWAYS.
That language — capitalized, isolated, at the top — held across every subsequent generation. This became better along with the character brief that it created as well.
Problem 3: The Bomb Sequence Was Completely Wrong
What happened
Two separate errors in the bomb sequence.
First version: Ayla grabbed the bomb off the wall and planted it herself, then ran. She was the one setting the bomb.
Second version: The last panel of the page — Ayla’s white-gloved hand hitting a door control panel — was read as a massive alien hand reaching toward her. She got “captured.”
Why
The prompt said “discovers a bomb planted by the Robeni Pirates” but didn’t explain the sequence of events: that she goes through a door into a different room she hasn’t been in before, that the bomb is already there, that she stops and has a reaction beat before running. The AI compressed and reinterpreted.
The fix
Spell out every transition and cause-and-effect step by step. Don’t describe the image — describe what is happening dramatically.
Wrong:
She discovers a bomb in the next room and escapes through the window.
Right:
Ayla hits the door control panel — this is her hand on the panel, not anyone else’s. The door opens. She walks through into a NEW room she has not been in before. She stops. She sees a bomb already mounted on the wall. This is NOT her bomb. She did not plant it. The Robeni Pirates planted it. Her expression shifts from tactical focus to genuine alarm. Then she runs.
Every transition needs its own sentence. “She escapes” is not enough — the AI will invent the route.
Problem 4: The Finale Was Bright Daylight
What happened
The window-crash and explosion scenes — which should be deep night, with the explosion as the only major light source — kept generating as bright blue-sky daylight. Sunny city, clear afternoon, completely different movie from the dark interior scenes before it.
Why
The prompt described the interior as dark but didn’t explicitly state what the exterior looked like when she crashed through the window. The AI defaulted to daytime.
The fix
State time of day as a hard global constraint, not an incidental detail.
GLOBAL LIGHTING: Deep night throughout the entire video. No daylight anywhere. The exterior is a dark nighttime cityscape — glowing windows below, black sky above. The explosion is the primary light source in the finale — orange-red firelight illuminating Ayla from below as she soars away.
Once it was stated as a global constraint rather than implied by the interior mood, it held.
Problem 5: Title Cards and Credits Appeared in the Video
What happened
The AI added an “OVERMIND” title card and credits at the end of the video — text overlays over the final explosion shot.
Why
It wasn’t told not to. A 60-second animated video “looks like” a trailer or episode, so the AI applied trailer conventions.
The fix
One explicit line in the prompt:
No text anywhere in the video. No title cards, no credits, no captions, no speech bubbles, no overlays of any kind.
Problem 6: Costume Colors Drifted
What happened
In earlier generated sections, Ayla’s armor came out as washed lavender and steel-gray instead of bright white with warm gold. The colors were close but wrong — she looked like a different character.
Why
“White armor with gold pauldrons” is vague about tone. “White” can read anywhere from bright white to off-white to pale lavender depending on the lighting model.
The fix
Be more specific about the exact character.
Bright white/silver armor — not off-white, not lavender, not gray. Warm gold shoulder pauldrons — not yellow, not silver. The colors should read as clean and high-contrast even in dark lighting.
Also: lock a canonical reference image. Point the AI at one specific panel or piece of art and say “match this exactly in every shot.” Text descriptions of costume colors will always drift. A reference image cuts the drift dramatically.
Where It Landed
It took about ten rounds of revision before Page 1 locked. Some passes fixed one problem and broke another. But none of those rounds meant redoing the whole video from scratch, that’s the key thing.
The approach that worked: watch the full video all the way through, write down every single problem you see, then go scene by scene and regenerate only the specific clips that are broken. Scenes that are working get left alone. This matters because regenerating the whole video resets clips that already look right: you fix one thing and break three others. It also costs dramatically more because of tokens. Surgical fixes are faster, cheaper, and protect your good work.
The actual workflow: watch the output → list every problem → fix only the broken scenes → stitch them back in → watch again. Repeat until clean. Never regenerate a scene that already looks right.
The final cut is a real 60-second motion comic: dark interior fight, bomb discovery, window crash, night explosion. Glasses stay on. Mercs shoot at her. No title cards. Colors hold. It’s not perfect frame-for-frame with the page, but it plays as the same story.
Every problem above came from something the prompt left unsaid. Even with the actual comic pages uploaded and visible to the AI, it doesn’t interpret your narrative intent from artwork alone. It can see the panels — it still needs you to spell out the relationships, timing, conflict, and cause-and-effect in the brief. The AI sees the pages. It reads the words. Write the words accordingly.
The New Ayla Rin Comic is also coming soon. Sign up for the Ayla Rin: Gateway to Doom Crowdfund, coming soon for great cosmic comic books with a cover by legendary Green Lantern artist, Ethan Van Sciver:
Impressive. Cool video.