How we built the first Tanglish AI comic

A profession nobody believed in

A decade ago one of us told his sister he wanted to be a cinematographer. She said that wasn't even a profession. But the pull was never the job title — it was the passion to make something that is felt, not measured or merely seen.

Over time we became designers. How different is that from cinematics, really? It's all vision and direction — except this time we realised we wanted to direct whatever came next. Doing anything at scale costs like hell in production, so in the meantime we experimented on social media and with whatever we could make alongside like-minded friends. When AI turned up as a way to project ideas instead of keeping them at a distance, we jumped in.

Three goals, one toolkit: direct a sitcom, draw like a comic artist, publish a book.

A premise, some quirks, a moodboard

How did we start? One of us wrote what it should be about. Gave the characters their quirks. Mapped how the story should flow. Built a title and concepts on top of that, then a moodboard around the vision.

Then we pitched it to a friend, Sumith, who was already actively playing with content creation. The moment he heard it, he hopped on with his own expertise — and that's when it stopped being one person's idea.

Decoding the characters from both sides

Sumith changed the horizon. He wrote a ton of prompts to generate characters and decode them through his own interpretation. While one of us was generating on Gemini, he was running the same ideas through ChatGPT — two labs, two models, the same cast being pulled into focus in parallel.

We kept iterating until the designs converged. Once we were close to what we wanted each character to look like, we finalised the art-style prompt and locked it.

Six hours for one intro

By now we were impatient. We wanted a trial run — to know whether we could actually pull this off or not. Around the same time, Sumith ran a single shot, a single prompt, to get the whole story out in one go.

So we spent a complete day writing and editing prompts to get the pictures for the pilot episode. Just getting the intro scene established took six-plus hours — along with the assembly in Figma, which we used to add frames and adjust the text and images. It worked.

Fanboys who grew up on anime and manga

The friends came in one by one. From being fanboys, to visiting Japan, to actually bringing the culture into the work — that's always been an adventure. Sreekar, Ambi, Shan, Anand — each of them landed with the same energy.

What we hadn't said out loud until then was how much we all have a thing for anime and manga. That shared thread is really what held the group together.

The expressions the panels lived on

As the pilot came along, we felt a ton of expressions were missing — the ones that were subtle, but important for the frames. Sumith started building expression sheets to check each character's range.

But to actually beat around the comic-book expression — the exaggerated, readable kind — we landed on chibi-style expressions. Sumith led those sheets too, and that's what finally gave the panels the range they needed.

Consistency and composition, the new walls

As Sumith pushed toward a stable daily rhythm in the story, he started hitting roadblocks — either the characters weren't consistent, or the scene composition was off. A single prompt to “put it all out” wasn't good enough: consistency and composition kept failing.

So the tasks split cleanly. One of us set up the prompts. Sumith worked on making the artwork unique and in-line with the style.

Agreeing on an order, not a timeline

Eventually we sat down and talked through how we actually move forward with generating the comic. The first realisation was just how much work is ahead — episodes, then characters, then environments, then props, then the generations themselves.

Episodes: one of us takes the writing and reviews it thoroughly before anything is treated as final. The generation side: the other takes the Higgsfield side — learning it properly before writing the character art-style specifically for it.

The sequence: write and review episodes → learn the tool and write the art-style → generate characters, environments and props together → set up voice-prompting → start planning the generations.

What each tool was actually good at

We wrote prompts with Claude, generated characters with Higgsfield — a ChatGPT base character, then Seedream 5.0 for the full character sheet. Scene and environment construction was an absolute time-killer, even though the model and flow were the same.

Google Flow gives you consistency between characters; ChatGPT is good at maintaining the artist look. We combined both for the outcome. And because we wanted panels to be exact, we did the assembly ourselves in Figma, dialogue boxes and all.