How to Turn a Bollywood Scene Into a 90s Anime Using Only AI

Screenshot a scene → style-transfer it into anime → animate it with dialogue using Google Flow (Veo 3). Three tools. No animation software. No team.

Deewar's confrontation scene between Vijay and Ravi is one of the most iconic dialogues in Indian cinema. The staging, the lighting, the silence before the reply — it's already cinematic.

That's the insight: great source material makes great AI output. You're not prompting from scratch. You're giving the model something to hold onto.

1Screenshot the Source Scene

Find the moment in the original film. The one frame where the emotion is at its peak — before the dialogue, during it, or just after. Composition matters. Lighting matters. Don't grab a blurry mid-motion frame.

For Deewar, I grabbed two:

• —Vijay facing his brother — gray suit, stone alley, contemptuous stillness
• —Ravi's close-up — the quiet before the kill shot

2Style Transfer Into Anime

Feed each screenshot into ChatGPT Image Generation or Nano Banana Pro with a tight style prompt.

The goal: preserve the character's face and lighting, shift the aesthetic to 90s Japanese cel animation.

Convert this into a 90s hand-drawn Japanese anime cel illustration.
Thick ink outlines, hard cel shading, muted noir palette — City Hunter /
Golgo 13 / Crying Freeman era. Preserve the face, hairstyle, skin tone,
costume, and lighting composition exactly. No moe. No modern anime style.
No big shiny eyes. Keep the cinematic gravity of the original.

You'll likely need 3–5 attempts to lock the face. When it's right, export it.

3Write the Video Prompt

This is where craft separates you from prompt-slop generators. Your prompt needs to do three things:

1. 01Anchor the visual style (cel animation era, linework, palette)
2. 02Lock the character's face to the reference image you just generated
3. 03Carry the emotional weight — the anger, the restraint, the pause

For Vijay's line:

90s Japanese anime cel animation, hand-drawn, thick ink outlines, hard cel
shading, muted noir palette — City Hunter / Golgo 13 / Crying Freeman era.

Static medium two-shot, no camera movement.

Two men face each other in a dim stone alley at night. RIGHT: Vijay — thick
black wavy hair, sharp arched brows, brown skin, gray suit jacket over dark
navy shirt. Chin lifted, jaw locked, cold contempt. Match the exact face,
hairstyle, suit, and lighting of the attached reference image.

Action: Completely still from the shoulders down. Only mouth and jaw move.
Nostrils flare on the last word. A single hard exhale.

Dialogue — spoken in Japanese, cold cutting fury, controlled but rising,
biting each word, ending on a sharp challenging tone:
"今日、俺にはビルがある、財産がある、銀行の預金がある、屋敷がある、車がある…
お前には何がある?"

Single hard rim light from screen-right. Deep noir shadows. Muted brown
and indigo cel grade. Hand-drawn linework. Subtle paper grain.

Negative prompt: avoid CGI smoothness, modern anime style, moe style, big
shiny eyes, camera movement, identity drift, slow monotone delivery.

No background music.

For Ravi's reply:

90s Japanese anime cel animation, hand-drawn, thick ink outlines, hard cel
shading — City Hunter / Golgo 13 era.

Static tight close-up, no camera movement.

Ravi — thick black wavy hair, brown skin, soft dark eyes, gray-purple suit
jacket. Face fills the frame. Looking up slightly, off-screen right.
Match exact face, hair, and lighting from attached reference image.

Action: One beat of complete stillness. Eyes lift and lock. Corner of mouth
trembles once. Then speaks — not soft, not slow. Restrained fire.
The last two words land like a blade.

Dialogue — spoken in Japanese, low voice trembling at the start, landing
hard and clear at the end:
"俺には… 母さんがいる。"

Hard rim light from screen-right. Shadow swallowing the left half of the
face. Background dissolves into black.

Negative prompt: avoid CGI smoothness, modern anime style, camera movement,
identity drift, tearful sobbing, weak delivery, dull affect.

No background music.

4Animate in Google Flow (Veo 3)

Upload your reference image. Paste your prompt. Set to 9:16 or 1:1 depending on your platform. Generate.

Veo 3 handles lip sync better than most models right now, especially for non-English dialogue. The Japanese works in your favour — it sounds cinematic, it holds the frame, and the model has strong training data on anime speech patterns.

Generate 10. Keep 1. That's not a suggestion — it's the workflow.

What Makes This Different

Most people generate anime characters from scratch and get generic output.

You're doing the opposite — grounding in a real source performance, locking the face through style transfer, then animating with intentional prompting. The model has something real to hold onto at every step. This is the writing-first, craft-forward approach. The quality gap is visible.

Tools Used