From script to video: an AI workflow for Indian creators

Why order matters

Video generation is the slowest and costliest step. If you change a line of dialogue after generating twelve clips, you may lose several of them. Editors learn this on real shoots: you do not roll the camera until the script is locked. AI filmmaking is the same discipline with a different camera.

The example: "Aakhri Local"

Meera, a nurse, boards the last local train after a night shift and falls asleep. An old woman wakes her gently at her station. On the platform, Meera realises she nursed that woman's husband weeks ago. No dialogue except two lines. Forty-five seconds.

Step 1: Lock the script in two columns

Write what we hear on the left and what we see on the right. Hindi scripts work exactly the same way.

SEE                                              HEAR
Platform at night, Meera runs, train doors open   Train horn, footsteps
Meera sits, head against the window               Rattle of the train
Old woman opposite, watching her                  Soft rain on the roof
Old woman touches her hand                        "बेटी, तुम्हारा स्टेशन आ गया।"
Meera on platform, turns back, recognises her     Silence, then the train pulls away
Close on the old woman at the window, smiling     "ध्यान रखना अपना।"

Step 2: Break it into a shot list

If shot sizes and camera moves are new to you, learn the basic set first. It is the vocabulary every AI video model understands.

Step 3: Storyboard before you generate

A storyboard is a rough frame for each shot. It shows you, cheaply, whether the story reads without words. If you cannot follow "Aakhri Local" from six frames, a generated video will not fix it. Look for missing shots (do we see her get off the train?) and fix them now.

Step 4: Lock the characters and the look

Consistency is the biggest problem in AI films. Write a character sheet and paste the exact same words into every prompt.

MEERA: Indian woman, early 30s, navy hospital scrubs under a grey cardigan,
hair in a low bun, small steel wristwatch, tired eyes, no jewellery.
OLD WOMAN: Indian woman in her 70s, white cotton saree with a thin blue border,
silver hair, round glasses, kind face.
LOOK: late night Mumbai local train, warm tungsten interior light,
cool blue night outside, light film grain, muted colours.

Then make one reference image for each character and reuse it in every image-to-video generation.

Step 5: Generate stills first, then animate

Still images are faster to iterate than video. Get the frame right as an image, then turn it into motion. Here is a prompt for shot 3:

Close-up of an Indian woman in her 70s in a white cotton saree with a thin blue
border, silver hair, round glasses, seated by the window of a late night Mumbai
local train, watching someone across from her with a gentle, knowing expression.
Warm tungsten interior light on her face, cool blue city lights blurred outside,
light film grain, muted colours, 50mm lens, shallow depth of field.

For the motion pass, describe only what moves: "she blinks slowly and her eyes soften, rain streaks on the window behind her, the train sways gently". One clear movement per shot looks more real than three.

Step 6: Voice and dialogue

Record the two Hindi lines yourself or with an actor, or use Hindi voice-over. Use lip sync only where a face is visibly speaking on screen. In "Aakhri Local", line 1 can be off-screen over the hand insert, which avoids lip sync entirely. Fewer talking faces means fewer things that can look wrong.

Step 7: Edit to the voice

Lay the voice and key sounds on the timeline first, then cut pictures to them. Hold shot 3 longer than feels comfortable; the silence is the story. Use a J-cut so the train horn starts before shot 1 appears. Add room tone under every shot so the audio never drops to dead silence.

Step 8: Music, sound and captions

Add soft rain and train ambience, and keep music low. Then caption the dialogue: Devanagari for Hindi audiences, Roman Hinglish if your audience reads that way, English if you want reach outside Hindi speakers.

Step 9: Label and export

Export 9:16 for Reels and Shorts, 16:9 for YouTube. Because "Aakhri Local" has realistic AI people, turn on YouTube Studio's altered or synthetic content setting, use Instagram's AI label, and write "Made with AI" in the description. Labels build trust; viewers feel cheated when they discover it on their own.

Doing this in Kritiverse

The whole order above is how the Kritiverse AI film maker is laid out: idea, script, storyboard, shots, edit, captions and voice, export. You can write the script in Hindi, Hinglish or English, send it to Storyboard, and set each shot with Direction Mode, either by choosing options yourself with Add Direction or letting Auto Direction suggest shot size, lens, camera move and lighting. Generate lets you pick the model for each shot. The cost is shown before every click, and failed generations are refunded automatically. The Video Editor works in the browser with your local files, and Captions and Hindi voice-over finish the film.

Common mistakes

• Generating before the script is locked. Every rewrite costs clips.
• Describing characters differently in each prompt. Copy the character sheet word for word.
• Too much camera movement. A push in, a pull out and static shots carry most short films.
• Ignoring sound. A silent AI clip feels fake. Ambience makes it believable.
• Making every shot a talking face. Use inserts, reactions and off-screen lines.
• Skipping the label. Realistic AI people need a clear AI disclosure.