Mahabharat in AAA cinematic.

AAA cinematic looks (the kind in Black Myth: Wukong, Sekiro, God of War Ragnarök) are a specific aesthetic — hyper-detailed PBR materials, dramatic key lighting, atmospheric haze, cinematic depth of field, micro-detail in costumes.

AI models know this aesthetic well. They've been trained on thousands of frames. So "AAA fantasy warrior" lands in one or two passes. But ask for "Bhishma in AAA cinematic style" and the model has no idea who Bhishma is. It defaults to generic Asian fantasy. Or worse, to Hollywood's idea of Indian mythology — Bollywood and gold trim. We had to teach it.

Original Sanskrit descriptions. Raja Ravi Varma paintings. B.R. Chopra's 1988 TV series for costume and props. Amar Chitra Katha for iconography. Indian temple sculpture for body proportion and pose.

Then we built character sheets for 12 main characters:

Each character took 80 to 150 image generations to lock the sheet. Total for character sheets alone: about 1,300 images.

We weren't going to default to Chinese palaces or European castles. We built reference packs for five world spaces:

• Hastinapur palace exteriors — sandstone, lotus motifs, elephant pillars, intricate jharokhas.
• Kurukshetra battlefield — flat dusty plain, distant mountains, dawn light.
• The dice hall — Mughal-influenced indoor architecture, gold inlays, marble.
• Khandav Van forest — dense tropical, banyan, mist.
• Indraprastha — dreamlike marble, mirror floors, surreal scale.

Architectural references pulled from real heritage sites — Hampi, Khajuraho, Amer Fort, Mehrangarh — fed alongside game references for lighting and atmosphere.

Five lighting principles, pulled from Black Myth: Wukong and Sekiro, built into a "lighting bible" with reference images for every scene type:

• Volumetric haze in every scene for depth.
• Strong rim lighting to separate characters from backgrounds.
• Warm key, cool fill colour contrast in interiors.
• Practical light sources — oil lamps, torches, sunlight through windows — as motivated lights.
• PBR materials with high specular response on armour and jewellery.

Every shot in the final film references this bible.

Most attempts at Indian mythology AI fail here. They get the characters approximately right and then put them in generic fantasy environments. The mismatch breaks everything.

We built specific set pieces from scratch:

• The Sabha hall (dice game): 100-pillar marble hall, elephant carvings on every pillar, suspended oil lamps, chess-board floor. 200+ environment shots to lock this single location.
• Kurukshetra plain at dawn: flat reddish-brown earth, distant Himalayas, rolling fog, vultures circling. ~150 generations.
• Krishna's chariot: specific design with white horses and lotus motifs. ~80 generations.
• Forest of exile: dense banyan canopy, hanging roots, soft god-rays. ~100 generations.

Total environment generations: roughly 800 images.

Even with sheets locked, character consistency breaks when dropped into a new environment. Our pipeline:

1. Generate base shot with character + environment via multi-reference input.
2. Check character consistency (face, armour, jewellery — must match the sheet).
3. If broken, regenerate with stronger reference weighting.
4. Once locked, generate motion (img-to-vid).
5. Cut motion to 4-second clips usable in edit.

We used Kling 3.2 for complex motion (chariot rides, battle sequences) — best long-term consistency. Seedance 2.0 for static or slow-moving shots — multi-file reference input (up to 12 files). Veo 3.1 for tight close-ups — highest motion fidelity.