Can AI Actually Make A Bollywood Movie? Or Is It Just Hype (Full Breakdown)
A non-filmmaker challenges himself to create a Bollywood-style fight scene in a single night using AI tools. He compares GPT Image 2 vs Google's Nano Banana Pro for image generation and Seadance 2.0 vs Veo 3.1 for video generation, ultimately showcasing OpenArt's SmartShot as an all-in-one tool that combines the best of both. The video concludes that AI has made cinematic production accessible to anyone with a story idea, while acknowledging human creativity remains essential.
Summary
The video opens by contextualizing the massive effort behind iconic films — James Cameron waiting 15 years to release Avatar due to technology limitations, and Bahubali requiring 15,000 hand-drawn storyboards and 600 VFX artists across 18 countries. The host, who runs AI companies but is not a filmmaker, sets himself the challenge of creating a cinematic Bollywood fight scene in a single night from his laptop.
To understand filmmaking fundamentals, he consulted Claude AI, which explained the full production pipeline — pre-production, production, and post-production — and specifically broke down mood boards (reference images and color palettes created before filming) and storyboards (detailed shot-by-shot sketches). He emphasizes the staggering scale of traditional filmmaking, noting Bahubali took a full year just for pre-production planning.
The host then tests two leading image AI tools — GPT Image 2 and Google's Nano Banana Pro — across three tasks: a character reference sheet, a storyboard frame, and a hero mood board shot. GPT Image 2 wins all three tests based on consistency, prompt accuracy, and usefulness, notably generating full-body multi-angle character sheets with bonus production details like character names and color palettes.
Next, he tests two video AI tools — Seadance 2.0 and Google's Veo 3.1 — on a cinematic dolly shot, multi-shot character consistency, and a fight scene. Seadance wins all three, producing film-grade footage with smooth camera movement, consistent character faces across multiple angles, and realistic rain physics. A notable moment occurs when Seadance generates a Mumbai detective who uncannily resembles Aamir Khan without being prompted. Veo 3.1 fails entirely on the fight scene prompt, returning errors twice.
The host then introduces the video's sponsor, OpenArt's SmartShot, which he frames as solving the core problem of making image and video AI work together seamlessly. SmartShot combines GPT Image 2 and Seadance 2.0 into one pipeline and asks users for a scene description rather than a technical prompt, treating the user as a director rather than a prompt engineer. A single descriptive paragraph about a Mumbai detective fight scene generates a full pre-production document including character reference sheets for two characters, set design, a top-down floor plan, a complete storyboard with specific lens choices (40mm, 75mm, 50mm anamorphic), and then the actual video. The host demonstrates two additional examples — a full Bollywood trailer for a film called 'Aakhri Sanchar' and a cyberpunk reimagining of the Bahubali waterfall scene set in 2080 Mumbai — both generated from single paragraphs. He concludes by affirming that while AI cannot replace the human story and emotion behind great filmmaking, it has made every other part of the production process fast and affordable.
Key Insights
- Seadance 2.0 generated a Mumbai detective character that uncannily resembled Aamir Khan without any such prompt, raising unaddressed questions about the tool's access to celebrity likenesses in its training data.
- Veo 3.1 failed completely on the fight scene prompt, returning errors twice, while Seadance 2.0 successfully generated the requested action sequence — demonstrating a significant capability gap between the two tools on complex multi-character motion.
- The host argues that the fundamental problem in AI filmmaking is not generating images or videos individually, but making them work together — keeping character appearance, lighting, and framing consistent across both modalities — which is where most users give up.
- OpenArt's SmartShot generated a full pre-production document from a single scene description — including two-character reference sheets with front/side/back angles, set design, a top-down floor plan, and a storyboard with specific cinematographer-level lens choices like 40mm and 75mm anamorphic.
- GPT Image 2 won all three image generation tests over Nano Banana Pro specifically because it added unrequested production value — such as character names, color palettes, and equipment lists — making its output function as a real production reference sheet rather than just an illustration.
Topics
Full transcript available for MurmurCast members
Sign Up to Access