Case Study: Two Hindi Industrial Films for MAHA-GFC

Our team made two Hindi films in Hyperdream for Bharat Cluster Ventures (BCVPL) about MAHA-GFC, its planned furniture city in Palghar, Maharashtra. The second film, 22 shots long, took 1,200 credits, 40 video renders and 5.5 to 6.2 hours of logged work. Both final cuts were edited outside the app.

Key takeaways

• Film 1 (July 2026): 6 scenes, 9 shots, 7 characters, 4 places. 21 video takes, or 2.3 per finished shot. 2.9 to 4.8 hours of logged work.
• Film 2 (August 2026): 7 scenes, 22 shots, 4 named characters, 4 places. 40 video renders, or 1.8 per finished shot. 1,200 credits. 5.5 to 6.2 hours.
• 18 of Film 2's 40 video renders were thrown away. That is what 22 usable shots really cost.
• 24 of those 40 renders were reference-to-video: 60% of the renders and 69% of the video credits.
• Three of the models used have since been retired. The credit total is at summer 2026 prices.

Who is BCVPL and what did it need?

BCVPL calls itself "a DPIIT-recognized national initiative creating integrated, world-class industrial clusters" (bharatcluster.com, read 10 November 2026). Its flagship project is MAHA-GFC, Maharashtra Global Furniture City. Its site describes that as "India's first integrated industrial park dedicated to the furniture manufacturing value chain", with a "600+ acre vision" in Palghar (MAHA-GFC page). The site also lists an MoU with the Government of Maharashtra signed at Davos, and access to JNPT, Mumbai Port and the upcoming Vadhavan Port. These are BCVPL's own statements. We did not check them ourselves.

The offer to a furniture maker is to move into one shared city with suppliers, designers and exporters. That is hard to picture from a brochure. The films had one job: show a manufacturer where they would fit.

<!-- QUOTE: BCVPL -->

What are the two films?

They make the same case at two levels. Film 1 is about the whole industry. Film 2 is about one manufacturer.

Film 1: "MAHA-GFC Ep-1: अगर पूरा Furniture Industry एक जगह आ जाए तो?" ("What if the whole furniture industry came together in one place?"). Started on 22 July 2026. It is a cinematic documentary in 16:9, written for a single deep Hindi narrator, with 94 seconds of kept footage.

Its seven characters are types, not named people: The Carpenter, The Manufacturer, The Woman Entrepreneur, The Young Engineer, The Farmer, The Foreign Buyer and The Showroom Owner. The scenes, in the order they play:

1. THE FRAGMENTED INDUSTRY (बिखरा हुआ उद्योग), 2 shots
2. THE COST OF WORKING ALONE (अकेले चलने की कीमत), 1 shot
3. THE VISION EMERGES (कल्पना), 2 shots
4. EVERYTHING IS NEARBY (सब पास है), 1 shot. The project stores this as "SCENE 3A": a connector added between the vision and the family beat.
5. IT BECOMES A FAMILY (रिश्ता), 1 shot
6. THE REVEAL (रिवील), 2 shots

Film 1 as BCVPL published it on its YouTube channel on 25 July 2026, edited outside Hyperdream.

Film 2: "MAHA-GFC — विरासत से विश्व तक" ("From Legacy to the World"). Started on 21 August 2026. It follows Rajendra "Raj", a manufacturer, with Meera Kulkarni (production manager), Anil Pawar (senior technician) and Sameer Joshi (MAHA-GFC representative). Raj speaks to camera in Hindi, and his voice comes from the video model. It has 144 seconds of kept footage.

Its seven scenes run from a factory before dawn to an export route: Factory Before Dawn — A Family Legacy, A Business Built on Quality, The Distance to the World, Discovering MAHA-GFC, A Business Can Finally Move, The Route Outward, and From Legacy to the World.

Film 2 as BCVPL published it on its YouTube channel on 30 September 2026, as "MAHA-GFC: एक सोच से एक संकल्प तक".

Film 2, scene 4: the MAHA-GFC campus from the air. A still frame made before the video shot.

How did the characters and places stay the same?

Each character and each place got a reference image before any shot was rendered: 7 characters and 4 places in Film 1, 4 and 4 in Film 2. Those images were then passed to the video model with the shot.

The Carpenter's reference image, Film 1.

Raj's reference image, Film 2.

That was how the app worked in summer 2026. Today each character gets a reference sheet instead of a single image: four full-body views head to toe and four expressions. A place gets its own sheet too.

In Film 2, 13 of the 22 kept shots were reference-to-video, where the model is shown the character instead of being asked to invent them again. In Film 1 it was 5 of 9. The rest started from a still frame (image-to-video).

How long did the films take?

"Active time" is the sum of logged work sessions. The low figure ends a session after a 15-minute gap, the high one after 30. It leaves out the final edit.

Director, Hyperdream's film agent, ran in Agent mode on both. On Film 2 it took its first turn 8 seconds after the project was made. It started shot renders 19 times on Film 1 and 36 times on Film 2.

How many renders and credits did Film 2 take?

68 renders and 1,200 credits. No refunds exist for this project.

Veo 3.1 Fast, Veo 3 Fast and Nano Banana 2 are named here as history. The films were made in July and August 2026, and those three models have since been retired. Today's list is on the models page.

Video took 1,094 of the 1,200 credits, across two Veo models and Kling O3 Pro in two modes. The images came from two models of one family.

We don't have a full credit total for Film 1. The ledger only tagged spending by project from 28 July 2026, and 14 of its 21 video takes were made before that.

What went wrong?

18 of Film 2's 40 video renders were thrown away. Reference-to-video shots and image-to-video shots needed about the same: 24 renders for 13 kept shots, and 16 renders for 9 kept shots. Both come to about 1.8.

Film 1 needed more takes per shot: 21 for 9, or 2.3. Kling O3 Pro made 18 of them and the Seedance 2.0 family made 3. At least 4 attempts failed outright on 28 July. All 4 were refunded.

Neither film was finished in the app. No shot in Film 1 had sound switched on, and the records hold no music or narration render, and no export, for either film.

How was the final cut made?

Outside Hyperdream. The shots came out of the project as files and were cut in the team's own editing software.

Today the Edit step has a timeline, and Export makes one MP4 at up to 1080p plus a zip of the film's files. The zip is the route for anyone who wants to finish in their own editor. The Edit step does not replace professional editing software.

What would this cost on today's plans?

Director is on Pro and up. Pro is $99 a month with 2,000 credits (checked on /pricing, 10 November 2026), so Film 2's 1,200 credits would fit inside one month.

Don't read 1,200 as today's price for this film, though. It was spent at summer 2026 model prices, on a line-up that has changed. To estimate a film now, see the cost of an approved AI film.

How would you make a film like this in Hyperdream today?

1. Open Director, type the brief or paste a script, and pick 16:9.
2. Script and Look: settle the words and the look first. Film 2 went through 5 script versions before any picture was made.
3. Cast and Places: make one reference sheet for each character and place. Use types for an industry film, one named lead for a story.
4. Storyboard and Shots: check a still frame for every shot, then render. Budget about two renders per finished shot.
5. Edit: assemble and export, or download the zip and finish in your own editor.

Related: Shilp Group's Diwali film, a 30-second ad made the same way, and the 20-second two-character test you can run on any tool. To start a film, see the AI movie maker.

How we checked

Every number about the films comes from the two project records and the credit ledger, read on 10 November 2026. Company facts come from bharatcluster.com, read the same day, and are BCVPL's own claims. What we could not confirm: Film 1's total credits, which single shots took the most takes, and anything about the final edit, because it happened outside the app. The ledger and the take history disagree on which model made Film 2's last two shots; we used the ledger. An earlier write-up said "about two hours" and "about four hours". The logs don't support those, so we use the measured ranges.