The films in this post
In short. Between 12 and 19 November 2026 we finished more than 60 films and cuts on one PC: four short series, single films, 3D and clay films, and three ads for our own bag brand, Stashed. Most of it is AI films made in India, on a desk above a jewellery shop in Vaniyambadi. The films above are a few of them. The new ones are too new to have view numbers, so this post is a log of what we made and what we learned, not a results report.
What did we make in eight days?
The count comes from our project folders, with each version and language cut counted once.
Everything ran on one PC with an RTX 5070 Ti (16 GB), a Ryzen 7 9700X and 32 GB of RAM. A PC like this costs about ₹1.5 to 1.6 lakh, once (Nav's figure, November 2026 prices). The video model runs on it, so there is no per-video fee. The monthly costs that remain are a Claude plan, a ChatGPT plan for the storyboard images, and electricity. Here is how that setup works.
Which AI short series did we start?
An AI web series in Hindi, one in Tamil, and two drawn in code. Each series is a chain: every episode connects back to the last.
| Series | Language | Episodes | How it is made |
|---|---|---|---|
| Trishul | Hindi | 5 (14 to 18 Nov) | AI video from storyboard sheets with LTX-2.5 |
| Kolar | Tamil | 4 (13 to 17 Nov) | AI video from storyboard sheets with LTX-2.5 |
| Andar Ka Match | Hindi | 3 (17 Nov) | Drawn entirely in code; every voice is Nav's own recording |
| One Word Play | Hindi | 15 reels | Drawn in code, in Nav's cloned voice |
Trishul is a Hindi sci-fi story. Nav plays NAV-07, an android with a trishul locket where the heart should be, who fights KAAL robots that turn out to be copies of him. Episode 5, the finale, is the first film above.
Kolar is a four-episode Tamil street drama about a quiet boy who leaves Vaniyambadi for Chennai wearing his father's silver kada and one rule: keep your hands quiet. We are not showing its films yet.
Andar Ka Match has three episodes, the next three films above. Inside one man's head, dream (SAPNA) and home (GHAR) arm-wrestle. One Word Play is the word format from our 70,000 followers post, now 15 new reels, for example देर → दूर, बस → बेबस and दाम → दम. Their 10-second teasers are in the films library.
What else did we make besides the series?
- Single films. NETA, a fictional election trailer. Three Wonders. The Last Light and its making-of, Ek Lau. The Last Bus. The Room Upstairs, in English, Hindi and Tamil (the English cut is above). Basement, Please. Night Recovery.
- 3D and clay. The Pickup, a clay film of the two of us. Agnayya, a clay comedy set in the Mysore palace. The Last Bus, a clay story of a lost suitcase. 1 cut vs 25,000 cuts, a Blender physics film shot at a Chennai fruit cart. We wrote these up in 3D animation without a 3D team.
- Ads for our own brand. Three films for Stashed, our bag brand. Making an ad for a brand you own is the safest place to test a new method.
Do these films have numbers yet?
No. The only numbers we can give are older and already public on the site:
| Date | What | Number |
|---|---|---|
| 12 Nov 2026 | नशा → नाश | 6.2M views |
| First seven days | रख → राख | shared 76.7K times |
| 12 Nov 2026 | @_navaratan_ followers | 71.1K |
The films from this week are too new to have numbers. When they do, we will add them here, with the date.
What changed in how we work this week?
Four things became rules.
- Voice first. We record or generate the voice before any picture. The voice sets the clock, and every cut lands on a spoken word.
- One storyboard sheet per sequence. One image from ChatGPT Images through Codex, cut into panels, so every shot shares the same cast, light and grade.
- A critic before anything ships. A second agent watches the cut and scores it. On one film, two rounds took the score from 5.5 to 7.5 out of 10.
- Every series episode connects back. Every new episode connects back to the earlier ones. This is the main reason a series feels like a series and not a folder of clips.
Our render logs give a plain number for the cost. Across Trishul episodes 3 to 5 and Kolar, 87 LTX-2.5 renders took 5,554 seconds, an average of 63.8 seconds each. Roughly three to four renders go into every shot that stays in the final cut. The full method is in LTX-2.5 offline AI films on an RTX 5070 Ti.
What did not work this week?
- Faces drifted away from the cast between shots. We cut the shot or re-rendered it with a camera lock we call hold.
- Objects morphed and hands melted in some takes. We re-took them with a plain instruction.
- Text and logos garble in AI video, so we add them in the edit.
- On the Plus plan, ChatGPT Images stopped at 75 images in one usage window while we made NETA, and the reset came later than we estimated.
None of the fixes are clever. They are checks, repeated.
What are we changing on the site and in the course?
The Lab now teaches what we actually did this week, in this order: AI films offline on 7 December, short series on 14 December, 3D on 28 December. The prices do not change. On the site we are adding more of these films to the films library as they finish. The workflow behind all of it is on the director page.
Learn it
The 2BFT Lab teaches this. It is ₹10,000 once for the live course (live every Wednesday and Sunday, 7 PM IST, with every recording) or ₹1,000 for the recordings alone. The AI films session is 7 December, short series 14 December, 3D 28 December. You need your own Claude plan (Pro is about ₹1,800 a month, paid to Anthropic, not included). See pricing.
