Dolphin AI Turns a Script Into a Multi-Shot Video Without Losing the Costume

A long clip is not the same thing as a scene. That distinction is where most AI video tools still stumble, and it is the problem a new entrant is trying to solve directly.

Dolphin AI, built by the Indian social platform Eloelo, was unveiled in Mumbai at FICCI FRAMES 2026 on September 29. As of October 4 it is in early access with a public waitlist. The pitch is narrow and specific: feed it a script, get back a multi-shot video in which the characters, their clothes, and the locations stay consistent from one cut to the next. It produces clips up to 60 seconds in a single pass, with voice, music and edit points synced to the visuals.

Why multi-shot is the hard part

Generating a single good shot is largely solved for short durations. Generating five shots that belong to the same scene is not. A model that renders each shot independently has no memory of what the jacket looked like two cuts ago, or which side of the room the window was on. Continuity breaks, and the audience notices immediately.

Dolphin approaches this as an agentic workflow rather than a single model call. It plans the shot list from the script, then keeps the references that matter for each character and location locked across the sequence. The company also offers recasting, so a performer can be swapped after the fact without regenerating the whole video, and motion control for shots that need a specific camera move.

The tool supports lip sync in Indian languages, which is a deliberate choice given where it is launching. Generic models handle English well and treat other languages as an afterthought. A tool that ships with Indian language support is aiming at the local creator market first.

How it compares

The continuity problem is not unique to Dolphin, and it is worth being precise about what the alternatives do today. Runway's Multi-Shot mode produces sequences of about ten seconds and often defaults to no audio. Pika now supports multi-shot work up to around 30 seconds with storyboards and sound design. Luma's Dream Machine typically caps clips in the five to ten second range. Dolphin's claim is longer single-pass output with audio and continuity handled together, which, if it holds up outside a demo reel, is a step past the current options.

Whether it holds up is the open question. Company demos are edited for the best results. Independent tests will show how often a character's wardrobe drifts once the shot count climbs and the runtime stretches toward a minute.

The integration angle

Dolphin exposes an MCP server and a command-line interface, which lets other tools trigger generations directly. That matters more than it sounds. A studio that already runs a pipeline for scripting, asset management and review can call Dolphin from inside that system instead of exporting files to a web app and back. Pricing follows a subscription plus credits model, with Indian pricing still unannounced, and the terms name Starter, Pro and Creator tiers.

The agentic framing also answers a workflow complaint. An editor needs a video and, just as often, a version of it that can be revised without starting over. Recasting without regenerating, and motion control on a single shot, are both aimed at revision rather than first-generation. That is the part of the pipeline where time actually goes.

The workflow behind the continuity

Keeping a character consistent across cuts usually means building a reference set. The agent extracts the details that matter from the script and the first generated shot, then holds them as constraints for every shot after. A jacket has a colour and a cut. A room has a window on one side and a door on the other. Once those are fixed, later generations have less room to drift, and the video can be assembled without a reshoot.

The hard cases are the ones where the reference itself changes. A character who takes off a coat mid-scene, or a location shown from two angles that were never generated, forces the system to decide what is still consistent and what genuinely changed. That is where the planning step earns its keep, and where a simpler "generate each shot separately" tool falls apart.

Dolphin also lets a user recast a performer after generation, which implies the character identity is tracked separately from the performance. That separation is what makes revision cheap. If identity and motion are stored independently, changing one does not force a rebuild of the other.

The compliance layer

Anything aimed at Indian creators has to deal with disclosure rules, and Dolphin's material says so. YouTube requires labels when realistic content uses synthetic media, and repeated non-disclosure carries penalties. India's advertising regulator has draft labelling rules for AI-generated ads, and intermediaries must follow the country's IT Rules for unlawful content. The guidance that comes with the tool tells users to disclose synthetic shots, secure likeness permissions, track model licences, and keep prompt records for audits.

That guidance is the least glamorous part of the launch and probably the most consequential. Creators who skip it are the ones who get their accounts restricted.

Why the Indian launch matters

India is one of the largest creator markets in the world, and most of its output is in languages that global video models handle poorly. A tool built locally, with local language support and local pricing, is addressing a gap that the big studios have not prioritised. If Dolphin's continuity holds up, it could take a meaningful share of a market that has been underserved.

The pricing question is still open. Indian pricing has not been announced, and the tiers named so far are generic. What a creator pays per finished minute will decide whether the tool is a professional option or a hobbyist one. Comparable global tools charge by credits for generated seconds, and the effective cost per usable shot, after retries, is often several times the sticker price. Dolphin will be judged on the same basis.

There is also a talent angle. The company's own material points at roles that benefit: AI video editors, prompt-led story producers, brand storytellers and agency creatives, with guidance to build reels that demonstrate multi-shot consistency. That is a sign the tool is being pitched at people who will use it for client work, where a broken continuity means a reshoot and a missed deadline.

What to watch

Two signals will show whether Dolphin is more than a regional launch. The first is independent continuity testing on clips that approach the full 60 seconds. The second is whether the MCP and CLI integrations get adopted by studios that already have a pipeline, since that is the difference between a toy and a tool.

The Indian market angle is real, though. A local tool with local language support and local compliance guidance has an advantage that a global model shipping an English-first product does not. Whether that advantage survives contact with a fast-moving field is what the next few months will tell.