FrameKeeper

Project: FrameKeeper

Self-reported basis: Author's own description, unverified

Commercial due-diligence read: FrameKeeper appears to be a proof-of-concept tool for detecting continuity issues in generative video and generating correction prompts. It is built by one person using OpenAI’s Codex and GPT-5.6. The author states it was submitted to the OpenAI 2026 hackathon, but no evidence of revenue, customers or traction exists. The most important open question is whether this concept can scale into a production-grade tool that integrates with generative video platforms.

The description states that FrameKeeper is an AI continuity director for generative video. It analyzes scenes across a project and checks them against persistent story rules. It identifies continuity issues involving:

• Characters
• Clothing
• Environment and weather
• Props and object state
• Movement
• Visual style
• Scene-to-scene consistency

When an issue is detected, it shows:

• Expected state
• Detected state
• Explanation
• Confidence score

It can then generate a precise correction prompt for regenerating the affected scene.

The product also maintains a Continuity Bible, which contains persistent project rules for characters, environments, props, motion, and style.

Inference: The tool is described as being built using Codex and GPT-5.6, and it operates by sending scene context, continuity rules, and images to the model to receive structured analysis.

The author states that FrameKeeper was built to solve a problem in generative video: creating a complete multi-scene story with consistent characters, props, environments, and story states remains difficult. The tool is positioned as an AI continuity director that helps creators avoid issues like:

• Characters changing appearance
• Weather disappearing
• Objects switching state too early
• Scenes drifting from the intended story

It also claims to explore how AI can act not only as a generator but as a production supervisor.

Inference: The positioning is evolving from a hackathon prototype to a potential foundational layer for AI-native filmmaking. The author mentions future versions could include automatic regeneration, timeline-aware tracking, and integrations with generative-video platforms.

The description does not state who the target customer or ideal customer profile (ICP) is. It implies that FrameKeeper is aimed at creators working with generative video, but no specific segment (e.g., indie filmmakers, studios, content creators) is named.

Inference: The tool likely targets users of generative video tools, particularly those who are creating multi-scene projects and need consistency across shots. However, the lack of customer data or segmentation makes this unclear.

There is no evidence in the description of a business model or pricing structure. The author mentions that the hosted Build Week version includes Demo Mode for judges to test the workflow without an API key, but there is no indication of monetization plans.

Inference: If FrameKeeper becomes a product, it may be priced as an API or SaaS offering, but this is not stated.

The project was built using:

• Codex
• GPT-5.6
• Next.js
• React
• Tailwind
• TypeScript
• OpenAI

It uses Codex for development tasks (design, implementation, debugging, testing, documentation, deployment), and GPT-5.6 for structured continuity analysis.

The tool sends scene context, continuity rules, and images to the model and receives structured output.

Inference: The use of Codex suggests a developer-focused approach to building the application. The reliance on GPT-5.6 indicates that the core logic is AI-driven, with a focus on structured outputs for analysis and correction.

The project was submitted to the OpenAI 2026 hackathon, but there is no evidence of revenue, customers, or adoption beyond that. The author states it’s a prototype built in a short timeframe (Build Week), and future versions are planned.

Inference: No traction data exists. The tool is described as a proof-of-concept with no indication of usage or market validation.

The description does not mention any competitors. It implies that there is a gap in the generative video workflow between generation and editing, and FrameKeeper aims to fill that gap by acting as a production supervisor.

Inference: There are likely no direct competitors yet, but the space of generative video tools is growing rapidly, and this could be a niche opportunity for AI-native filmmaking workflows.

• Single-person team: The project is built by one person (Raghavendra Bhardwaj), which raises questions about scalability and long-term maintenance.
• No revenue or customer data: There is no evidence of monetization, adoption, or traction.
• Unproven AI model performance: GPT-5.6 is used for structured analysis, but the description does not include any validation or accuracy metrics.
• Limited scope in current version: The tool is described as a prototype with future features planned, suggesting it’s not yet production-ready.
• Contextual continuity challenges: The author notes that continuity is contextual and hard to reason about — this may be a technical limitation.

1. What specific generative video platforms or APIs does FrameKeeper plan to integrate with?
2. How does the tool handle edge cases in continuity (e.g., when story rules are ambiguous)?
3. Is there any internal testing or feedback from users of generative video tools?
4. What is the roadmap for moving from a prototype to a scalable product?
5. Are there any partnerships or early adopters in the generative video space?
6. How does FrameKeeper ensure that correction prompts are accurate and actionable?