The Problem
Making a short film — even a few seconds of consistent, photorealistic footage — used to mean hours or days of work, expensive tools, and a character that never quite looks the same twice. Every frame required manual attention. Nothing was connected end to end.
What We Built
FlixML is an end-to-end AI film studio that runs on your own GPUs. This is the part that matters: FlixML isn't a dashboard you click around in — it's built for an AI agent to operate. The agent calls one API to train, generate, animate, and render. It assembles scenes, sequences shots, and produces a finished cut on its own. You direct; the agent executes.
- Upload 15–25 reference photos — the agent trains a character LoRA using Flux.2, so the same face appears consistently in every frame
- The agent generates photorealistic images from script and scene descriptions, no manual prompting required
- Images are animated into video clips using Wan 2.2 I2V on your own GPUs — no cloud, no rate limits
- The agent assembles shots into scenes and scenes into a finished film, making editorial decisions on its own
The Facts
- Character consistency: Flux.2 LoRA trained on your photos, holds up frame to frame
- Image + video: photorealistic stills and animation on one pipeline
- Agent-ready: REST API + a single SKILL.md — no SDK required
- Self-hosted: multi-GPU routing, your infrastructure, your data
- Open source: MIT license — audit it, fork it, extend it
Why It Matters
FlixML is proof of how we build: AI agents doing real, end-to-end work — not demos, not chatbots that draft an email. A finished short film in minutes instead of days. If an agent can write, generate, animate, and render a film, imagine what one can do for your business.