TheFluxTrain

You describe the film or promo you want, and Director Zero builds the scenes, connects the pipeline, and runs generation. You start from a brief rather than an empty canvas, which cuts the setup that normally eats a first day.

Flow Studio runs a film as a graph of nodes. A dedicated tab-group node folds repeated subgraphs, like the ten scenes of a story, into one bounded workspace, so shared references and prompts wire once instead of fanning into every copy. Nodes outside the group stay connected, so folding is for focus, not isolation.

Iteration sinks route each output into the next command automatically, and can loop as long as you need, which removes long chains of one-off nodes when you step through variations. Generations show up as a carousel inside the generator node, so you browse results and promote keeps without leaving the graph.

The platform grew out of LoRA training, and it still trains custom models: Flux 1, Flux 2, Qwen Image, and an LTX video LoRA. Training a model on your subject is the route to keeping a character or product steady across scenes, which is exactly where clip generators tend to drift.

Under the studio you can generate with a range of video models, which is what makes it more than a single AI video generator. The list includes Google Veo 3.1, Kling O3, Seedance, FLUX 3, and Gemini Omni Flash, plus image models like Nano Banana and GPT Image. Voices and audio add text to speech, voice changing, dubbing, and sound effects, so a cut can ship with sound.

Scenes land on a timeline you can actually edit. Trim, subtitle, and export a family short or a promo cut, and hand off the pipeline as JSON. Long-form continuity across a whole film is the part still being built, and the studio covers the shorter pieces today.

A CLI (tft) and a REST API let you call generation, Flow Studio, and video-editor endpoints from your own tools, with API keys managed in the browser. Flow Studio graphs save to your account, so the same flow opens in the browser or from the terminal.