Agent One supports 7-minute short-film workflow with 50-minute director walkthrough
Creators published a 7-minute AI short made in 3 days with Agent One, then released a 50-minute walkthrough showing the shot-by-shot directing process. The update matters because it turns Agent One from a feature claim into a reproducible filmmaking workflow, though the evidence still comes from tutorial-style posts rather than broad user adoption.
TL;DR
- minchoi's post framed the strongest proof point here: a 7-minute AI short made in 3 days with Agent One, plus a linked director walkthrough that turns the claim into a repeatable process.
- minchoi's tutorial post and MayorKingAI's repost point to the same roughly 50-minute breakdown, covering setup, direction, and fixes instead of only showing a polished final cut.
- The workflow pitch matches what MayorKingAI's note on re-explaining emphasized, persistent context for characters, style, and worldbuilding, which Invideo's own Enterprise page describes as a "Context Engine" with persistent memory.
- Invideo's public product pages already position Agent One as a directing layer, not just a prompt box: the Kling page says you describe a vision, then Agent One writes the script, picks visuals, and iterates.
- The evidence is still tutorial-led, not ecosystem-wide. CharaspowerAI's animation post and MayorKingAI's repost add excitement, but they do not yet amount to broad adoption data.
You can watch the 7-minute short, jump into the full director walkthrough, and cross-check the memory claim on Invideo's enterprise page. The product pages also spell out the agent framing in plain language on Invideo's Kling integration page and the company's pricing page, which says the v4 agent can generate up to 30 minutes from a single prompt.
The 50-minute walkthrough is the real story
The finished short is useful evidence, but the more interesting artifact is the walkthrough. minchoi's tutorial post is a process document in video form, and MayorKingAI's repost describes it as a raw 50-minute run from setup to direction to fixes.
That changes the shape of the claim. Instead of "here's a cool AI film," the evidence becomes "here is the shot-by-shot directing loop someone actually used."
The clips in the evidence suggest three concrete stages:
- setup of the story and visual brief, per MayorKingAI's repost
- direction and iterative adjustment, per minchoi's tutorial post
- fixes after the initial cut, again per MayorKingAI's repost
Memory is the feature every post keeps circling
The most specific user complaint in the evidence is not render quality. MayorKingAI's note on re-explaining says the pain is having to restate characters, style, world, and references every time a creator switches tools or retries a scene.
That same post claims Agent One lets you brief once and keep directing from there. MayorKingAI's repost repeats the same angle for campaign consistency, while Invideo's Enterprise page describes a "Context Engine" with persistent memory for a brand, brief, or vision.
For AI filmmakers, that is the practical reveal in this batch of posts: continuity is being sold as retained context, not only as better image quality.
Agent One is being positioned as a director layer
Invideo's own language lines up with the demo. On the Kling page, the company says creators can describe a vision to Agent One, after which it writes the script, picks visuals, and keeps iterating until the video matches the brief.
That is much closer to directing than prompt-by-prompt clip generation. The same page pairs Agent One with multi-shot sequences and AI-powered directing, which fits the way minchoi's short-film post framed the short as something directed shot by shot over three days.
The evidence still sits in the tutorial lane. There is no hard production data here on failure rates, cost per finished minute, or how often the agent holds continuity without manual cleanup.
The stack behind the demo is already broader than one model
Invideo is not presenting Agent One as a single-model magic trick. Its pricing page says paid plans include access to 200-plus image, video, audio, and music models, and that the v4 agent can create up to 30 minutes of video from one prompt.
The AI models page lists a wide menu behind that claim, including Veo, Sora, Kling, Seedance, Pixverse, Wan, and multiple upscalers. Separately, OpenAI's customer story on Invideo AI says the product is built on GPT-4.1, gpt-image-1, and text-to-speech models, with OpenAI o3 handling planning and orchestration.
That makes the short-film demo more interesting than a one-off showcase. The visible output is a film tutorial, but the underlying product pitch is an orchestration layer sitting on top of a rotating model catalog.