I Don't Render a Feature. I Queue 15–30 Second Scenes and Keep the Cast Locked

I wasted a week trying to get a "short AI film" out as one long render. That is not how the tool I use is built. What I do now is treat the film as a queue of short scenes, lock the cast before any motion is generated, and only then batch the clips.

This is the scene-by-scene film workflow I actually run. When I want a show that keeps going, rather than a film told in parts, I use the series side of the same idea instead. The two look similar in the UI. The unit of work is different.

The constraint I accepted

It cannot make a full-length film in one go. Each episode is one short scene of 15 or 30 seconds, or Auto to fit the script. A story can run across many episodes, but each one is its own clip. I also cannot join those scenes inside Fableclip, edit footage I shot, drop a real actor into the frame, or get a timeline for frame-by-frame changes. If I want a single file, I download the parts and join them in any video editor. If I don't, I post them as parts.

Once I stopped pretending otherwise, planning got easier. A short film still needs a setup, a turn, a crisis and an ending. Episode one sets up the situation and the central question. Each episode after that opens where the last one ended and moves the story one real step. When I'm ready to end the film, I say so in the brief or in the next episode's script.

Script and cast, in that order

I start with one sentence and a genre I actually mean: mystery, sci-fi, fantasy, horror, action adventure, or a family film. Then I keep each scene small: one place, two or three characters, one clear action. Crowded scenes are where consistency dies.

Character consistency is a data problem. Each main character gets a written look and a reference sheet, with one detail that makes them unmistakable, because that look is reused in every scene. I can upload my own reference picture or describe them and generate one. Guest faces can be written fresh.

With a free account I can design the film and read the first two episode scripts. Making the episodes needs a paid plan. I use that gate as a review step. If the second scene doesn't open on the first scene's last beat, I rewrite it. I do not generate my way out of a broken handoff.

Batching scenes, and what they cost

Plans start at $19 a month. At 720p, a 15-second scene uses 8 video slots and a 30-second one uses 18, and I see the exact number before each episode is made. If a scene only needs 15 seconds, I don't automatically pay for 30. If the script is tighter as Auto, I use Auto.

Default frame is portrait 9:16, for TikTok, Reels and Shorts. For a wider, cinematic frame I pick Landscape 16:9 under Frame and leave it there for the whole film. The scenes do not come with burned-in subtitles; I add captions in the platform's own tools when I post. Finished parts can be downloaded or posted to TikTok, YouTube, Instagram and the other channels I connect, on a schedule.

When I don't use the film queue

If the project is a show — same cast, new episode, maybe a continuing storyline, maybe a standalone joke or case each time — I don't force it into a film shape. That's a series: a brief for how episodes connect, and a storyline if it should continue. The film queue is for a story that is trying to end. The series queue is for a story that is trying to keep going.

The pipeline is boring on purpose: premise, locked cast, two scripts I can read for free, then a batch of 15- or 30-second scenes whose slot cost I can see in advance. It is not a feature-length button. It's a job queue with a hard length cap, and that's what makes it usable.