The composer resolves unset layers against your prompt, not against your previous episodes — so consistency has to come from what you specify and what you've saved, never from what the tool remembers. Save the cast, the locations and the look as named references, call them by handle in every episode, and set every layer rather than leaving any to Director's Choice.
A prompt-first tool is faster per episode and worse at remembering. At seventy episodes those two facts pull in opposite directions, and the whole method is about which one you let win.
What's the structural problem?
Director's Choice resolves against your prompt. Not against episode three.
That's the sentence to internalise. Any layer you leave unset gets filled by the engine reading what you wrote for this episode — which is exactly right for a one-off and exactly wrong for episode forty of a series that has to match.
There's no memory between episodes, and there shouldn't be. A tool that silently carried decisions forward would be unpredictable in a different and worse way.
So consistency isn't something the tool provides or fails to provide. It's something you supply, every episode, through references and specification.
The composer has no memory of episode three. Consistency is a thing you supply, not a thing it keeps.
What has to be locked before episode one?
Four things, and none of them can be retrofitted cheaply.
| Lock | How | Why before episode one |
|---|---|---|
| The cast | Save each actor with a handle. Call the handle in every prompt, never a description | A re-described character comes back slightly different each time, and across seventy episodes those variations compound into a visibly different person |
| The locations | Save each recurring place as a location reference with a handle | The apartment your series lives in should be a handle, not a paragraph you retype and accidentally vary |
| The look | Save the visual treatment as a style reference | Episodes generated weeks apart drift otherwise, and the drift reads as a production problem rather than a choice |
| The aspect ratio | Set it on the first generation | Six options are available, but changing it mid-series means two versions of your series rather than one |
Source: Hexcoded Marketing Studio documentation and reference library, checked September 2026.
Where this falls short. Locking the cast and the look constrains the series in ways you may later want to change. A mid-series haircut means regenerating the reference rather than editing seventy prompts — which is the cheaper version of that problem, but it's still a decision you're committing to early.
Should you set every layer?
For a series, yes. All five.
The five layers are product, actor, format, hook and setting. For a one-off, leaving some unset is genuinely useful — the engine makes choices you wouldn't and that tells you what your prompt implies.
For a series that's the wrong trade, because every unset layer is a place episode forty can diverge from episode three. You're not gaining exploration; you're gaining variance in a format where variance between episodes reads as inconsistency rather than range.
The exception is format, which is worth setting for a different reason: it determines the whole shape of the piece, so a series where it varies isn't one series.
For a one-off, an unset layer is exploration. For episode forty, it's drift.
What does the cost structure reward?
Specificity, and it's free.
Only three things drive credits: length, which is proportional; resolution, which rises by tier; and whether you select an actor, which adds a small flat fee for a real human creator. Prompt length, custom layer phrases, the format, hook and setting selections and inline asset tagging all cost nothing.
So there's no economic reason to write short prompts or leave layers unset on a series. The detailed, fully-specified approach that consistency requires is also the free one.
What does cost money is length and resolution — which means the efficient pattern is working the episode out at fifteen seconds and a lower tier, then generating the keeper at what you need.
What about the episodes themselves?
The format has settled on conventions, and they apply regardless of which tool you generate in.
Episodes run thirty to ninety seconds, from scripts of roughly four hundred to six hundred words. Dialogue stays under twelve words a line. The first ten seconds re-anchor rather than establish, because a returning viewer needs reminding where they were rather than introducing to the premise.
Then one forward move per episode — a revelation, a confrontation or a decision. One. And end in suspension rather than resolution, because resolution is what makes a viewer stop watching.
The composer's four-to-thirty-second range sits inside that comfortably. A thirty-second episode is a single generation.
How do you actually run seventy of them?
The work moves out of generation and into preparation, and that's the whole efficiency gain.
Episode two is faster than episode one because everything you named persists. By episode ten the references are doing most of the consistency work and you're writing plot rather than describing a world.
What stays constant per episode is the prompt structure. Same layers, same handles, same style reference — with only the scene and the beat changing. That repetition is the point rather than a limitation; it's what makes episode forty match episode three.
Where this falls short. How much faster it gets depends on your shot count and your tolerance for re-rolls. We haven't measured it and we're not going to publish an estimate. Anyone quoting you hours per episode is describing their own tool stack.
What's the actual sequence?
Six steps, and the first three happen once.
Write the series bible
Characters, arcs, tone, world rules. The single source of truth everything else refers back to, including the rights record for every face.
Create and name the cast, locations and look
Each saved once with a handle. This is the only time you describe any of them, and it's the step that pays back seventy times.
Set the aspect ratio and lock the layer template
Decide the five layer values you'll reuse, and the aspect ratio. Changing either mid-series gives you two series rather than one.
Write the episode, changing only the scene and the beat
Same handles, same layers, same style. One forward move, ending in suspension. The repetition is what holds the series together.
Run Plan First before generating
Free, and on a series it's the check that catches a drifting look before you've paid for it.
Iterate short and low, then generate the keeper
Length and resolution are the only cost drivers that scale. Work the episode out cheaply, then render it properly.
When should you use Creative Studio instead?
When an episode needs shot-level control rather than a described scene.
The composer generates from a description. That's fast and it's the right tool for most episodes in a series where the beats are conversational and the coverage is simple.
Where it isn't enough: an episode with a specific shot sequence, a long continuous take, a beat that needs a particular model's duration, or anything requiring Edit or Extend. Those are shot-level jobs.
Mixing is normal rather than a compromise. The library and the credit balance are shared, so building the difficult episodes shot by shot and composing the rest is one workflow.
- Director's Choice resolves against your prompt, not your last episode. Consistency is yours to supply
- Lock four things before episode one: the cast, the locations, the look, the aspect ratio
- Save each as a named reference and call the handle. Never re-describe a character across seventy episodes
- Set all five layers on a series. An unset layer is exploration on a one-off and drift on episode forty
- Specificity is free. Only length, resolution and actor selection cost credits
- Keep the prompt structure constant and change only the scene and the beat
- Run Plan First every episode. Free, and it catches a drifting look before you pay for it
- Mix tools. Compose the simple episodes, build the difficult ones shot by shot
No. Any layer you leave unset resolves against the prompt for that episode rather than against earlier ones. So consistency across a series has to come from saved references and explicit specification, never from what the tool retains.
Four things: the cast, saved as actors with handles; recurring locations, saved as location references; the visual treatment, saved as a style reference; and the aspect ratio, set on the first generation. Each is expensive to retrofit and cheap to do first.
Yes, all five. Leaving layers unset is useful for a one-off, because the engine makes choices you wouldn't. On a series every unset layer is a place episode forty can diverge from episode three.
No. Prompt length, custom layer phrases and layer selections all cost nothing. Only length, resolution and selecting a real human actor affect credits — so the fully specified approach a series needs is also the free one.
Thirty to ninety seconds, from scripts of roughly four hundred to six hundred words, with dialogue under twelve words a line. The composer's four-to-thirty-second range covers a short episode in a single generation.
When an episode needs shot-level control — a specific shot sequence, a long continuous take, a particular model's duration, or anything using Edit or Extend. The library and credit balance are shared, so mixing the two across a series is one workflow rather than two.
Seventy episodes, one cast list
Save your actors, locations and look with handles, then call them in every episode. Plan First shows you the storyboard before it spends a credit — and prompt detail costs nothing at all.
Start a seriesMore on tools, features and how the platform fits together in Product.