I use Expand a lot, and it may be my favorite feature in Sudowrite. For the kind of drafting I do, I find it more useful than any of the Rewrite options.
My typical use case is taking a rough chapter or scene that has all the correct story beats but still feels flat—too much talking heads, summarized emotion, or scene direction—and using Expand to add the missing layers. With Muse 1.5 especially, it can turn that rough scene skeleton into something much closer to a polished chapter. I still need to revise the result into my own voice, but it gives me strong material to work with and is excellent for getting past writer’s block.
Expand is valuable enough that I could see myself using Sudowrite primarily for this feature if the boundary behavior were dialed in.
The main problem is that Expand does not reliably respect the narrative boundaries of the highlighted text.
In my testing, medium and larger selections often produce around 1,500 words of output regardless of the amount selected:
949 words highlighted → 1,520 words output → 21,938 tokens
245 words highlighted → 1,549 words output → 6,278 tokens
19 words highlighted → 344 words output → 3,305 tokens
So it clearly is not always targeting exactly 1,500 words, but the amount of expansion does not seem to scale naturally with the selected material. This creates two opposite problems.
With shorter selections, Expand can reach the final beat of the highlighted text and then continue inventing what happens next. Because it does not appear to consider the following chapter, that new material can take the story in a completely different direction from what is already written next.
With longer selections, the prose can be excellent, but the generation simply stops around the output limit while it is still partway through the selected scene.
Trying to solve this by highlighting several smaller sections and expanding them separately does not work very well either, because each section can continue beyond its ending beat. The pieces then overlap or introduce events that were never supposed to happen.
What I would really like is for the highlighted text to function as a narrative container.
Expand should essentially follow this instruction:
Start with the first beat represented in the highlighted text. Expand the lived experience between the existing beats. Preserve the events and direction of the selected material. Reach the final beat represented in the selection and stop there. Do not continue the story beyond that ending beat.
Even better would be an optional way to specify a Start Beat and End Beat, with the End Beat treated as a hard boundary. For example:
Start Beat: Main character arrives at the storage unit.
End Beat: Main character finds the old key.
Expand everything between those beats, but do not continue beyond the discovery of the key.
I don’t necessarily need Expand to produce a particular number of words. In fact, I think output length should be subordinate to covering the selected material. A 250-word scene skeleton might need 800 or 1,200 words. A 950-word rough scene might need 2,000 or more. A tiny selection may only need a few hundred.
A Light / Medium / Deep expansion setting would be useful, but the more important feature is simply: fully expand the selected material without going beyond its ending beat.
I have also tried using Key Details under Write Settings to explicitly tell the model what beat it must end on. That does not seem to reliably constrain Expand; Muse will still continue beyond the ending beat. If Key Details is supposed to apply to Expand, there may also be a bug or limitation there.
My Write settings are currently set to one card at approximately 50 words, so Expand appears to use its own generation settings. I’m not suggesting it necessarily needs to inherit the Write settings—I’d actually prefer dedicated Expand controls if possible.
The prose quality is already there. That’s why I’m requesting this. Expand is very good at taking a flat rough draft and adding the texture, interiority, description, and connective material that makes it feel like a real chapter. What prevents me from relying on it more heavily is that I can’t trust it to stop where my selected scene stops.
If Expand could treat the selected ending as a hard story boundary, it would become dramatically more useful for chapter-level fiction workflows.
