How to read this
Three buckets rather than quarters. Now is being built at this moment. Next is decided and unstarted. Later is wanted, unscheduled, and honest about being unscheduled.
Nothing moves from Next to Now because time passed. It moves because the thing before it finished, or because enough people told us the order was wrong.
Now
Being built at the moment.
Consistency across longer sequences
Holding a character and a place across more shots than a single scene. This is the hardest problem in the product, most of the research effort goes here, and it is the one thing that would most change what the tool can be used for.
Provider breadth
More models behind the same adapter interface, so a node gains options rather than the graph gaining special cases. Each one includes evaluation against real user cases before it appears as a choice.
Collaboration
Shared projects that behave correctly when two people edit the same graph — including what happens when they disagree, which is the part that makes this genuinely difficult rather than merely fiddly.
Next
Decided, not started.
Creator revenue share
Cash alongside the credit rewards that already exist. Terms published in full — split, threshold, schedule, what happens to accrued earnings — before anyone is asked to opt in.
Deeper spatial output
Scenes that survive being viewed from more than one position, and a delivery format a headset can read.
Workflow versioning
Published workflows that can be updated without breaking the graphs people built on top of them, which is currently the reason publishing feels riskier than it should.
Timeline interchange
Getting a sequence out to an editorial tool. Requested often enough by production teams that it has earned a place here.
Later
Wanted, unscheduled, and honest about it.
On-premise deployment
For teams whose material cannot leave their network — studios under a publisher security requirement, institutions under policy. A real request and a large piece of work.
Plugin surface
Third-party node types, once the node contract is stable enough that publishing one is not a promise we are going to break in six months.
Mobile review
Not authoring — a way to look at and comment on a sequence from a phone, which is where a lot of client feedback actually happens.
Not planned
Asked for often enough to be worth answering here rather than individually.
- Training our own foundation model
- Not our advantage and not our thesis. We route, evaluate and build structure. A worse in-house model behind a nicer interface would help nobody.
- A social feed
- Explore exists to find workflows worth running. Turning it into a place to accumulate followers would change what gets published, and not for the better.
- An asset marketplace
- Selling generated images is a different business with different incentives, and the incentives are the problem — it rewards volume over usefulness.
- Character or persona chat
- Not this product.
How this list changes
It moves when we learn something, and we usually learn it because somebody told us. The Discord and the GitHub org are where that happens. What breaks for people in public tends to move up this list faster than what is requested politely in private, which is not a policy so much as an observation about evidence.
The most useful thing you can send is not a feature request. It is the thing that stopped you, with what you were trying to do when it stopped you.