Concept work is volume work
A character needs turnarounds, expressions, outfit variants, weapon states and a silhouette that still reads at thumbnail size. An environment needs times of day, weather states and a mood pass for every major beat. A single character can generate forty images before anyone has agreed on the design.
All of it has to stay recognisably the same character and the same world. Generated independently, it will not. That is not a criticism of the models — it is what happens when nothing in the system records that image four and image thirty-one were supposed to be the same person.
The graph is built for exactly this shape: one identity node, many derived views, and a change to the identity propagating to all of them instead of starting the set again.
What teams use it for
- Character exploration
- Turnarounds and variants from one reference, so the set stays coherent while the design is still moving. The reference is what changes; the set follows.
- Environment mood
- Biome and lighting passes that share a framing, which makes them comparable rather than merely different — and comparable is what a direction review needs.
- Cinematic previs
- Beats boarded on the director stage, read as a sequence from the same graph the concept art came out of.
- Key art and marketing
- Explored against the same references the concept team is using, rather than diverging from them and needing to be reconciled later.
- Pitch material
- A vertical slice of the visual target, built early enough to inform whether the target is achievable.
Not asset production
If your studio has a policy on generated material in shipped assets — and increasingly they do, sometimes because a platform or a publisher requires it — this sits on the concept side of that line, which is usually where the policy permits it. Check yours rather than assuming, and note that some policies cover concept material too.
Style consistency across a large world is still hard. A graph gives you shared rule nodes, which is the best available answer, and it is not a complete one.
Questions from studios
Can concept artists actually use this, or does it replace them?
It is a tool for concept artists, and the ones who get the most out of it are the ones with the strongest fundamentals — because the useful part is deciding what to hold constant, which is an art direction skill rather than a prompting one.
Does it work for stylised art, or only realism?
Both, but style consistency is harder than realism consistency and needs a dedicated treatment node that everything inherits. Budget a session on establishing that before generating anything you intend to keep.
Can we keep our IP off third-party servers?
Not in this build — generation runs against hosted providers. On-premise deployment is on the roadmap under Later. If this is a hard requirement, say so; it is the request that moves it.
Is our concept work used for training?
No, not by us. Individual model providers have their own terms, and where one retains or trains on inputs we surface that rather than leaving you to find it in their documentation.
How does this fit with an existing art bible?
The art bible becomes rule nodes — palette, material vocabulary, silhouette language, scale reference. That is close to the best case for this tool, because the rules already exist and only need encoding.