Frequently asked questions
Is your question not here? Just ask it — we will answer it and add it to this list.
Can the existing context from ChatGPT be moved across?
Not as a file. There is no technical export of the context of an AI account to another model: what a model appears to have learned is the result of a chat history, the custom instructions and the files pasted in along the way. What can be done, and what is precisely why this application exists, is to take that knowledge out, make it explicit and store it neutrally so any model can be given it again. The import wizard helps: it reads the data export, groups the image prompts by subject and makes a proposal per character. Only what a human confirms becomes canon.
Which model handles fixed characters best?
On paper, the model that is alone in having a separate character reference category next to object and style references, because that is exactly the split the canon makes too. But paper is not a verdict: that is why the bake-off sits in the application. The same fixed regression set goes to several models, the results are judged blind, and only the cost per usable image gives the real answer. A model that is half the price but has to be redone three times as often is the more expensive one.
What does generating images and video cost?
An image costs a few cents to a few tens of cents per generation, depending on model and resolution. Video is an order of magnitude more expensive: a few seconds at high resolution quickly costs a couple of euros, and on average you throw more away than you keep. That is why the estimated price is on screen before you generate, why the application tracks a budget per project and per month, and why there is a per-user, per-day limit.
Can an AI approve an image by itself?
No. Automatic checking may reject and warn, but approving is always done by a person holding the Art Director role. That applies to agents in particular: no agent approves and no agent publishes to a production system. This is a hard design rule, not a setting.
May the generated imagery be used commercially?
That differs per provider and is therefore recorded per model profile: may the output be used commercially, is our input used to train the model, and which visible or invisible provenance marking does the provider put in the file. That marking is not removed, but it is recorded — a watermark in an image that gets printed on a product is something you would rather discover beforehand.
What happens to imagery that may not become public yet?
It carries the embargo flag, with a date and a reason. Material under embargo cannot be published to a shared folder, to an app or in a supplier package unless someone holding the Art Director role deliberately lifts the embargo with a reason. Material for a product that still has to be registered as a design carries the embargo by default.
Can several people work at the same time?
Yes, and that is one of the reasons the application exists. Roles decide who may do what, two people reviewing the same asset at once see each other's action, and a generation is a task in a queue so nobody has to wait on a running job. On top of that, agents can handle sub-tasks in parallel within a configured limit per provider.