Your GPU paints the character, and the world around them.
A character chat that is only text asks you to imagine everything. privstory runs an image model on your own machine alongside the chat model: it paints your character's portrait when they're created, and repaints the scene backdrop as the story moves. No credits, no queue, no uploads.
Two kinds of image, both generated locally
The first is the portrait. Describe a character in a sentence and the model fills in their persona and appearance, then paints a portrait to match. It becomes that character's face for every scene you open with them.
The second is the backdrop. A scene starts from an opening situation, and as the conversation moves somewhere new the backdrop is repainted to follow it. The visual doesn't stay a fixed illustration while the story walks away from it.
No credits, no queue, no upload
Hosted image generation is metered, because a GPU somewhere is rendering for you: a credit balance, a monthly image allowance, or a queue that free users wait in. Generate locally and none of that applies. There's no allowance to run out of and no queue but your own.
Nothing is uploaded either. The prompt derived from your character and scene, and the image that comes back, both exist only on your disk. Generated images are never scanned, moderated server-side, or used as training data, because they are never transmitted.
The module decides how images look
The image model isn't fixed by the app. A module bundles a chat model, an image model, and the prompts that drive character and scene creation, so switching modules changes the whole visual character of what gets painted.
Build your own or import someone else's, download its weights once, and pick it when you create a character. Each character stays bound to the module it was made with, so its look stays consistent across every scene.
What it needs from your machine
Image generation is the heavier of the two models, and it's what benefits most from a GPU. Apple Silicon uses Metal automatically; Windows and Linux use Vulkan, which covers NVIDIA, AMD, Intel and integrated GPUs from the same installer.
There is a CPU fallback and it does work, but image generation on CPU is slow. The chat and image models are never resident at the same time, so peak memory is set by the larger of the two rather than their sum.
Common questions.
How many images can I generate?
As many as your hardware will produce. There is no credit balance, no monthly allowance, and no per-image charge, because the rendering happens on your own GPU.
Are my generated images uploaded or moderated?
No. Both the prompt and the resulting image stay on your device. There is no server-side moderation pass, because there is no server involved in generating them.
Can I change the image model?
Yes, by choosing a different module. A module bundles the chat model, the image model, and the prompts that drive generation, so you can build or import one with the image model you prefer.
Do I need a powerful GPU?
A GPU is recommended, and image generation is the part that benefits most. Apple Silicon uses Metal, and Windows and Linux use Vulkan across NVIDIA, AMD and Intel. A CPU fallback works, but expect image generation to be slow.