Most AI roleplay forgets what your character looks like the moment the scene changes. LumaBrowser's Roleplay mode doesn't. A local language model writes the story; a local image model paints each beat, keeping the same face, remembering the outfit, and framing the shot like a scene from a film. It all runs on your own machine, so nothing you write or generate ever leaves it.
Entirely on your own hardware. Both the writing and the images are produced by local models running on your GPU. There's no roleplay service in the cloud, no chat logs on someone else's server, and no per-message or per-image billing. Your scenes are yours alone.
Anyone can generate a pretty picture once. The difficulty in illustrated roleplay is making the same character recognizable across dozens of images as the pose, outfit, mood, and setting all change. Roleplay mode is built around that problem.
Each character is pinned to a stable seed and a reference portrait. When a new moment needs a different pose, the pipeline keeps the crisp, established face and composites it onto the new figure, instead of re-rolling from scratch and getting a stranger.
A wardrobe pass decides what each present character is actually wearing for the place and occasion, and tracks hats, glasses, and masks as separate items, so an edit that changes an expression doesn't accidentally strip an accessory.
A director pass chooses the pose, the framing (a wide establishing shot or a close-up), and where the focus lands, then stages up to two characters into the scene together with shared, believable lighting.
An expression catalog across many categories drives the character's face. The mood is repainted with a feathered mask over just the face, so the expression changes while the rest of the character holds steady.
Roleplay is a chat mode. You set up the scenario once, then just write, the illustration happens on every turn without you leaving the conversation.
Pick "Roleplay" in the chat mode picker and fill in the scenario, characters, and settings in a quick setup form.
Roleplay in plain prose. The local model narrates back, moving the story and the characters forward.
After each turn, the wardrobe and director passes interpret the text, deciding outfit, pose, expression, and framing.
A reaction image is painted and posted inline with the reply, with a live preview as the shot is staged and refined.
Instead of a spinner, Roleplay mode shows its work: the backdrop lays in, the characters are posed and dressed, then the moment is painted and the lighting matched, all narrated as it happens.
Reading the scene…Staging the shot…Dressing Mara in a field jacket…Painting the moment…Matching the light…Roleplay mode builds on LumaBrowser's local AI stack, so it appears only once both pieces are in place.
The narration runs on your own local LLM via the built-in one-click setup, no API key, nothing sent to a provider.
The art is produced by local image generation. Roleplay pairs a base illustration model with an editing model that places and harmonizes characters into scenes.
Roleplay ships as an optional add-on rather than a bundled feature. Enable it from the Extensions area, and "Roleplay" shows up in the chat mode picker once your models are ready.
Mature content possible notice. Roleplay mode is uncensored: it applies no content filter of its own, and what it can depict depends entirely on the local image model you choose. It is intended for adults, and the image pipeline is constrained to adult character proportions. You are responsible for the models you load and the content you generate on your own machine.
Immersive, illustrated roleplay with characters that hold together, running privately on your own GPU with no service in the loop. LumaBrowser is free; add the Roleplay mode when you're ready. Set up Local AI and Local Image Generation first, or head home to see everything it does.