Skip to main content

Character image generation needs a scene-only mode (and the prompt enhancer breaks character prompts)

Feature RequestsImageCharacters

Category: Characters / Image Generation

Summary:
Generating images inside a character chat is unreliable because the app's auto-prompt-writer re-describes the character and the prompt enhancer rewrites everything in generic style language. The result: the character drifts, goes glossy/CG, and drops key details. There's no way to get a clean, consistent render without manually rewriting the prompt every single turn — which breaks roleplay immersion.

The problems, specifically:

    The enhancer overwrites character prompts. With "Automatically Enhance Prompt" on, a clean prompt like Star, standing in the room gets rewritten with "glowing, shimmering, hyper-detailed, cinematic" — which produces an over-glossy, semi-CG look, and drops details like hair colour.

    The auto-writer re-describes the character every time, lossily. It says things like "silvery hair" and "glowing shimmering skin," losing the actual details (violet/magenta streaks, cyan eyes, rainbow marks) and adding gloss words.

    No scene-only option. For image consistency, the ideal prompt is just what the character is doing — the uploaded character image should carry the look. Right now the text re-describes the character and fights the reference.

    Outfit drifts on bare prompts because clothing isn't carried from the character.

Requested changes:

    A "scene-only" prompt mode for character image generation — output only pose, action, camera angle, and setting; let the character image carry appearance.
    Use the uploaded character image as a reference/seed automatically so renders stay consistent to the character.
    An editable prompt template per character, so creators control what the auto-writer includes.
    Don't let the enhancer rewrite character-derived prompts (or make it opt-out per character).
    Carry the character's clothing/outfit into the prompt.

Evidence:

    Enhancer example: "a dude holding a beer" → a long paragraph adding "photorealistic cinematic, golden hour lighting, shallow depth of field, high-resolution photography style." Same happens to character prompts.
    Result: characters render glossy, inconsistent, and off-model compared to the uploaded reference.

Impact:
Character chat is a core feature, and image generation inside it is a major draw. Making it consistent and low-friction would make roleplay-with-images genuinely usable instead of requiring manual prompt engineering every turn.

1 comment

Log in to comment and vote

Comments1

  • Blue Macaroni

    •

    Oct 14

    1. Consistency across renders. Same character, same prompt → a different person each time

    Request: the character image should anchor identity so every render is recognisably the same person.

    2. The enhancer toggle is global-only. I had to turn it off everywhere to fix character prompts — which means i lose the enhancer for normal image gen too. Request: a per-character enhancer toggle, so creators don't have to choose.

    3. Pose/anatomy distortion. Complex poses come out broken — merged legs, distortions, hands in the wrong place,wrong genders. Request: better anatomy handling, or at least a camera-angle field.

    4. "Content Unsupported" false errors. NSFW character renders throw "content unsupported" randomly and need re-rolling. Request: better NSFW handling for character images.

        The enhancer "drops details like hair colour" — The enhancer adds style words.
        In my testing the auto-writer "re-describes every time"

    The specific behaviour i found — typing the character name in the prompt pulls the character image, but the button doesn't.