AI companion guides

Choose character personalization when the voice or relationship context is unclear. Choose the conversation prompt collection when you have a persona but need a specific opening scene. For a new visual output, use the generator workflow before the visual brief examples. Each page owns one task so you can identify the next correction without mixing the entire workflow together.

The common method is to make one clear decision, inspect the result and correct the specific mismatch. A more elaborate profile or prompt is not automatically a better one. If the output contradicts your brief, a short record of what happened is more useful than changing every detail and hoping the next variation works.

Use the personalization guide when the mismatch belongs to the character rather than the current scene. A formal voice in one reply and casual banter in another may reflect conflicting style instructions, a changing context or the generation itself. The first useful question is whether the stable brief requests both styles. Editing the biography is reasonable when the conflict is there; rewriting an image prompt would address a different task.

Do not infer an orientation or personality from an illustration. Those belong in the fictional character note. The personalization guide helps separate identity, interests, voice and relationship context. If your character is already coherent but the first exchange has no direction, move to the opening-scene guide instead of adding more background.

Use the conversation prompts when the persona is defined but the exchange has no shared topic. An everyday activity gives the character a reason to respond: discussing a book, planning an afternoon or comparing a hobby project. The opening does not need to establish an entire relationship history. A small scene lets you inspect how the generated voice handles one difference of opinion or follow-up question.

The conversation examples include both new-acquaintance and established-relationship contexts. They keep the fiction non-graphic and leave the user’s actions open. Choose one situation, then correct the behavior you actually see. If the conversation repeatedly contradicts the profile, return to personalization rather than collecting a longer list of opening messages.

Use the generator workflow when you need to inspect and revise an output. A cropped face, duplicated accessory or merged figure is a concrete visual problem. Changing the mood language may produce another variation without resolving that problem. A small revision record identifies the brief, the visible error and the next change; it does not assume that image generation is deterministic.

The image workflow separates choosing the tool, checking the charge, writing the brief and inspecting the result. Read it before projecting the cost of a larger set of images. If your question is simply how to describe the first scene, the visual examples are a more direct starting point.

Use the visual prompt examples when you have not yet made a first request. Separate the adult character identities, setting, style and composition. A single portrait and a two-person romantic scene need different framing details, even when they belong to the same fictional world. Keep the first activity simple enough that an output can be compared with the brief.

The fictional-adult visual briefs describe distinct characters and non-graphic everyday scenes. They do not use real-person likenesses or private photographs. If an output later fails the frame, move to the revision workflow; if the conversation voice is wrong, move to the persona guide. Each route keeps the correction attached to the result it can actually influence.

A character profile establishes identity, background and speaking style. A conversation opening establishes the current situation. A visual brief establishes the people, setting, image style and composition. Putting all three into one paragraph makes it harder to diagnose errors. Keep a short stable adult character note and write the changing scene separately.

A first conversation or image project should be small enough to evaluate. Decide which fact matters most: a recognizable voice, clear agency, a usable frame or a transparent operation cost. If the experiment cannot answer that question, do not treat a pleasant output as proof of the missing capability. Keep unknowns visible and compare only the claims the evidence supports.

Our review directory contains dated access observations and limitations for all five subjects. It does not rank measured conversation quality, because no paid sessions were performed. The review method explains how a documented feature differs from a recorded result and why we publish factual statuses rather than unsupported ratings.