Character Reference Sheet for AI Social Campaigns

A character reference sheet for AI social campaigns should document approved views of the character, physical identity, wardrobe, expressions, poses, scene families, voice notes, forbidden changes, disclosure requirements, and review criteria. It should be detailed enough that a reviewer can reject an off-character image without guessing.

AI characters drift when the team relies on memory or a single hero image. A model may change the face, age, proportions, hair, wardrobe, or expression while keeping the image attractive. That is not acceptable for a recurring brand persona, AI influencer, mascot, or product guide.

The reference sheet is not only for generation. It is for review. It lets writers, image generators, designers, and approvers compare every social post against the same standard before publishing.

The physical identity lock is the non-negotiable description of the character. It covers age range, face structure, hair, skin tone, body type, proportions, recognizable features, and realism level. The description should avoid resembling a real private person or public figure.

For a synthetic influencer, the lock might include face shape, eye shape, hair length, styling, body proportions, and recurring accessories. For a mascot, it might include silhouette, color patches, eye shape, material style, and animation posture. For a brand educator, it might include wardrobe and camera language.

The lock should be paired with images. Text alone is too ambiguous. Images alone are too easy to reinterpret. Together they give the generation and review process a stronger target.

1. 1

   Approved core views

   Front, three-quarter, side, full-body, closeup, and neutral expression views.

2. 2

   Recognizable features

   Hair, face structure, silhouette, accessories, proportions, color marks, or recurring wardrobe details.

3. 3

   Realism level

   Photoreal, editorial, illustrated, 3D, mascot, anime-inspired, or simplified vector style.

4. 4

   Forbidden similarity

   Do not make the character resemble a real person, public figure, customer, employee, or influencer without explicit rights and review.

Wardrobe and scene rules keep the character flexible without becoming random. A persona can appear in different contexts, but the outfit families and environments should match the account's niche. A productivity guide may rotate desk, commute, and planning scenes. A beauty persona may rotate bathroom, product shelf, and ingredient education scenes.

Do not give every post a new world. Three to five recurring scene families are usually enough for a month of content. The audience recognizes the character faster when the environment feels familiar.

Wardrobe should support the content job. Educational posts can use simpler outfits and cleaner backgrounds. Product reviews can use a recurring shelf or studio setup. Conversion posts can use a cleaner graphic layout with the character as guide, not decoration.

Expressions and poses carry personality. If the character looks serious in one post, childish in the next, seductive in another, and corporate in a fourth, the account loses continuity. The reference sheet should show approved emotional range.

Create an expression row: neutral, friendly, focused, excited, curious, skeptical, and celebratory. Create a pose row: speaking, pointing, holding product, using app, comparing options, reacting to a question, and presenting a CTA. Each pose should be appropriate for the brand and audience.

This matters for trust-sensitive categories. A health, finance, or education persona should not use exaggerated expressions that trivialize the topic. A playful consumer mascot can have a wider range, but it still needs boundaries.

1. 1

   Approved expression range

   Show what friendly, focused, excited, skeptical, and thoughtful look like for this character.

2. 2

   Approved pose range

   Show how the character presents, points, holds products, uses devices, and appears beside text.

3. 3

   Forbidden emotional range

   Ban expressions that feel too young, too sensual, too aggressive, too corporate, or too unrelated to the brand.

A character reference sheet should include trust rules, not just visual rules. TikTok provides AI-generated content disclosure guidance, Meta has introduced AI labels across Facebook, Instagram, and Threads, and FTC endorsement guidance matters when a persona promotes products or has a material brand relationship.

The practical rule is simple: viewers should not be confused about whether the character is synthetic, whether a recommendation is commercially influenced, or whether a shown event actually happened. Put that rule into the character system before content production scales.

For product-led characters, add commercial boundaries. What can the character recommend? What requires source support? What requires legal review? What disclosure phrase is used in captions, profile text, or post labels?

A character reference sheet becomes useful when it is used in review. Place all generated campaign assets in a grid beside the reference sheet. Look for face drift, body drift, wardrobe drift, scene drift, expression drift, and voice drift.

Review the batch as a sequence too. The character should feel like the same persona across education, proof, product, and CTA posts. If the visual identity holds but the voice changes from post to post, the campaign still fails continuity.

AttentionClaw can help teams turn approved character systems into repeatable carousels and slideshows. The reference sheet keeps the system from becoming a random image generator.

1. 1

   Identity review

   Compare face, body, style, wardrobe, and scene against the sheet.

2. 2

   Role review

   Check whether the character's expression, pose, and content role fit the post.

3. 3

   Trust review

   Check disclosure, claims, product recommendations, and source requirements.

4. 4

   Batch review

   Review the full campaign grid so continuity problems are visible before scheduling.

A character reference sheet created once and never updated becomes a liability rather than an asset. AI image generation models are updated, prompts drift over iterations, and campaign creative direction evolves. A sheet that was accurate six months ago may produce a noticeably different character today if it is not maintained.

A lightweight versioning workflow involves two practices: dating every approved asset with the prompt and model version used to generate it, and scheduling a quarterly review where a selection of new generated images is compared against the reference sheet's physical identity lock. If drift is visible — the character is consistently appearing with different facial structure, hair, or age range — the approved reference images may need to be regenerated and the prompt language tightened.

When a campaign runs across multiple months or involves multiple team members generating assets, a shared folder organized by approval status (in review, approved, rejected, archived) keeps everyone working from the same baseline. Rejected images should be retained with a note explaining why they failed the reference check — this builds a team understanding of the drift patterns that are most common for this specific character.

Using a consistent AI-generated character across social campaigns creates compliance obligations that vary by platform and use case. The reference sheet is the right place to document these obligations so that anyone producing campaign assets knows the rules without having to research them independently.

At minimum, the disclosure section of the reference sheet should specify: which platforms the character will appear on and what each platform's current AI disclosure requirement is for that content type, whether the character will appear in paid advertising (which typically triggers stricter rules than organic posts), and whether the character will be presented as a real person, a branded mascot, or clearly fictional. Each of these choices changes the disclosure approach.

A character presented as clearly fictional or as a branded mascot typically has simpler compliance requirements than one designed to appear as a real person. If the character is designed to look human and speak in first person about real experiences, the disclosure bar is higher and the stakes of non-disclosure are greater. The reference sheet should make this distinction explicit so that writers and designers do not inadvertently create content that crosses into deceptive presentation.