Pose Library: 30 Reusable JSONs for Influencer Content

I rewrote the same pose prompt for "casual mirror selfie, phone in hand, soft window light" about forty times before I admitted I needed a library. Forty times. Each one was slightly different because I would forget which version had worked best, and I would tweak the wording, and then I would have to test the new version, and the whole loop ate hours every week. The AI character pose library is the answer to that loop, and once I built mine I never went back.

The trick is not the prompts themselves. The trick is treating each pose as a structured object you can drop into any persona's workflow without thinking. A JSON entry per pose, thirty entries for the categories that cover most influencer content, and the locked character executes. That is the library. Below is the structure, the curation logic, and the actual thirty entries I use.

Quick Answer: An AI character pose library is a JSON file where each entry encodes one pose as a reusable scene clause with framing, pose verbs, prop list, eye direction, and a notes field. Thirty entries split across selfie, mirror, action, editorial, outfit-of-the-day, and travel categories cover ninety percent of AI influencer content needs. You drop the JSON into your prompt pipeline, the locked character renders the pose, and you stop rewriting the same descriptions from scratch.

The Pose Vocabulary Problem and Why a Library Solves It

Here is the thing nobody tells you when you start making AI influencer content. The hard part is not the model. The hard part is keeping your pose vocabulary consistent across hundreds of generations without losing your mind.

You write a prompt. It works. Three weeks later you want a similar pose for a different scene. You cannot remember the exact wording that worked, so you write a new one. The new one is close but not quite, and you spend twenty minutes fiddling. Multiply this by every post you make and you have hours per week burned on rewriting things you already figured out once.

The library solves this by giving every pose a stable address. "Casual mirror selfie" is no longer a phrase you reconstruct. It is a JSON entry with a fixed prompt fragment and a notes field reminding you what works and what does not. You call the entry by name, the prompt fragment gets inserted into the larger prompt, and you generate.

In my own workflow, I tracked the time spent on pose prompt writing before and after I built the library. Before, I was spending about two hours a week just on pose prompts. After, it dropped to maybe fifteen minutes a week, mostly when I add a new pose to the library or update an existing one. That is roughly a hundred hours a year saved on a task that does not produce creative value. The library is one of those tools that does not feel exciting until you live with it for a month and realize how much friction it removes.

The other thing the library does is make pose decisions visible. When you have to write a fresh prompt every time, you tend to default to whatever you wrote last. The library forces you to look at the full menu and pick consciously, which means your content variety goes up just from the act of choosing from a list instead of typing from memory.

Thirty Categories: Selfie, Mirror, Action, Lifestyle, Editorial, Outfit-of-the-Day, Travel

Thirty pose entries is the number I landed on after trying fifteen, then forty-five, then fifty. Fifteen was too few because some scenes had only two options and you got repetition fast. Forty-five was too many because you spend more time browsing the library than you save. Thirty hits the sweet spot where each category has enough variety without becoming a chore to scroll through.

Here is how the thirty split across categories:

• Selfie poses: 5 entries
• Mirror poses: 5 entries
• Action and movement: 5 entries
• Editorial and magazine energy: 5 entries
• Outfit-of-the-day: 5 entries
• Travel and outdoor: 5 entries

Five per category is enough to never repeat within a week of posting. If you post once a day in a given category, you cycle the same five poses every week, but the lighting and outfit and background will be different, so the cycle is invisible to your audience.

The categories themselves were chosen by auditing what actually gets posted. I looked at fifty AI influencer accounts and tracked the pose categories they used most. Selfies, mirror shots, and outfit posts dominated. Travel and editorial showed up consistently but less frequently. Action shots were rare but high-engagement when done well. That audit shaped the split. If your niche is fitness, you would probably want more action and fewer editorial. If your niche is fashion, you want more outfit-of-the-day and editorial. The library structure scales to any niche.

I also bucketed each pose by face visibility, because that affects which consistency tools you need. Full-face poses need stronger persona reinforcement. Side-profile and back poses are more forgiving. The library notes field tracks this so you know which entries to run with extra reference weight.

JSON Schema for a Pose Entry

The schema for a single pose entry has five fields. I tried more. The extra fields were noise. Five is the minimum that captures enough information to be useful without becoming a chore to fill in.

Here is the schema:

{
  "id": "selfie-01-handheld-phone-soft-light",
  "category": "selfie",
  "framing": "medium closeup, phone slightly above eye level, three-quarter angle",
  "pose": "right hand holding phone, left hand near face or hair, slight smile, eyes on camera",
  "eye_direction": "looking into the lens",
  "props": ["phone in right hand", "optional coffee cup in left hand"],
  "notes": "Works best with soft window light. Avoid full harsh sun. Body weight on the back leg."
}

The id is descriptive enough to read at a glance. Category lets you filter the library. Framing covers the camera setup. Pose covers what the body is doing. Eye direction is its own field because models handle eye contact instructions specifically when called out. Props is an array so you can branch on prop availability. Notes is where you record what works and what fails, so the library teaches you over time.

When you drop the entry into a prompt, you concatenate the relevant fields into the scene clause. A typical render looks like:

"[Fixed hero clause]. [Framing]. [Pose]. [Eye direction]. [Props joined with commas]. [Background and lighting clause]."

The hero clause comes from your locked persona. The framing, pose, eye direction, and props come from the JSON entry. The background and lighting clause is the one variable part you write per post. Three locked fragments plus one variable fragment is the entire structure of a render.

For batch generation, I keep a small Python script that reads my pose library JSON, picks N entries by category, joins them with my hero clause and a background variant, and writes out the full prompts to a file. That file feeds my batch render. I can produce fifty post-ready prompts in about ten minutes of selection plus a few seconds of script runtime. The same task without the library used to take me a full afternoon.

Five Selfie Variations Worth Saving

Selfies are the bread and butter of influencer content. Five well-curated selfie entries cover almost every selfie context you would post.

Selfie one is the classic handheld phone selfie with eye contact. Phone slightly above eye level, three-quarter angle, the hand holding the phone partially visible in frame. This is the "good morning" post, the "check this out" post, the casual greeting. Soft front-side light works best. Avoid hard overhead sun, it makes the under-eye area look harsh.

Selfie two is the chin-up confident selfie. Camera at chest level looking up slightly, chin raised, slight smile, eyes either at camera or off into the middle distance. This is the "today's outfit hits" energy. Works well with strong overhead light or a top-down side light. The pose reads confident without being aggressive.

Selfie three is the candid laugh. Face turned slightly off camera, mouth open in a real laugh, hand near face or hair. Hardest to generate cleanly because real laughs vary, but the locked persona helps. Use it sparingly for high-engagement candid posts. Notes field reminds you to add "natural laugh, eyes crinkled" to push the model toward a believable expression.

Selfie four is the close eye-contact serious. Tight on the face, eyes locked on camera, neutral or slight pursed lip expression. This is the "we need to talk" post or the editorial-feel close-in shot. Stronger persona reference weight required because tight face shots drift the easiest. I covered tight-face weight tuning in detail in my IPAdapter FaceID v2 weight tuning guide.

Selfie five is the over-the-shoulder selfie with environment context. Phone held to the side, looking back over the shoulder at the lens, background visible behind the head. Used for "where I am" posts. The trick is the background prompt. The pose itself is short, the environment carries the post.

Five Mirror-Pose Variations Worth Saving

Mirror shots add a layer of complexity because the model has to render both the person and the reflection, and the reflection needs to feel believable. Five mirror entries cover the common contexts.

Mirror one is the standard outfit-check. Standing in front of a full-length mirror, phone held at chest height pointing toward the mirror, body slightly angled, weight on back leg. Reflects the full outfit cleanly. Works for the everyday wardrobe post.

Mirror two is the bathroom mirror selfie. Camera slightly tilted, hand on the counter or holding a product, eyes on the camera or in the mirror. This is the morning routine post or the getting-ready post. Bathroom lighting tends to be overhead and warm, which softens features.

Mirror three is the angled mirror with one foot forward. Body turned three-quarters to the mirror, one foot in front of the other, hip slightly cocked. This is the fashion-pose mirror shot. Reads more editorial than casual. Use it when the outfit is the focal point.

Mirror four is the side-by-side mirror where the figure stands beside a vertical surface, half profile to camera, half profile to mirror. Splits attention between the real subject and reflection. Used for "two angles, one outfit" content.

Mirror five is the seated mirror pose, usually a vanity or low mirror, applying lipstick or fixing hair. The motion gives it candid energy. Generally easier on the persona because the face is at a forgiving angle and partially obscured by the action.

For mirror poses, the notes field reminds me that the reflection direction must be physically plausible. A right-handed phone hold reflects as left-handed in the mirror. The model sometimes gets this wrong. A note flag in the JSON keeps me from posting an obvious error.

Five Action Poses That Do Not Distort the Face

Action poses are where face drift happens fastest because the body is in motion and the model prioritizes pose execution over identity reinforcement. Five entries that I have tested to hold the face cleanly.

Action one is walking toward camera, mid-stride, arm swing visible, slight forward lean. The classic "main character walking down the street" pose. Stay around medium-full body framing. Closer than that and the motion blur the model adds can smear the face.

Action two is the casual leap or skip. Both feet off the ground at the peak of the motion, arms slightly out for balance, smile or open mouth. Higher risk of face drift. Use it sparingly when the post warrants the energy. The fix is to keep framing slightly wider so the face is not the focal point at full pixel density.

Action three is the seated lean-in. Sitting on a step or low wall, elbows on knees, leaning forward, eyes on camera. Reads engaged and conversational. Body weight forward gives it momentum even though the figure is static. Easy on the face because the head is steady.

Action four is the laughing turn. Walking or standing, head turned over the shoulder, laughing or smiling. The turn adds movement without putting the body in full motion. Face stays clean if the eye direction prompt is explicit. Note in the library says "eye direction must specify 'looking back over shoulder' or model will face camera."

Action five is the active sit, one knee up, one foot flat, hands resting on the raised knee, looking ahead or at camera. This is the "thinking moment" pose with energy underneath. Reads thoughtful but engaged. Stable on the face because the upper body is steady.

The pattern across all five action poses is that the face is either steady or partially obscured by the motion. That is the trick. Big body motion plus a stable head position protects the locked identity. I covered the face protection strategies for pose-heavy generations in my character sheet workflow, and those same protections feed every entry in the action category.

Five Editorial Poses With Magazine Energy

Editorial poses are different from selfies because the camera is at a deliberate distance, the framing has compositional intent, and the model is not engaging the camera the same way. Five editorial entries that hold up.

Editorial one is the three-quarter standing portrait, looking off camera into middle distance, body angled, weight on back leg, hands relaxed at sides or one in a pocket. The classic magazine cover pose. Works at three-quarter to full body framing.

Editorial two is the seated editorial, often on a step, chair, or low ledge. Body folded into a clean shape, hands resting on knees or thighs, gaze either at camera or off. Cleaner than standing because the model has fewer body parts to track.

Editorial three is the leaning editorial. Leaning against a wall, doorframe, or column, body angled, one foot crossed over the other or one foot flat against the surface behind. Looks effortless but the pose communicates confidence and presence.

Editorial four is the walking editorial in motion, but at editorial pace, which means slower and more deliberate. Eyes off camera, focused on something past the lens, arms relaxed. Reads cinematic rather than casual. Wider framing than selfie so the figure takes about a third of the vertical frame.

Editorial five is the close intimate portrait, tight on the face and shoulders, hand resting against the jaw or near the hair, soft gaze either at camera or down. This is the moodiest of the editorials. Tight face shot, so reference weight needs to be high. Pair with editorial lighting from my lighting prompts guide for best results.

The editorial category is where the locked persona pays off most visibly. A drifting face in a casual selfie looks like a slightly different friend. A drifting face in an editorial close portrait looks broken. The library notes for editorial entries always include "use highest persona reference weight" as a reminder.

Five Outfit-Of-The-Day Poses

Outfit-of-the-day posts are about the clothes, but the pose still matters because it shapes how the outfit reads. Five poses optimized for showing wardrobe clearly.

Outfit one is the front-facing full body, feet shoulder-width apart, hands relaxed at sides, eyes at camera or slightly above. The classic "here is what I am wearing" pose. Clean and direct. Shows the full outfit without distortion.

Outfit two is the side profile full body, weight on back foot, slight forward lean, hands in pockets or relaxed at sides. Shows the outfit from the side, which is where shape and silhouette read most clearly. Used for outfits where the cut matters more than the front print.

Outfit three is the walking outfit, mid-stride, arms swinging naturally, eyes ahead. Shows the outfit in motion, which is how it actually looks when worn. Best for outfits with flow, like dresses, long coats, or wide-leg pants.

Outfit four is the seated outfit, perched on a step or low surface, legs at a flattering angle, hands on knees or beside hips. Often used for shoe-focused outfit posts because the seated pose puts the footwear at a natural focal point.

Outfit five is the detail-shot outfit, cropped to focus on a specific piece, like a closeup of shoes, or a half-body shot showing the jacket. Used for outfits where one piece is the hero and the rest is supporting. The framing is tighter than the full-body entries.

These five rotate cleanly across a posting week. Combine them with the five looks method for wardrobe identity and you have five wardrobe identities times five outfit poses, which gives you twenty-five distinct outfit-of-the-day post compositions before any repetition.

Five Travel and Outdoor Poses

Travel content is its own beast because the location is usually the star and the figure is the supporting element. Five travel poses that work across most location types.

Travel one is the wide environmental walk, figure smaller in frame, walking away from camera or across the scene, environment dominant. Used for sweeping landscape posts. Persona reference weight can be lower because face is not the focal point.

Travel two is the looking-at-view shot, figure with back to camera, hands at sides or one hand raised slightly, gazing at a vista. Iconic travel content shape. The mystery of the unseen face is part of the appeal. Easy on the persona because face is not generated.

Travel three is the candid travel selfie, phone held out, big smile, landmark in background. The "I am here" post. Standard framing rules apply, the location is the variable part of the prompt.

Travel four is the seated travel, on a bench, step, or ledge, looking out at the view or down at a map or coffee. Reads contemplative and lifestyle. Good for slower-pace travel content like coffee shop scenes or quiet morning posts.

Travel five is the action travel, walking or moving through the scene with purpose, often with a bag or luggage in frame. Reads journey-in-progress. Used for transit content like train stations, airports, or arrivals at a hotel.

Each travel entry has a notes field reminding me to write the location prompt as detailed as possible. The pose entry handles the figure. The location prompt handles the place. The model treats them as complementary, and the result reads like real travel photography instead of a figure pasted into a generic outdoor scene.

Importing the Library Into an Apatero Workspace

Building the library is one half. Using it efficiently is the other half. Without a way to call entries quickly, the JSON sits in a folder and you go back to typing prompts from scratch out of habit.

The way I use my library across tools varies depending on what I am running. For ComfyUI, I have a small node that reads the JSON file and exposes the pose entries as a dropdown, so I can pick a pose by id and the framing, pose, eye direction, and props fields get concatenated into the prompt automatically. For local quick generations, I run a small terminal script that prints the prompt fragment when I pass the id.

For Apatero AI, the import is built in. You drop the JSON file into the workspace and the pose library appears as a dropdown next to your locked persona. Pick the persona, pick the pose entry, add your background and lighting clause, generate. The friction goes to near zero. When I am batching a week of posts, I can knock out five outfit shots, five selfies, and five action shots in about twenty minutes of setup plus the generation runtime.

The other Apatero AI feature that matters here is library sharing across personas. If you run two or three different AI characters, you do not need a separate pose library per persona. The same thirty entries work with any locked persona because the pose JSON is independent of the identity. One library, multiple characters, no duplication.

For starter libraries, the community pose pack on Civitai is a reasonable starting point if you want to skip the curation step and edit a base library down to your needs. The community packs tend to be larger than thirty entries, but you can prune them to your actual content patterns within an hour. I started from a community pack myself and trimmed it down over a few weeks of usage.

The deeper version of this workflow connects the pose library to the captions and the posting schedule. I covered the captions side in my work on voice consistency for AI influencers, and combining a pose library with a voice profile gets you most of the way to a full content engine that runs on selection rather than typing.

Frequently Asked Questions

How Many Entries Should My Pose Library Have?

Thirty is the sweet spot. Fifteen is too few because some categories run out fast. Forty-five is too many because you spend more time browsing than generating. Aim for five entries per category and six categories that match your niche.

Can I Share a Pose Library Across Multiple AI Personas?

Yes, that is the point of structuring the library as a JSON of scene clauses rather than as a folder of images. The pose entry is identity-agnostic. Drop in any locked persona and the same pose entry works.

What Format Should the Library Be In?

JSON is the cleanest because it is structured and any tool can parse it. I have seen people use YAML, CSV, and Markdown tables. They all work. JSON wins on portability between tools and ease of programmatic use.

Do I Need to Train a LoRA Before I Can Use a Pose Library?

No. The pose library is independent of how you lock your character. It works with a trained LoRA, with IPAdapter FaceID v2, with Flux Kontext editing, or with a hosted persona on Apatero AI. Whatever holds your identity, the library fills in the pose.

How Long Does It Take to Build a Thirty-Entry Library From Scratch?

About four to six hours if you have an established niche. The bulk of the time is testing each entry to confirm the framing and pose verbs render cleanly. The JSON itself is quick to type once you know what works.

Should Action Poses Use a Lower or Higher Persona Reference Weight?

Higher, usually. Big body motion competes with identity reinforcement, so a slightly stronger reference helps the model maintain the face. Notes field in the library should flag which entries need the boost.

Can I Use a Pose Library With Midjourney or DALL-E?

Yes, but you lose some of the structured benefits because those tools do not have a clean reference image system the same way ComfyUI or Apatero AI do. The pose clauses still work as prompts, but you cannot lock the identity as tightly.

What Categories Should I Prioritize for a Fashion Niche?

Outfit-of-the-day, editorial, and mirror. Five entries each gives you fifteen poses that cover most fashion content. Add three action and two travel entries to round out the variety, and skip selfie and lifestyle for a stricter fashion focus.

Do Pose Libraries Work for Non-Human Characters Like Brand Mascots?

Yes, with modifications. Some pose entries assume human anatomy. Mascot characters need framing-focused entries rather than pose-verb-heavy entries. Build a separate mascot library with simpler entries that focus on framing and prop placement.

Is Apatero AI's Library Import Compatible With My Existing JSON?

If your JSON follows the schema in this article, yes. The fields map directly to the import dropdown structure. If your JSON is in a different format, the workspace has a converter that handles most common shapes, but the cleanest path is to use the documented schema from the start.