Character Creation Act 1: Image Prompts

<h1>Character Creation Act 1: Image Prompts (Making Your Bot Look Like <em>Them</em>)</h1> <p>MiocAI characters live or die on one thing: <strong>visual consistency</strong>. If your character’s face, hair, and body “reset” every time you generate an image, the roleplay starts feeling like you’re swapping actors mid-season. (baywatch..)</p> <p>This first act is a firm look into <strong>appearance prompts</strong>,how to write them, how to keep them short without losing detail, how to avoid the classic “why is it doing <em>that</em>” mistakes, and how to iterate fast using MiocAI’s built-in testing.</p> <p><img src="https://miocai.com/media/uploads/697b88aecd2d2_paste.png" alt="4 images of a consistent character" /></p> <hr /> <h2>0) Before We Start: The Non-Negotiables (So You Don’t Lose the Character)</h2> <p>MiocAI has a few hard rules that matter for image prompts:</p> <ul> <li><strong>No real people</strong> (no photoreal clones of streamers, celebrities, exes, politicians, etc.)</li> <li><strong>No minors / under-18 appearance</strong> (no “looks young,” “petite teen,” “loli,” etc.)</li> <li><strong>Public characters</strong> have stricter rules than private ones (especially around explicit avatars/opening messages).</li> <li><strong>Illegal content stays illegal</strong> even if you label it “fictional.”</li> </ul> <p>Keep your prompts clearly adult, fictional, and within platform boundaries,especially if you plan to publish.</p> <p><img src="https://miocai.com/media/uploads/697b8900e676b_paste.png" alt="we dont play about that" /></p> <hr /> <h2>1) The Mental Model: What Your Prompt Actually Does in MiocAI</h2> <p>A lot of people write prompts like they’re trying to describe the <em>entire image</em>. On MiocAI, that’s usually the wrong target.</p> <h3>MiocAI images are built from two layers:</h3> <ol> <li><strong>Appearance Prompt (Character Prompt)</strong> This is the “casting sheet.” It defines what the character looks like <em>every time</em>:</li> </ol> <ul> <li>hair, eyes, face features</li> <li>body type / silhouette</li> <li>outfit (or outfit system fields)</li> <li>signature accessories</li> </ul> <ol start="2"> <li><strong>Action / Scene Prompt (Context-Aware Image Prompt)</strong> This is generated from chat context (recent messages) and decides:</li> </ol> <ul> <li>pose, expression, actions</li> <li>framing/angle (sometimes)</li> <li>props relevant to the scene</li> </ul> <p><strong>Your appearance prompt should NOT try to do the action prompt’s job.</strong> If you cram “smirking, lying on bed, cinematic lighting, city skyline, holding sword” into the appearance prompt, you’re basically telling MiocAI:</p> <blockquote> <p>“Do this. Every time. Forever.”</p> </blockquote> <p>That’s how you end up with a character who’s always shot from the same angle or always holding the same object.</p> <p><img src="https://miocai.com/media/uploads/697b898f67972_paste.png" alt="action vs character prompt" /></p> <hr /> <h2>2) Studio vs Built-In Avatar Generation: Start With a Strong Base Image</h2> <p>One of the fastest ways to get consistency is to start from a <strong>clean reference image</strong> you actually like.</p> <h3>Recommended workflow (especially for new characters)</h3> <ul> <li>Generate images in <strong>Studio</strong> (or use the prompt test images in the character editor).</li> <li>Pick a winner.</li> <li>Use that as the character’s avatar / wallpaper / reference.</li> </ul> <p>Why? Studio generation tends to be better for <strong>design exploration</strong> because you’re not fighting roleplay context. You’re just designing.</p> <p><img src="https://miocai.com/media/uploads/697b89e6eef6f_paste.png" alt="image studio" /></p> <h3>Important: Your uploaded avatar does <em>not</em> steer image generation</h3> <p>Uploading a profile picture is mostly cosmetic. <strong>MiocAI doesn’t use your avatar upload as a “reference image” during normal image generation.</strong> What controls appearance is the <strong>appearance/outfit prompt</strong> (and any outfit system you’ve set up).</p> <p>So if your character’s prompt says “short silver hair,” but your avatar has long black hair,the generator will follow the prompt, not the avatar.</p> <p><strong>Exception (tooling):</strong> Some flows let you turn an image into a prompt (image-to-prompt), which is useful. But that’s <em>converting the image into text</em>, not using the image directly as a reference every time. More on this when we actually implement this.</p> <p><img src="https://miocai.com/media/uploads/697b8a9272af5_paste.png" alt="you look nothing like your dating profile" /></p> <hr /> <h2>3) Finding the Prompt Editor (And What Those Toggles Actually Mean)</h2> <p>In the character creator/editor you’ll see:</p> <ul> <li><strong>Enable Images</strong> toggle</li> <li><strong>Style toggle</strong>: anime ↔ realistic</li> <li><strong>Outfit prompt system</strong> (structured 4-field mode)</li> <li><strong>Advanced model options</strong> (model picker)</li> <li>Prompt input + a <strong>Test Prompt</strong> button that generates multiple sample images</li> </ul> <p><img src="https://miocai.com/media/uploads/697b8ad968e78_paste.png" alt="the interface" /></p> <h3>Style toggle: Anime vs Realistic</h3> <p>Pick one and commit early. Switching style later often requires prompt cleanup because:</p> <ul> <li>tags that work well in anime may behave weirdly in realistic</li> <li>realistic tends to punish long prompts harder</li> </ul> <h3>Structured prompt mode (Outfit prompt system)</h3> <p>This splits your appearance prompt into:</p> <ul> <li><strong>Clothes</strong></li> <li><strong>Body</strong></li> <li><strong>Face</strong></li> <li><strong>Extras</strong> <em>(often reserved / may be limited depending on implementation)</em></li> </ul> <p>Pretty simple, and tt helps prevent your outfit and face details from stepping on each other.</p> <hr /> <h2>4) The Golden Rule: Your Prompt Is a Permanent Tattoo</h2> <p>If you put something in the appearance prompt, you’re asking for it to appear <strong>in every image</strong> unless the action prompt overrides it.</p> <p>That includes:</p> <ul> <li>props (“holding a microphone” → you’ll get a microphone addiction)</li> <li>framing (“full body” → you’ll keep getting full body)</li> <li>specific body parts (“bare feet” → surprise: feet forever)</li> <li>specific camera angles (“from behind” → enjoy backs)</li> </ul> <p><strong>So the question isn’t “Would this look cool?”</strong> It’s: <strong>“Do I want this in 90% of my images?”</strong></p> <p><img src="https://miocai.com/media/uploads/697b8b9f19a7f_paste.png" alt="are we being fucking fr" /></p> <p>Like what the fuck is that.</p> <hr /> <h2>5) Prompt Format: Tags Beat Sentences (Most of the Time)</h2> <p>You <em>can</em> write natural language, but MiocAI image models generally behave better with <strong>tag-style prompts</strong>:</p> <h3>Good pattern</h3> <ul> <li>comma-separated descriptors</li> <li>minimal filler words</li> <li>important identity traits early</li> </ul> <h3>Example formats</h3> <p><strong>Natural-ish (works sometimes):</strong></p> <blockquote> <p>“A tall woman with pale skin and short white hair wearing a black jacket.”</p> </blockquote> <p><strong>Tag-style (usually stronger):</strong></p> <blockquote> <p>“1girl, pale skin, short white hair, black jacket, confident expression”</p> </blockquote> <p>The tag version is clearer and tends to create fewer “interpretation detours.”</p> <p>Here an example with natural language:</p> <p><img src="https://miocai.com/media/uploads/697b8c4336bc3_paste.png" alt="all natural cows milk" /></p> <p>Versus with simpler tags:</p> <p><img src="https://miocai.com/media/uploads/697b8cc431422_paste.png" alt="bad example but eh" /></p> <hr /> <h2>6) Order Matters: Put “Identity” First, Put “Flavor” Later</h2> <p>Most models weight early tokens more strongly. So do this:</p> <h3>The recommended ordering</h3> <ol> <li><strong>Count / subject</strong>: <code>1girl</code>, <code>1boy</code>, <code>solo</code> (optional)</li> <li><strong>Identity anchors</strong>: hair color/style, eyes, skin tone, signature features</li> <li><strong>Silhouette</strong>: tall/short, athletic/curvy/slim, wide hips, broad shoulders</li> <li><strong>Outfit core</strong>: main clothing items + color palette</li> <li><strong>Accessories</strong>: 1–3 max</li> <li><strong>Optional vibe tags</strong>: “elegant,” “punk,” “gothic,” etc.</li> </ol> <h3>What <em>not</em> to do (IMPORTANT)</h3> <p>Don’t start with a shopping list of tiny details. Your character’s identity should win the first half of the prompt.</p> <hr /> <h2>7) Shorter = Cleaner (And Usually Higher Quality)</h2> <p>Long prompts don’t just “add more detail.” They also:</p> <ul> <li>create contradictions</li> <li>dilute priorities</li> <li>push your strongest features into the “ignored” zone</li> </ul> <h3>A practical strategy: build a 3-layer prompt</h3> <p><strong>Layer A (Core Identity):</strong> must never change <strong>Layer B (Signature Look):</strong> usually consistent <strong>Layer C (Nice-to-have):</strong> optional, removed first when quality drops</p> <p>Example (not for copying,just structure):</p> <ul> <li><strong>A:</strong> <code>1girl, short black hair, amber eyes, beauty mark under eye</code></li> <li><strong>B:</strong> <code>black turtleneck, long coat, silver earrings</code></li> <li><strong>C:</strong> <code>subtle eyeliner, patterned scarf, rings</code></li> </ul> <p>If images get messy, delete Layer C first.</p> <hr /> <h2>8) Common Trap #1: “Rear Size” and Forced Camera Angles</h2> <p>Some body-part emphasis tags accidentally steer <strong>composition</strong>.</p> <p>A classic example is explicitly emphasizing rear size,models frequently interpret that as “show it,” which nudges the camera behind the character more often than you want.</p> <h3>Better ways to communicate “voluptuous” without forcing the camera</h3> <p>Try silhouette descriptors instead of explicit focal commands:</p> <ul> <li><strong>“wide hips”</strong></li> <li>“defined waist”</li> <li>“curvy build”</li> <li>“full figure”</li> <li>“thick thighs” <em>(use sparingly; it can become a focus magnet in some models)</em></li> </ul> <p>If you want your character to read curvy but still get varied shots, silhouette terms are your friend.</p> <hr /> <h2>9) Common Trap #2: Shoes (And Why Feet Suddenly Take Over Your Life)</h2> <p>If you specify footwear, many models treat it as a hint that you want to see it, which increases:</p> <ul> <li>full-body framing</li> <li>downward camera bias</li> <li>“visible feet” frequency</li> </ul> <h3>If you don’t care about shoes, don’t mention them</h3> <p>If shoes matter, use them intentionally:</p> <ul> <li>keep footwear simple (“black boots”)</li> <li>avoid extra foot-related tags unless you truly want feet visible a lot</li> </ul> <h3>Better framing control (without shoe obsession)</h3> <p>Instead of specifying shoes, control framing via:</p> <ul> <li>“portrait”</li> <li>“upper body”</li> <li>“cowboy shot”</li> <li>“close-up”</li> </ul> <p>These control composition without locking you into “full body every time.”</p> <p><img src="https://miocai.com/media/uploads/697b8dca51f9b_paste.png" alt="focuses on the wrong thing" /></p> <h2>Like what are these face feet</h2> <h2>10) Don’t Put Props You Don’t Want in Every Shot</h2> <p>This is the #1 reason people get “samey” images.</p> <p>If you write:</p> <ul> <li>“wearing headphones”</li> <li>“holding a sword”</li> <li>“smoking”</li> <li>“glasses” …you’re requesting those items constantly.</li> </ul> <h3>The accessory rule: 1–3 signature items max</h3> <p>Pick a small set of iconic elements and let the scene do the rest:</p> <ul> <li>one signature accessory (glasses OR earrings OR choker)</li> <li>one signature clothing element (coat OR hoodie OR dress)</li> <li>one signature feature (scar OR beauty mark OR hairstyle)</li> </ul> <p>Everything else belongs in scene prompts or outfits.</p> <hr /> <h2>11) Using the Outfit Prompt System Like a Pro (Structured Mode)</h2> <p>Structured mode is basically “prompt hygiene enforced by UI.”</p> <h3>Clothes field: keep it wearable and consistent</h3> <ul> <li>Use <strong>main outfit + palette</strong></li> <li>Don’t list 12 items</li> <li>Avoid scene-specific outfits unless you intend that outfit to be default</li> </ul> <p><strong>Good clothes field style:</strong></p> <ul> <li>“black bomber jacket, white t-shirt, dark jeans”</li> <li>“long red dress, gold trim, black gloves”</li> </ul> <h3>Body field: silhouette and proportions, not anatomy fixation</h3> <ul> <li>“athletic build, tall, long legs”</li> <li>“curvy build, wide hips, relaxed posture”</li> <li>“lean build, narrow shoulders”</li> </ul> <h3>Face field: identity anchors only</h3> <ul> <li>hair style + color</li> <li>eye shape/color</li> <li>key features (freckles, scar, mole)</li> <li>keep it stable</li> </ul> <h3>Extras field: treat as “rare spices”</h3> <p>If extras is available, keep it minimal and avoid camera/scene locks:</p> <ul> <li>“soft lighting”</li> <li>“clean lineart”</li> <li>“muted palette” Not: “night city rooftop, rain, neon reflections” (that’s a scene)</li> </ul> <hr /> <h2>12) Model Choice: Obedience vs Aesthetic (Pick Your Poison)</h2> <p>Different MiocAI image models have different personalities.</p> <p>A common pattern you’ll notice:</p> <ul> <li>Some models are <strong>high obedience</strong> (follow prompts literally)</li> <li>Some are <strong>high aesthetic</strong> (prettier by default, but drift more)</li> </ul> <p>So you choose based on what you’re doing:</p> <h3>When you want strict consistency</h3> <p>Pick the model that:</p> <ul> <li>locks hair/eyes reliably</li> <li>respects outfit items</li> <li>doesn’t “reinterpret” accessories</li> </ul> <h3>When you want the best-looking single image</h3> <p>Pick the model that:</p> <ul> <li>composes beautifully</li> <li>has strong lighting and rendering</li> <li>may freestyle details if your prompt is too long</li> </ul> <p>A practical approach:</p> <ul> <li><strong>Design phase:</strong> use obedience model to finalize the character sheet</li> <li><strong>Art phase:</strong> switch to aesthetic model once your prompt is clean and proven</li> </ul> <p><img src="https://miocai.com/media/uploads/697b8e91bdaa0_paste.png" alt="two different models" /></p> <hr /> <h2>13) Iteration Workflow: How to Fix a Prompt Without Losing Your Mind</h2> <p>MiocAI gives you a built-in advantage: <strong>test images</strong>.</p> <h3>The “4-image test loop”</h3> <ol> <li>Write/adjust your prompt.</li> <li>Click <strong>Test Prompt</strong> (or generate test images).</li> <li>Look for:</li> </ol> <ul> <li>face consistency</li> <li>hair consistency</li> <li>outfit consistency</li> <li>unwanted recurring composition (same angle / same framing) <ol start="4"> <li>Make <strong>one change at a time</strong></li> <li>Re-test.</li> </ol></li> </ul> <p>If you change 6 things at once, you won’t know what fixed (or broke) it.</p> <h3>Keep a tiny changelog</h3> <p>Even a simple note helps:</p> <ul> <li>v1: “hair keeps changing” → moved hair tags earlier</li> <li>v2: “too many full-body shots” → removed shoes tag, added “upper body”</li> <li>v3: “glasses disappear” → moved “glasses” earlier</li> </ul> <hr /> <h2>14) Debugging Guide: Symptom → Cause → Fix</h2> <h3>Symptom: Hair color changes every other image</h3> <p><strong>Cause:</strong> Hair tags too late, prompt too long, model freestyling <strong>Fix:</strong> Move hair tags into the first 25–40% of the prompt; delete weak extras</p> <h3>Symptom: Outfit becomes inconsistent</h3> <p><strong>Cause:</strong> Outfit list is too detailed or contradictory <strong>Fix:</strong> Reduce outfit to core items + palette; move “fashion details” to outfit system/outfits</p> <h3>Symptom: Always full-body</h3> <p><strong>Cause:</strong> Shoes, “full body,” overly detailed clothing, body-part focus tags <strong>Fix:</strong> Remove footwear; add framing tag like “portrait” or “upper body”</p> <h3>Symptom: Always same angle</h3> <p><strong>Cause:</strong> Angle tags in appearance prompt (“from behind,” “from below,” etc.) <strong>Fix:</strong> Remove angle tags from appearance prompt; use scene/action prompts instead</p> <h3>Symptom: Random props appear</h3> <p><strong>Cause:</strong> Vague tags that imply props (“idol,” “performer,” “soldier”) <strong>Fix:</strong> Replace vague role tags with visual traits (“stage outfit” vs “performer”)</p> <hr /> <h2>15) Example Prompts You Can Learn From (Don’t Copy Blindly,Steal the Structure)</h2> <p>Below are example prompt <em>styles</em> designed to show structure. Swap details to fit your character.</p> <h3>Example A: Minimal “stable identity” prompt (anime)</h3> <blockquote> <p><code>1girl, short auburn hair, green eyes, pale skin, freckles, black turtleneck, long grey coat</code></p> </blockquote> <p>Why it works:</p> <ul> <li>short</li> <li>identity first</li> <li>outfit simple</li> </ul> <p><img src="https://miocai.com/media/uploads/697b8f15134ce_paste.png" alt="goddamn mate" /></p> <hr /> <h3>Example B: Balanced prompt with one signature accessory</h3> <blockquote> <p><code>1boy, messy silver hair, blue eyes, small scar on cheek, slim build, white shirt, dark vest, thin tie, round glasses</code></p> </blockquote> <p>Why it works:</p> <ul> <li>scar + glasses are strong identity anchors</li> <li>still not overloaded</li> </ul> <p><img src="https://miocai.com/media/uploads/697b8fc3039e0_paste.png" alt="men 🤤" /></p> <hr /> <h3>Example C: Structured mode example (good separation)</h3> <p><strong>Clothes:</strong></p> <blockquote> <p><code>black hoodie, dark cargo pants</code></p> </blockquote> <p><strong>Body:</strong></p> <blockquote> <p><code>athletic build, broad shoulders, relaxed stance</code></p> </blockquote> <p><strong>Face:</strong></p> <blockquote> <p><code>short black hair, grey eyes, sharp jawline</code></p> </blockquote> <p><strong>Extras (optional):</strong></p> <blockquote> <p><code>soft lighting</code></p> </blockquote> <hr /> <h3>Example D: “Curvy without forced framing”</h3> <blockquote> <p><code>1girl, long brown hair, hazel eyes, curvy build, wide hips, fitted sweater, high-waisted skirt, hoop earrings</code></p> </blockquote> <p>Why it works:</p> <ul> <li>silhouette term (“wide hips”) instead of “focus on body part”</li> <li>outfit readable, not encyclopedic</li> </ul> <blockquote> <p>📷 <strong>Image placeholder:</strong> “Curvy silhouette with varied camera angles.”</p> </blockquote> <hr /> <h2>16) Advanced Tips (Use Only After You’re Getting Consistent Results)</h2> <h3>A) Emphasis syntax (if supported by your model/tools)</h3> <p>Some systems support strengthening/weakening tokens (often <code>{}</code> or <code>[]</code>). If your model respects it, it can help when one detail keeps getting ignored.</p> <p>Use sparingly:</p> <ul> <li>too much emphasis makes the image “clingy” to that detail</li> </ul> <h3>B) Quality boosters (optional)</h3> <p>Some models respond to generic “quality” tags. They’re not magic, but can help sometimes:</p> <ul> <li>“best quality”</li> <li>“high quality”</li> <li>“very aesthetic”</li> </ul> <p>Tradeoff: they consume prompt space and can be placebo on some models. Test once,keep only if it clearly improves results.</p> <hr /> <h2>17) The “Do / Don’t” Cheat Sheet</h2> <h3>Do</h3> <ul> <li>Put identity traits early (hair, eyes, signature features)</li> <li>Keep the appearance prompt focused on <strong>what should persist</strong></li> <li>Use structured mode if you like clean control</li> <li>Use the test images loop and change one thing at a time</li> <li>Prefer silhouette descriptors over “camera bait” anatomy tags</li> </ul> <h3>Don’t</h3> <ul> <li>Add props you don’t want forever</li> <li>Add shoes unless you want more full-body / feet visibility</li> <li>Add camera angles to the appearance prompt</li> <li>Write a 400-character prompt and expect stability</li> <li>Expect avatar uploads to influence generation automatically</li> </ul> <hr /> <h2>18) What Act 2 Will Cover (So You Don’t Misuse Act 1)</h2> <p>Act 1 is about <strong>designing a stable character sheet</strong>.</p> <p>Act 2 will focus on personalities. thats all for today, take care!</p>