Plan a disclosed synthetic-character workflow.

Craft Before any image, write one sentence: who this character is for and what she is for. “A street-style diary for 20-something creatives in Berlin” is a position. “An AI girl” is not.

Craft Set three goals and let them pick the formats: awareness → posts and stories; demonstration → reels with process; community → captions that reply. Goals also decide how often you publish.

Product In the studio, positioning lives with the character: the passport is a data plugin, so the character’s description, appearance and rules stay in one place instead of being scattered across prompts.

Workflow Disclosure is part of positioning, not an afterthought: from day one, the bio and the caption template say the character is AI. This presentation identifies its synthetic-character studies, and the FAQ states what the studio does and does not do.

Craft A character concept is more than a face: personality, taste, tempo, boundaries. Write three things she would post and three things she would never post. The “never” list steers scenes more than any prompt trick.

Craft Hard line: do not model a real, identifiable person. No real name paired with a generated face, no “age them up / make them younger” games with public figures, no cloning a creator’s recognizable identity. A synthetic face is fine; a synthetic claim to a real person’s identity is not.

Craft Two more boundaries that save months: no minors in any suggestive context, and no impersonation of professions that require credentials (medical, legal, financial advice) unless the fiction is stated in the caption itself.

Workflow Keep the “never” list next to the appearance data. The demo passport template on the landing shows the shape: identity blocks, what to preserve, device, references — free-text fields like these are the right home for boundaries.

Product The studio’s character passport is a six-block canon: identity at three depths (micro / mid / full), anomalyLock — the features to preserve (moles, freckles, asymmetry, real skin texture), device, referenceSet and faceAdherence per mode. The full demo template is public on the landing.

Product Anomaly Lock exists because recognition lives in “imperfections”. Beauty-mode faces drift; pinned moles and skin texture hold the same person from frame to frame.

Craft Pick constants you can hold for months: at most two hairstyles, a limited palette, two or three recurring locations, one device look (an “iPhone 15 Pro Max” shoots differently than a cinema camera — pick one and stay).

Workflow Keep the passport short enough to re-read in one minute before every session. If it grows past a screen, move scene-level detail into scene packs and leave the person-level canon in the passport.

Craft The identity base must be clean: your own generated frames, licensed stock, or fully synthetic boards. Do not reuse a real person’s photographs or likeness without the appropriate permissions and disclosures.

Product The passport takes four or five references: front, ¾, profile, one pose. That is enough to anchor identity without over-fitting the face to a single angle (see Quick start).

Craft Build a small style board per project: light, wardrobe, locations. For every board item write one provenance line — where it came from and what allows you to use it. Screenshots of other creators’ feeds are not references; they are liabilities.

Workflow One working reference per scene — “photo 5”: the pose, the location, the light. Engines listen to references more than to text; the scene reference carries the mood while the passport carries the face.

Product You describe the scene in Russian or English; the studio assembles the final prompt through per-engine adapters (Nano Banana Pro, Kling, Seedance, Veo, Omni Flash). Auto-tags catch location, light, pose and outfit; Prompt Doctor tells you what to strengthen; the strength meter estimates completeness; the cliché guard flags the “neural” look; the realism panel adds UGC texture markers. Four formats cover the feed’s jobs: Frame, Series, Shoot, Feed.

Craft Iterate one variable at a time: composition → light → wardrobe → camera. Generate, grade, keep or revert. Batching five changes into one regeneration teaches you nothing about what worked.

Workflow Grow a personal library of scene structures that worked for you (“window light, half-turned, phone at chest height”). Reuse structures, not whole prompts — the passport plus a structure rebuilds any scene.

Product The studio’s consistency tools are the passport itself: anomalyLock pins the features that must survive every regeneration, and faceAdherence is set per mode — a value for stills, a slightly lower one for motion. The principle is visible in the landing’s demo shoots: “one character · 4 angles”, “one character · different light” (Showcase).

Craft Wardrobe and light are identity too. Lock a light direction and a palette per series; change one constant per shoot, never all three at once. The same “device” (one phone look, one grade) does more for recognition than any single perfect frame.

Workflow Run a weekly identity audit: open five random recent frames next to the passport and mark drift (jaw softer? eyebrows thicker? skin too clean?). Fix drift by tightening the anomalyLock wording — not by stacking more prompt adjectives on every scene.

Product The same passport feeds the video engines: prompts go through the Kling, Seedance, Veo and Omni Flash adapters; number keys 1–4 switch the active video engine; a multi-shot flow for Kling is built into the feed planner. Video prompts are engine-ready — you paste them into your own engine accounts.

Craft Design motion from stills, not from zero: take a working still, add one motion idea — a turn, a step, hair in wind, a hand entering frame. Keep the face task identical to the passport; motion amplifies whatever identity is already there, including drift.

Craft Pick the honest duration: 2–4-second loops hold feed attention and hide temporal artifacts better than long takes; dialogue-length shots are for talking moments; B-roll is for atmosphere between posts.

Workflow Shot list first, generation second. The studio exports a shot list with day markers; generate shot by shot, grade each, and only then cut a reel. The landing’s showcase videos were produced exactly in this order.

Craft Decide how the character “speaks”: captions only, text-on-screen, or synthetic voice. All three are performance choices and all three need the same thing — a consistent register (short lines, one humor style, one way of starting posts).

Craft A synthetic voice must never imitate a real, identifiable person — that is the same hard line as the face rule in chapter 02. Label voice clips as AI in the caption or via the platform’s audio label.

Product Honest scope: the studio works with text — prompts, captions and shot lists. It does not generate or clone audio. Voice production, if you use it, happens in your own tools under their terms. Caption voices in the studio (photo-dump, unsent sms, archive numbering) shape the “speech” rhythm of the feed without any audio.

Workflow Keep two pillars separate. PRODUCT_REELS demonstrate the craft: process, before/after, how a prompt becomes a frame, engine comparisons. MODEL_WORLDS are the character’s life: diary, raw, staged. Mixing them in one post teaches the audience nothing and reads as an ad in a diary.

Product The studio gives each pillar a visual system: worlds A/B/C (Diary / Raw / Staged) and twenty scene packs. The site itself follows the split — the Showcase page is product proof, the Cast page is a model-worlds reference.

Craft Alternate pillars on a schedule (for example 4:1) and never mix worlds inside one feed — the studio flags a mixed-world feed because a character who is “raw underground” on Monday and “staged couture” on Tuesday stops being anyone (FAQ).

Workflow Every published asset traces to exactly one internal master file: generate → grade → name (date · engine · scene) → store → publish from the master only. No re-exporting from chats, no screenshots of your own work. When a platform asks “where is this from?”, the answer is one file and one prompt session.

Product The studio keeps the loop local and reviewable: bundle export as JSON/TXT, a Series Card PNG, the hit journal with per-engine results, and backups that carry presets. Your data stays in your browser — the license check is the only network call (FAQ).

Craft Keep a simple register: file name, date, engine, scene pack, grade, where published. Ten columns in a spreadsheet beat any “I’ll remember” system — and they are what chapter 13’s provenance check feeds on.

Product The proof you can trust is the kind you can inspect: real application evidence and real outputs, each labeled with where it came from. Media shown as studio proof carries its provenance; anything without provenance is presented as craft reference, not as studio output.

Craft When you show “how it’s made”, separate the tool from the output: a screen of the studio (passport, adapters, strength meter) next to the frame it produced. Never present another studio’s or engine’s UI as yours — audiences and platforms both check.

Craft Say plainly where generation happens: the studio builds prompts; images and videos come out of your own engine accounts. The site keeps that statement in the FAQ and in the footer of every page — the same honesty belongs in your posts.

Craft Check the classic failure points, every time: faces (identity drift, eye artifacts, teeth), hands (finger count, joints), temporal artifacts in video (flicker, morphing background, jewelry that changes), captions (spelling, contrast, safe areas), crops (heads, feet, platform ratios), sound (sync, noise floor), mobile readability (look at the frame at phone size before publishing).

Product Inside the studio, the cliché guard flags the “neural” look before it ships, and 150 automated self-checks run against the scene. The hit journal records what actually generated well, per engine.

Workflow Grade every frame S / A / B / C. Publish S and A only; B goes to rework; C is archived with a one-line reason — nothing is deleted, because bad frames teach too (chapter 14 uses them).

Workflow A provenance note for every asset: prompt session, engine, date, references used, master file. The register from chapter 10 is the source; the note is what you can show if a platform or a rights holder asks.

Craft Consent and rights: no real person’s likeness, voice or identifiable property without permission; no trademarks worn or shown in a way that implies endorsement; music, fonts and stock imagery under licenses that actually cover commercial posts. The characters on this site are fully synthetic and unnamed for exactly this reason.

Craft Platform rules differ and change: AI-content labels, disclosure wording, audio rights, branded-content tags. Check the current rules of each platform before publishing, not after a takedown. This guide is education, not legal advice — for commercial publication, get an actual rights review.

Workflow A five-point publication gate, every post: disclosure present → provenance recorded → QC grade ≥ A → third-party rights checked → platform rules checked. Only then publish.

Craft Track a few honest metrics per post — reach, saves, watch-through, the quality of comments — and compare formats over weeks, not hours. Day-by-day swings on a new account are noise.

Workflow Learn forward-only: after each week, write down what to repeat and what to drop next time — then act on exactly that list. No uncontrolled A/B on a synthetic persona: running parallel “versions” of a character multiplies identities, fragments consistency and reads as fake to the audience.

Product The hit journal keeps learning attached to scenes and engines: what generated well, with which passport and scene pack. The archive of B/C grades from chapter 12 is the other half of the lesson.

Roadmap Nothing in this guide promises future outcomes. Capabilities marked Roadmap are ideas, not shipping dates; confirmed behavior is always marked Product and matches the current version’s changelog on the landing.