learn Media-production skill.md
-------------------------------

General useful tips for media assets (images / videos) generation:

- Split every prompt into an immutable half and a variable half
Immutable = character bible + locked style suffix + locked negative prompt. Variable = shot type, pose, action, setting, lighting-of-the-moment. Once episode 1 is done, writing episode 2's prompts becomes editing one clause instead of writing from scratch.

- Never use the words "face", "eyes", "expression" to describe emotion. Describe emotion through body language + manga effect lines instead: trembling lines, sweat drops (on the mask surface), speed lines, dark aura, sparkle effects. With no face, emotion has to be carried by everything else
Body posture (clenched fists, hunched shoulders, wide stance), hand gestures, camera angle (low angle = powerful, high angle = small/vulnerable), environmental storytelling (rain, broken glass, fire, wind), color-temperature shifts between scenes, motion blur or speed lines — and, unique to this concept, the state of the tag itself. A flickering or cracking tag can read as fear; a tag blazing brighter can read as triumph. The tag is your face-acting substitute.

- The hardest technical part: getting the nickname to render cleanly
Legible in-image text used to be the weak point of every image model. It's much better now but still needs well prompting, focus thinking, and precise generation. As we need featureless surface, completely blank, absolutely no eyes, no nose, no mouth, no eyebrows, no facial features.

- Avoid making shadows that suggest facial;
(no nose shadow, eye socket shadow, and lip shadow).
- Masking the whole face oval and regenerating it cleanly is more reliable than pixel-precise spot-fixing.

- Keep nicknames short: 3–9 characters, one word, no special characters.
Describe the lettering in plain words ("thick bold stencil lettering," "condensed sans-serif caps") rather than asking for elaborate script.
Put the exact nickname in quotes every time, with identical phrasing across scenes.
keep nickname ALL CAPS, and repeat it twice in the prompt ("a solid mask with the words THE DREAMER written in bold letters across it — the mask has no other features, only the text THE DREAMER").

Some examples of nickname (you can use or imitate like them based on content)

أبو سريع / بو سريع: The Speedster (Rushed, impatient, and does everything in double-time)
وسواس: The Worrywart (Anxious, paranoid, and obsessively overthinks worst-case scenarios)
ابنة الضغط: The Panic Inducer (High-strung, perpetually stressed, and elevates everyone’s blood pressure)
غشاش: The Hustler (Sneaky corner-cutter who always looks for unfair shortcuts)
لئيم: The Petty Saboteur (Spiteful, holds grudges, and takes joy in petty payback)
المعلم: The Big Boss (Traditional, street-smart patriarch who commands the room)
الحكيم: The Voice of Reason (Calm philosopher who drops proverbs at the most inconvenient times)
مصمم: The Mastermind Engineer (Brilliant planner obsessed with blueprints, gadgets, and tech)
دقيق: The Perfectionist (Meticulous nitpicker obsessed with precision down to the millimeter)
كسلان: The Sloth (Master of doing nothing who turns minimal effort into an art form)
نشيط: The Energizer (Hyperactive go-getter whose boundless energy exhausts everyone around)
ناعم: The Velvet Viper (Sweet and angelic on the outside, ruthlessly cunning and two-faced inside)
حكاكة: The Clingy Flatterer (Shameless sycophant who brown-noses and clings to anyone in power)
حارس: The Iron Watchdog (Unblinking, hyper-vigilant sentinel who tracks every single movement)
أم فضيحة: The Drama Queen (Loud, sensationalist gossip who turns minor mishaps into viral scandals)

- Standard negative-prompt / error checklist — use on every generation (Ensure following mistakes not present in the visuals)
deformed hands, extra fingers, fused fingers, missing fingers, mutated limbs, extra limbs,
disproportionate anatomy, bad anatomy, asymmetrical body,
visible eyes, visible nose, visible mouth, visible eyebrows,
realistic facial features, blurry, low resolution, pixelated,
jpeg artifacts, oversaturated noise, watermark, signature, random text, garbled text, duplicate character, cloned figure, extra character, cropped, out of frame, low detail background.

The "visible eyes/nose/mouth" lines matter specifically — most models default toward drawing a face because their training data is dominated by faces. You have to actively suppress it, not just leave it undescribed.

- Avoid making shadows that suggest facial;
(no nose shadow, eye socket shadow, and lip shadow).
- Masking the whole face oval and regenerating it cleanly is more reliable than pixel-precise spot-fixing.

- Lock one color/lighting "grade" phrase for the whole episode
Repeat it verbatim in every scene — e.g. cinematic teal-and-orange grade, high contrast, saturated, volumetric lighting. This one repeated phrase is what makes many separate generations read as one continuous piece of media instead of disconnected images.

- Universal quality-booster suffix
Append to every prompt: ultra-detailed, sharp focus, dramatic lighting, rich saturated colors, professional composition, 4K render, masterpiece quality.

- Batch, don't settle
Generate at least 3–4 variants per scene and pick the one closest to the locked character bible. Regenerate a whole scene rather than accept visible drift — it compounds fast across an episode, and viewers notice continuity breaks especially hard in faceless-character series, since the tag/silhouette is all they have to track identity with.

- Keep characters, props, and settings original
Avoid naming copyrighted characters, franchises, logos, or real people in prompts.
- Reference Weight Tuning If the character drifts across generations — outfit changes, proportions shift, face gains features — adjust the reference influence strength in your tool:
Too much drift: Increase the reference weight.
Too rigid or stiff: Decrease the reference weight.
Background detail bleeding into the face area: Lower the reference weight and strengthen the environment description in the prompt.

- Avoid making shadows that suggest facial;
(no nose shadow, eye socket shadow, and lip shadow).