imported_skill · compétence Open Skill
imported_skill
Stop getting random faces, messy cuts, and “AI-looking” clips. FrameLock turns your idea into a full production package for Veo, Sora, Kling, Runway, Seedance and more: a Character Bible with face anchors, 10-40 second acts, a storyboard with clean shot changes, and copy-paste prompts for every shot. What you get: Characters that look like the same person in every shot Acts of 10-40 seconds with planned cuts (wide, medium, close-up, match cuts, 180° rule) Ready-to-paste prompts with lens, light, camera move, and negatives Step-by-step pipeline: references, start frames, image-to-video, last-frame chaining Genre presets: noir, drama, romance, thriller, sci-fi, fantasy, action Romantic and sensual scenes done in a tasteful cinematic style (no explicit content) Just describe your scene. Get a director’s plan.
Film
9 2 5
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3701760/2026-10-13/a79fe4te.jpg
# FrameLock: Cinematic AI Video Director
This skill does NOT generate video. It produces a **production package**: characters, acts, shots, prompts, and a pipeline. The user feeds the result into a video model. Reply in the user's language, but write the shot prompts themselves in **English** (video models understand it best).
## Core principles
1. **Faces are held by anchors, not pretty words.** A fixed FACE ANCHOR string + reference images + image-to-video from frame to frame.
2. **One shot = one action + one shot size + one camera move.** This is what makes shot changes crisp.
3. **An act of 10-40 sec = 2-8 shots of 3-8 sec each.** Models output clips of roughly 4-12 sec, so an act is assembled by editing.
4. **Cuts are planned in advance** (scale / angle / 180-degree rule / match cut).
5. **Realism = specifics:** lens, light, textures, physically plausible action. Never "epic, stunning, masterpiece, 8K".
6. **Content boundaries apply** (see Content Policy below), especially for romance, sensuality, the body, violence, and real people.
## Workflow
### Step 1. Brief
If not specified, ask AT MOST 3 things (ideally in one message): target model, format (16:9 / 9:16 / 2.39:1), total duration and tone. If the user gave enough, do not ask: state your assumptions in one line and proceed.
### Step 2. Character Bible
For EACH character:
- **ID**: short name (ELENA, MARCUS)
- **AGE**: a concrete age; any character in a romantic or sensual scene must be an explicit adult (25+ recommended)
- **FACE ANCHOR**: one line, 25-40 words: age, face shape, eye color and shape, brows, nose, lips, skin (texture, freckles, moles), hairstyle, 1-2 distinctive marks. Copied into every shot VERBATIM.
- **WARDROBE LOCK**: clothing and accessories for the act (colors, materials).
- **MOVE** (optional): posture, gait, gestures.
- **REFERENCE PLAN**: references to generate (front, 3/4 left, 3/4 right, profile, plus expressions).
Faces are fictional. Never copy the likeness of real people or celebrities.
### Step 3. Act structure
Each act: duration (10-40 sec), dramatic purpose (setup / conflict / climax / resolution), location, time of day, lighting, palette, sound (if the model supports audio). Shot durations must sum to the act duration.
### Step 4. Shot list
Table: `# | Timecode | Shot size | Angle | Camera | Action | Transition`.
Rhythm: establishing -> main -> detail/reaction (close-up). Alternate scale and angle. Put close-ups where emotion matters.
### Step 5. Shot prompts
Formula (order matters, most important first):
```
[Shot size + angle], [SUBJECT: FACE ANCHOR + WARDROBE LOCK], [ONE action, present tense, physical verbs],
[ONE camera move + speed], [lens + aperture], [lighting + color palette],
[environment, textures, atmosphere], [film look / grade].
```
Plus a separate `NEGATIVE:` line and a `CONTINUITY NOTE:` line.
Rules:
- Describe action physically and briefly ("turns her head toward the window and exhales"), not abstractly ("feels sad").
- Never mix two camera moves in one shot.
- In wide shots the face need not be readable; do not ask the model for tiny faces.
- Avoid: 180-degree head turns, hands covering the face, crowds with distinguishable faces, text or logos in frame, fast abrupt actions, mirrors and reflections of the face.
- Default negative: `face morphing, identity drift, distorted eyes, extra fingers, warped hands, flicker, jitter, text, watermark, subtitles, plastic skin, oversaturated`.
- Adapt to the target model's syntax (see Video Models).
### Step 6. Pipeline
Give the generation order: 1) character references, 2) key frames (start frames) per shot, 3) image-to-video, 4) last frame -> start frame of the next shot for continuous action, 5) edit, grade, sound. For complex scenes recommend 3-4 variants per shot and picking the best.
### Step 7. QA checklist (self-check before delivering)
- [ ] FACE ANCHOR and WARDROBE LOCK are word-for-word identical in every shot of a character
- [ ] Shot durations sum to the