The Problem with Prompting for Style
A text prompt for visual style looks like this: "cinematic, warm golden hour lighting, film grain, muted palette." That description might get you close on one frame. Across a sequence of twenty clips, it will produce twenty subtly different interpretations. Diffusion models are probabilistic — the same prompt generates a distribution of outputs, not a deterministic one.
The more you rely on language to specify a visual style, the more drift you get. Language is imprecise. Image distributions are wide.
What a Style Pack Actually Is
A Style Pack is a collection of conditioning inputs that together describe your visual language more precisely than text can. It includes:
- A palette — 4–6 dominant hex values extracted from source footage or a reference image. These get baked into the IP-Adapter conditioning so every frame is pulled toward the same color profile.
- A lighting summary — not a text description but a conditioning weight that tells the model how strongly to preserve the source frame's luminance structure via ControlNet.
- A prompt fragment — a short, tight descriptor that is appended to every frame prompt. Not a full description — a modifier. "warm grain, analog" not "a cinematic scene with warm golden hour lighting and visible film grain."
- A negative prompt fragment — what to exclude. "overexposed, digital noise, clean CGI" is a different negative than "blurry, low quality." Match it to your aesthetic.
How to Build One From Source Footage
ShotLock's analysis pipeline extracts a Style Pack suggestion automatically from your clips. Run an analysis job against 2–3 representative clips — ideally from different lighting conditions in your project — and the system will:
- Extract keyframes at a configurable interval
- Run k-means clustering on pixel values to identify dominant palette colors
- Build a suggested Style Pack with those colors already populated
Review the suggestion in the Style Packs tab, adjust the prompt fragment to match your aesthetic intent, and save. From that point forward, every render job that uses this Style Pack will be conditioned on those specific inputs.
Using Multiple Style Packs in One Project
A single project might need more than one visual language — a present-day storyline and a flashback sequence, for example. Style Packs are designed for exactly this. Create one per look, assign them to different clips in your Queue, and the render pipeline applies the correct conditioning to each.
This is the difference between a tool built for single-image generation and one built for editorial work.