Prompt Engineering

Prompt Guide for Creator Flow (Image + Video)

This guide focuses on practical prompt writing for Cannon Studio’s Creator Flow pipeline. It covers opening-frame image prompts, motion-focused video prompts, and reference usage for continuity.

Core Prompt Principle

Write prompts as production instructions, not prose. Treat each prompt like a compact shot brief: what the camera sees, what matters most, and what must stay stable. The highest-quality outputs usually come from clear constraint order, not longer text.

  • Put hard constraints first (subject identity, framing, environment)
  • Put style and atmosphere second (lighting, tone, texture)
  • Put strict negatives last (no text/logos/watermarks/UI artifacts)

High-Performance Prompt Formula

For the reference-aware image model used in Creator Flow, prompt order strongly affects output stability. Use this sequence and keep each part concise.

  1. Shot + lens intent: close-up / medium / wide, lens feel, camera height
  2. Primary subject lock: exact identity anchors, wardrobe anchors, silhouette anchors
  3. Environment lock: location, zone, key props, weather/time-of-day
  4. Lighting and color: contrast level, temperature, mood direction
  5. Composition constraints: foreground/background depth and subject placement
  6. Negative constraints: no text, letters, logos, watermarks, signatures, UI

Image Prompt Structure (Shot Pre-Production)

  • Frame type: wide, medium, close-up, overhead, POV, etc.
  • Subject: who or what is dominant in frame
  • Composition: position, depth cues, lens feel, perspective
  • Lighting and atmosphere: time of day, contrast, color temperature
  • Material detail: texture, weather, environment specifics

Keep image prompts focused on frame zero only. Do not describe future movement in the image prompt. The more motion language you add here, the more likely composition quality drops.

Copy-Ready Opening Frame Template

[shot type + lens feel], [primary subject with identity anchors], in [specific location/zone], [lighting + atmosphere], [composition depth and placement], [texture/material details], cinematic quality, no text, letters, logos, watermarks, signatures, or UI elements.

Video Prompt Structure (Shot Production)

  • Action: what physically changes from frame 0 onward
  • Camera behavior: push-in, pan, tilt, track, lock-off
  • Pacing language: subtle, gradual, sudden, controlled
  • Continuity constraints: maintain setting, subject identity, and scene logic

Motion prompts should describe change over time, not restate the entire frame. Keep one dominant action beat, one camera behavior, and one continuity lock line.

Copy-Ready Motion Template

Start from the existing opening frame. [subject action] with [pace], camera [specific move], preserve [identity + environment continuity anchors], keep motion physically plausible and stable.

Using References Effectively

Reference images are best used when continuity has high risk. Keep usage intentional: use them to anchor identity and environment, not to replace prompt clarity.

  • Character references: maintain identity under motion and partial visibility
  • Location references: preserve environment continuity during camera movement
  • Start frame references: keep generated video aligned with the opening composition

Use references aggressively for profile turns, occlusions, entrances/exits, and fast motion near faces. Use lighter prompting without references for simple static shots where continuity risk is low.

Continuity Locking That Actually Works

  • Repeat 2-4 exact identity anchors across connected shots (not 12 variants of the same idea)
  • Keep location nouns identical across prompts for the same place
  • Reuse one style anchor phrase across the sequence for color/texture consistency
  • Avoid introducing new wardrobe props mid-scene unless that change is intentional
  • When in doubt, lock fewer things but lock them clearly

Prompt Quality Checklist

  • Clear subject priority: primary subject is obvious
  • Visual coherence: lighting and environment do not contradict each other
  • Action plausibility: movement is physically reasonable for the shot length
  • No ambiguity: avoid vague terms like “cool” or “nice” without specifics
  • Continuity intent is explicit when needed
  • Negative constraints prevent overlays/artifacts (text/logos/watermarks/signatures/UI)

Common Prompt Failure Modes

Failure: too many competing actions in one short shot.

Fix: simplify to one dominant action plus one secondary beat.

Failure: style drift between connected shots.

Fix: repeat key continuity descriptors and use shared references.

Failure: cinematic words without scene-specific detail.

Fix: pair style language with concrete subject/environment detail.

Failure: output contains text, glyphs, logos, or overlay artifacts.

Fix: add explicit negative constraints at the end of the prompt and keep them consistent.

Failure: identity drift after camera motion begins.

Fix: keep one dominant action, reduce competing descriptors, and use character reference support.

Fast Optimization Loop (High Signal)

  1. Generate with the compact formula and clear negatives
  2. Identify the single biggest miss (identity, framing, lighting, or motion)
  3. Edit only the constraint that controls that miss
  4. Regenerate and compare against the exact same shot objective
  5. Promote winning language into your reusable style/continuity anchors

The goal is deterministic iteration: one change per pass, one measurable improvement per pass.

Continue Learning

Pair prompt design with Creator Flow and editing guides to improve end-to-end output quality.