The AI Director’s Playbook: Training Custom Midjourney v6 Styles for Brand Actor Consistency

In the rapidly evolving landscape of generative video production, the holy grail for directors, agency creatives, and workflow engineers is temporal and visual consistency. While tools like Runway Gen-3, Luma Dream Machine, and Kling AI have democratized high-fidelity motion, they all suffer from a fundamental vulnerability: upstream asset drift. If your source keyframes lack precise character and stylistic alignment, your generated video sequence will inevitably break down.

To solve this, elite AI filmmaking workflows leverage Midjourney v6 as the upstream “casting director” and “production designer.” By mastering Midjourney’s advanced parameter matrix, specifically Character Reference (--cref) and Style Reference (--sref), we can effectively “train” and lock in a custom Brand Actor and aesthetic style without writing a single line of PyTorch code. This guide details the exact, production-proven pipeline for establishing absolute character and style consistency for enterprise-grade generative video campaigns.

—

Understanding the Midjourney v6 Consistency Engine

Unlike Stable Diffusion, which relies on local training methods like LoRAs (Low-Rank Adaptations) or Textual Inversions, Midjourney v6 utilizes a highly sophisticated, closed-source attention-steering mechanism. To “train” a model in Midjourney, we must build a deterministic prompt and reference structure that forces the latent space to align with our specific Brand Actor and visual identity.

The Core Parameters

• Character Reference (--cref [URL]): Directs Midjourney to mimic the facial structure, hair, and primary physical features of the target actor in the reference image.
• Character Weight (--cw [0-100]): Controls the fidelity of the transfer. --cw 100 (default) copies face, hair, and clothing. --cw 0 focuses exclusively on the face, allowing for seamless costume changes—a critical feature for narrative filmmaking.
• Style Reference (--sref [URL]): Guides the aesthetic, color grading, lighting, texture, and camera optics of the output image.
• Style Weight (--sw [0-1000]): Adjusts the strength of the style injection. The default is 100; pushing this to 800-1000 yields hyper-consistent art direction across varied prompts.

—

Phase 1: Generating the “Anchor Asset” Library

Before launching a multi-shot campaign, you must establish your Brand Actor’s “Anchor Assets.” These are the master images that Midjourney will reference for every subsequent generation. Do not use low-resolution, poorly lit, or highly stylized images as your anchors.

Step-by-Step Anchor Generation Workflow

—

Phase 2: “Training” the Style and Character Matrix

With your Anchor Assets hosted, you are ready to construct the master prompt template. To achieve true brand consistency, we must combine --cref (for the actor) and --sref (for the visual brand identity).

The Anatomy of a Production-Ready Prompt

To generate consistent scenes, use the following structural formula:

/imagine prompt: [Cinematic Shot Type] of [Brand Actor Name] [Action/Environment Description], [Camera/Lens Specification], [Lighting Setup], [Color Palette] –cref [Actor_URL] –cw 0 –sref [Brand_Style_URL] –sw 250 –ar 16:9 –v 6.0 –style raw

Why --cw 0 is Your Best Friend

In a narrative video sequence, your actor cannot wear the same outfit in every scene. By setting --cw 0, you instruct Midjourney to only lock the facial geometry. You can then describe different clothing in the text prompt:

• Scene 1 (Boardroom): ...a woman wearing a tailored charcoal grey blazer... --cref [URL] --cw 0
• Scene 2 (Activewear): ...a woman jogging in a neon green windbreaker... --cref [URL] --cw 0

The face remains identical, but the wardrobe adapts perfectly to the narrative context.

—

Phase 3: Creating Custom Command Shortcuts

Typing out long URLs for every generation is inefficient and prone to syntax errors. Midjourney allows you to package your “trained” actor and style into custom command shortcuts using the /prefer option set command.

How to Set Up a Custom Option

1. In the Discord message bar, type /prefer option set and press Enter.
2. In the option box, type a memorable name (e.g., brand-actor-v1).
3. In the value box, paste your parameter string:
   --cref https://link.com/actor.png --sref https://link.com/brand-style.png --cw 0 --sw 300 --ar 16:9 --v 6.0 --style raw

Now, when you want to generate a new shot, your prompt becomes incredibly streamlined:

/imagine prompt: A close-up shot of a woman smiling in a sunlit cafe, holding a coffee cup --prefer_option brand-actor-v1

—

Phase 4: Transitioning from Static Midjourney Frames to Generative Motion

Generating a consistent static image is only 50% of the battle. The final step is translating these highly consistent Midjourney keyframes into dynamic video without losing character fidelity.

The Image-to-Video (I2V) Pipeline

1. Upscaling Protocols: Before sending your Midjourney image to a video generator, upscale it. Use Magnific AI or Topaz Gigapixel AI with “Face Recovery” features enabled. This sharpens skin textures, eye reflections, and fine details, preventing the video generator from muddying the actor’s face.
2. Motion Prompting in Gen-3/Luma: Upload your upscaled image as the first frame. Write a motion-focused prompt that describes *only* camera movement and micro-expressions. Do not redescribe the character.
   • Good Motion Prompt: “Slow push-in on the woman’s face as she subtly smiles, soft wind blowing her hair. Photorealistic, 8k, cinematic camera motion.”
   • Bad Motion Prompt: “A blonde woman in a cafe holding a cup of coffee and smiling.” (This invites the AI to re-interpret the character, causing facial drift).
3. Post-Generation Face Locking: If the video generator introduces minor facial warping over a 4-second clip, run the output through LivePortrait or FaceFusion, using your original Midjourney Anchor Asset as the source face. This locks down temporal consistency with pixel-perfect accuracy.

—

Pro-Tips for Enterprise Generative Workflows

—

Summary & Key Takeaways

Achieving brand actor consistency in generative video production requires a disciplined, structured approach to prompting and asset pipeline management. By implementing Midjourney v6’s --cref and --sref parameters, utilizing --cw 0 for wardrobe flexibility, and integrating high-end upscalers and face-locking tools, filmmakers can scale their creative output without sacrificing visual integrity.

As generative video tools continue to mature, those who master the art of the upstream asset pipeline will lead the industry in producing commercial-grade, narrative-driven AI cinema.