Issue #10 | Platform: Midjourney v7 (primary) / Flux 2 Pro / GPT-Image-2 | May 20, 2026
Don’t stop halfway. At the end, you’ll find a free bonus download: 10 production-grade digital assets packed into one vault.
You generated a perfect character last week. Same prompt. New session. Completely different person.
This issue gives you the Character DNA framework, so every scene you generate looks like the same person, same world, same brand.
[02] THE FIRST 80 WORDS
AI image models are stateless. Every prompt is a fresh start. No memory of what came before. The word "character" means nothing to them without detailed visual data to anchor it. Most prompts describe appearance at the surface level: hair color, outfit, pose.
But models need structural anchors the specific combination of facial geometry, feature ratios, and visual markers that make someone recognizable from image to image. Without those anchors, you'll get a different face every time.
By the end of this issue, you'll have a repeatable system for locking in your character's visual DNA, so every scene you generate looks like the same person, same world, same brand.
[03] PLATFORM SPOTLIGHT MIDJOURNEY V7
Platform: Midjourney v7 by Midjourney, Inc. Launched: March 2025. The newest major version of the most widely used AI image platform. Built on a native image generation architecture with significantly improved prompt adherence, character coherence, and stylistic control over v6.1. Behavioral observations noted: March–May 2025. Both platforms update on irregular cycles. Verify before acting.
How Midjourney v7 interprets prompt language differently from others in scope
Midjourney v7 reads prompts holistically, it weights descriptive language about structure and relationship over surface-level adjectives. Unlike v6.1, it rewards detailed character descriptors and penalizes vague references. Example: "woman in red jacket" produces random results; "35-year-old East Asian woman, angular jawline, short blunt bob, structured crimson blazer with gold hardware" produces a consistent, repeatable face.
The one technical behavior that surprises most Midjourney v7 users
Character Reference Image (--cref) is Midjourney v7's standout capability for character consistency. Feed it a reference image URL and it locks onto the face, maintaining structural identity across different scenes, lighting, and poses. Works best when combined with a detailed text prompt. The higher the --cw (character weight) value, the stronger the face lock use --cw 100 for maximum fidelity.
The one parameter that produces the biggest quality jump in Midjourney v7
--cw (Character Weight) is Midjourney v7's most impactful consistency parameter. It controls how strongly the model anchors to the reference face from your --cref image. At --cw 0, only style is transferred; at --cw 100, full facial structure is locked. Example 1: Same --cref image at --cw 50 gives flexible resemblance. Example 2: Same --cref at --cw 100 locks the face precisely, even across different outfits, backgrounds, and lighting.
What Midjourney v7 does better than any other model in scope
Midjourney v7 leads in photorealistic character portraiture, editorial fashion imagery, and cinematic scene-building with human subjects. Its highest-value use case for creators: generating a consistent character across 10+ scenes with recognizable facial identity, something no other model does as reliably at this quality level.
What Midjourney v7 consistently fails at
Midjourney v7 struggles with consistent hand anatomy and text rendering in scene-integrated characters. Its main failure mode for character work: when a scene is compositionally complex (e.g., crowded background, extreme angles), facial coherence degrades even with --cref active. Flux 2 Pro handles complex scene composition with characters more reliably when face-lock fidelity is less critical.
INCOME SIGNAL: Character consistency is a paid skill. Brand managers, indie authors, and content studios are actively seeking creators who can produce a recognizable character across 10–20 scenes. Deliverable: a character image pack (10 scenes, 1 character). Client type: personal brand, fiction author, product brand. Revenue model: $150–$500 per pack, retainer for ongoing content.
[04] THE TECHNIQUE: The Character DNA Framework
DIAGNOSIS
Most creators treat character prompts like costume descriptions. They list what the character is wearing and what color their hair is, and wonder why every generation looks like a different person wearing the same outfit. The actual problem: they're describing a character's surface, not their structure. Without anchoring the model to facial geometry and identity markers, you'll never get the same face twice.
MECHANISM
The Character DNA Framework works by separating a character into four layers: Facial Architecture (bone structure, feature ratios, unique markers), Identity Signals (hairstyle, skin tone, eye shape), Signature Elements (outfit components, accessories, brand markers), and Scene Context (lighting, setting, pose). By explicitly encoding all four layers in every prompt, and pairing it with a --cref reference image in Midjourney v7, you give the model enough structural data to reconstruct the same person reliably across any scene.
Note: Observations based on testing conducted in Midjourney v7 and Flux 2 Pro. Platform behavior changes frequently. Always test with current model version before client delivery.
[05] THE STRUCTURE - WHERE THIS TECHNIQUE LIVES
Where the Character DNA Framework lives in the Five-Part Prompt Structure framework:
Before Composition: The Character DNA Framework applies at Layer 1 (Subject/Character Definition), before you write any scene, action, or style instruction. In Midjourney v7, pair your Character DNA prompt block with --cref [image URL] --cw 100 for maximum consistency. Rebuild the full DNA block for every new scene; never assume the model remembers.
BEFORE / AFTER
BEFORE FAILS (Viewer Language):
female CEO, red jacket, confident expression, looking at camera
Before Technical Parameters: In Midjourney v7, append --cref [your reference image URL] --cw 100 --ar 2:3 --v 7 to lock character identity. For Flux 2 Pro, use img2img workflow with IP-Adapter at strength 0.7–0.9.
AFTER, Midjourney v7 (optimized with the Character DNA Framework):
[CHARACTER DNA BLOCK]
FACIAL ARCHITECTURE: 35-year-old East Asian woman, angular jawline, high cheekbones, almond-shaped dark brown eyes, straight nose bridge
IDENTITY SIGNALS: short blunt black bob, light golden skin tone, minimal makeup
SIGNATURE ELEMENTS: structured crimson blazer with gold-tipped lapels, single pearl earring, no rings
SCENE CONTEXT: standing at floor-to-ceiling boardroom window, late afternoon directional light, sharp shadow contrast
--cref [reference_image_url] --cw 100 --ar 2:3 --v 7
AFTER, Flux 2 Pro (optimized with the Character DNA Framework):
[CHARACTER DNA BLOCK]
FACIAL ARCHITECTURE: 35-year-old East Asian woman, angular jawline, high cheekbones, almond-shaped dark brown eyes, straight nose bridge
IDENTITY SIGNALS: short blunt black bob, light golden skin tone, minimal makeup
SIGNATURE ELEMENTS: structured crimson blazer with gold-tipped lapels, single pearl earring, no rings
SCENE CONTEXT: standing at floor-to-ceiling boardroom window, late afternoon directional light, sharp shadow contrast
[IP-Adapter: reference_image, strength 0.80] [aspect ratio: 2:3]
WHAT TO WATCH FOR
Three visual signals that confirm the technique worked:
The same facial structure appears across multiple generations, same bone angles, same eye shape, same jawline definition, even when the background, outfit, and lighting change.
The signature outfit element (crimson blazer, gold lapels) carries over accurately into each new scene same color, same cut, same hardware detail without being re-described in scene-specific terms.
When you swap only the SCENE CONTEXT line while keeping the full DNA block identical, the character's face, hair, and outfit remain locked. The model successfully transferred visual identity to a completely new environment with zero facial drift.
[06] PROMPT VAULT ENTRY
VAULT ID: PTP-CHARACTER-CONSISTENCY-010
PLATFORM: Midjourney v7 (primary) / Flux 2 Pro (secondary variant)
TECHNIQUE: The Character DNA Framework
QUALITY SCORE: [9/10]
TAGS: character-consistency, portrait, brand-character, --cref, facial-lock, commercial-ready
---
PROMPT (Midjourney v7 primary)
[CHARACTER DNA BLOCK]
FACIAL ARCHITECTURE: [age]-year-old [ethnicity] [gender], [jawline shape], [cheekbone], [eye shape and color], [nose bridge]
IDENTITY SIGNALS: [hairstyle and color], [skin tone], [defining feature]
SIGNATURE ELEMENTS: [outfit description with specific details], [accessory 1], [accessory 2]
SCENE CONTEXT: [location and setting], [lighting type and direction], [mood/atmosphere]
--cref [reference_image_url] --cw 100 --ar [ratio] --v 7
---
VARIABLES
[VARIABLE 1] Replace "[age]-year-old [ethnicity] [gender]" with your character's specific demographic.
[VARIABLE 2] Replace "[hairstyle and color]" with your character's signature hair descriptor.
[VARIABLE 3] Adjust "--cw 100" to --cw 50-80 for looser resemblance, keep at 100 for strict face-lock.
[VARIABLE 4] Adjust "--ar [ratio]" for portrait (2:3), landscape (3:2), or square (1:1) compositions.
---
USAGE NOTES
Best for: Brand mascot packs, fiction character series, personal brand imagery. Run your first generation at --cw 100, then test at --cw 75 for scenes with complex composition where full lock may fight the scene geometry. Always rebuild the full DNA block per generation.
---
INCOME PATHWAY NOTE
This vault entry produces a sellable deliverable: a 10-scene character image pack. Target clients: indie authors (book covers + social), personal brands (consistent content character), product brands (mascot series). Price anchor: $150–$500 per pack depending on client scope.Platform-honest bottom line: For character consistency work, Midjourney v7 wins for photorealistic face-lock via --cref + --cw 100. Flux 2 Pro becomes the better choice when compositional complexity outweighs strict facial fidelity.
[07] INCOME SPOTLIGHT CALLOUT
THE INCOME SIGNAL: Character consistency is the single most commercially valuable AI image skill in the current market. The ability to produce a recognizable character across multiple scenes, same face, same identity, different context, is exactly what brand teams, indie authors, and content studios cannot do reliably on their own.
MINIMUM VIABLE WORKFLOW: Build one character using the Character DNA Framework. Generate 5 scenes using Midjourney v7 with --cref + --cw 100. Package as a PDF or Drive folder. Price at $150-$250 for a starter pack. Offer a 10-scene expansion for $350-$500.
PLATFORM MATCH: Midjourney v7 for photorealistic characters. Flux 2 Pro for complex scene characters. GPT-Image-2 for illustrated/stylized brand mascots.
MINIMUM VIABLE WORKFLOW: Build one character using the Character DNA Framework. Generate 5 scenes using Midjourney v7 with --cref + --cw 100. Package as a PDF or Drive folder. Price at $150–$250 for a starter pack. Offer a 10-scene expansion for $350–$500.
[08] YOUR ASSIGNMENT BEFORE THE NEXT ISSUE
TASK: Build your first character using the Character DNA Framework. Create the four-layer DNA block (Facial Architecture, Identity Signals, Signature Elements, Scene Context). Generate your character in three different scenes using Midjourney v7 with --cref + --cw 100. The character must be visually recognizable across all three scenes.
THE SUCCESS SIGNAL: When you look at all three generated scenes side by side, you can immediately recognize it's the same person, same facial structure, same signature outfit element, same hair. Someone who didn't know your prompt should be able to say "that's the same character" without being told.
THE FAILURE SIGNAL: Your three scenes show three noticeably different faces. The most common error mode: you used a vague surface description ("woman with dark hair, red jacket") without full Facial Architecture encoding, and didn't include a --cref reference. Prompt fix: Go back to the DNA framework, add the full Facial Architecture layer with geometric descriptors, and attach your best generation as --cref image URL with --cw 100.
INCOME TEST: If your three-scene character set passes the success signal, take this one commercial action: Save your three best outputs in a folder labeled "[Character Name]. Sample Pack." Post one image on LinkedIn or your portfolio with the caption: "Available: Custom character image packs starting at $150. Same face. Different scenes. DM to order." That's your first income gate.