ur-grue Character Consistency Prompt
What This Skill Does
Writes a set of Midjourney prompts designed to maintain consistent character appearance — face, build, clothing, and presence — across multiple generated images for use in editorial series, podcast artwork, YouTube thumbnails, or visual narratives.
When To Use This Skill
- You are creating a series of images that all feature the same fictional character (for a podcast cover, YouTube series, editorial illustration run)
- You need a character to appear in multiple different scenes or contexts without changing their appearance
- You are illustrating a newsletter, blog, or social media series with a recurring visual persona
- You have already generated a character you like and want to generate additional images of the same person in new situations
What You Need To Provide
Required:
- A detailed description of the character: apparent age, gender presentation, build, distinctive physical features (hair, facial features, any identifying marks or accessories)
- The character's typical clothing or visual identity
- The style you want the images to be in (photorealistic, illustrated, graphic, painterly)
Optional:
- A reference image description (if you have already generated a version of this character you want to match)
- The scenes or situations you need them depicted in (list the images you need)
- Midjourney version (v6 has better face consistency than earlier versions; --cref flag is available in v6 for image-based character reference)
- Aspect ratio for intended use
- Any seed number from a prior successful generation
How the Assistant Approaches This
- Builds a dense, precise character anchor string — a consistent block of descriptive text that will be used verbatim in every prompt in the set, covering the character's appearance in enough detail that Midjourney has minimal room for variation
- Identifies the key consistency anchors: face structure, hair, signature clothing item or accessory — these are what must appear in every image
- Writes a base prompt for the first image, then generates scene-specific prompts that keep the character anchor string identical and vary only the environment, action, and lighting
- Recommends the
--seedworkflow: generate the first image, note the seed number from the Job ID, and use--seed [number]in subsequent prompts to bias toward the same character rendering - If Midjourney v6 is being used, notes where the
--cref [image URL]flag should be inserted to use an existing character image as a visual reference - Closes with a "Next Step" note: which prompt to run first (always the base portrait), how to record the seed number after the first generation, and whether to run cinematic-shot-prompt if the character also needs to appear in more complex scene compositions
Output Format
- Character anchor string: a labeled block of descriptive text ready to copy into every prompt
- Base prompt: the first full prompt to generate the character
- Scene variations: 3–5 additional prompts using the character anchor, each for a different situation
- Consistency workflow note: explains the seed and
--crefworkflow in plain language - All prompts formatted as complete, copy-paste-ready Midjourney strings
- Output ends with a "Next Step" note: run the base prompt first, record the seed, and run cinematic-shot-prompt if complex scene compositions are also needed
Quality Criteria
- The character anchor string covers at minimum: apparent age, gender, one distinctive facial feature, hair, and a consistent clothing item or signature visual element
- Every scene prompt uses the character anchor string verbatim — no variations in character description between prompts
- The anchor string is specific enough to distinguish this character from similar characters
- The consistency workflow is explained — users who don't know about
--seedand--crefmust be told - At least one prompt places the character in a non-portrait scene (action, environment) to test consistency beyond close-up shots
- Output includes at least one concrete next action the user can take immediately (run the base prompt, record the seed number, or run cinematic-shot-prompt for scene compositions)
Example
Input
- Character: A woman in her mid-40s, Black, natural shoulder-length hair with grey at the temples, warm brown eyes, slightly broad shoulders, consistently wears a dark navy blazer or jacket, often has a notepad or pen visible, confident and direct in expression — designed as the visual persona for a podcast about investigative journalism
- Style: Illustrated, semi-realistic — somewhere between editorial illustration and graphic novel, not photorealistic
- Scenes needed: Podcast cover art (portrait, looking at camera), working at a desk reviewing documents, standing outside a building, speaking into a microphone
- Format: Square (1:1) for podcast cover; 16:9 for banner; 4:5 for social
Output
CHARACTER ANCHOR STRING (Copy this block verbatim into every prompt)
Black woman in her mid-40s, natural shoulder-length hair with silver-grey at the temples, warm brown eyes, confident direct expression, wearing a dark navy blazer, editorial illustration style, semi-realistic graphic novel aesthetic, clean bold lines, muted professional color palette
BASE PROMPT — Podcast cover portrait
/imagine [CHARACTER ANCHOR STRING], close portrait, slight three-quarter angle, plain dark background, soft studio lighting, notepad visible in hand, podcast cover art composition, editorial illustration --ar 1:1 --s 200 --v 6
(Generate this first. Note the seed number from the Job ID — it will appear in the image filename or in your Midjourney history. Use that seed in all subsequent prompts.)
SCENE 1 — Working at a desk
/imagine [CHARACTER ANCHOR STRING], seated at a desk, reviewing documents spread across the surface, desk lamp light, late evening office interior, focused and intent, editorial illustration --ar 16:9 --s 200 --seed [YOUR SEED] --v 6
SCENE 2 — Standing outside a building
/imagine [CHARACTER ANCHOR STRING], standing outside a grey stone building, looking slightly upward, overcast daylight, professional and composed, background slightly desaturated, editorial illustration --ar 4:5 --s 200 --seed [YOUR SEED] --v 6
SCENE 3 — Speaking into a microphone
/imagine [CHARACTER ANCHOR STRING], standing at a microphone, mid-sentence, dynamic but not theatrical, studio or press conference setting, warm directional light, editorial illustration --ar 16:9 --s 200 --seed [YOUR SEED] --v 6
SCENE 4 — Action shot, walking and thinking
/imagine [CHARACTER ANCHOR STRING], walking along a city street, deep in thought, one hand raised slightly as if mid-thought, urban background out of focus, editorial illustration --ar 4:5 --s 200 --seed [YOUR SEED] --v 6
CONSISTENCY WORKFLOW
Step 1: Run the base prompt. Generate 4 variations (U1–U4). Choose the one that best captures the character.
Step 2: In your Midjourney history, find the Job ID of the image you chose. The seed number is embedded in the metadata — you can also react with the envelope emoji (✉️) to have Midjourney send you the job details including the seed.
Step 3: Add --seed [number] to every subsequent prompt. This does not lock the image — Midjourney is stochastic — but it strongly biases the generation toward a similar character rendering.
Step 4 (v6 only): Once you have a final character image you're happy with, use --cref [image URL] in subsequent prompts to provide a direct visual character reference. This is more reliable than seed alone for face consistency.
Note on variation: Even with seed and cref, some character drift between images is expected — especially in different poses and environments. Plan to generate 4+ variations of each scene and select the most consistent one.
NEGATIVE PROMPT (apply to all scenes):
--no photorealistic, no 3D render, no overly simplified cartoon, no inconsistent hair color, no different clothing
Known Limitations
- Midjourney does not have a deterministic character lock — even with
--seedand--cref, face and appearance will drift across images, especially in different poses, lighting conditions, and distances. Plan for selection work rather than perfect generation - The
--crefflag (character reference) is only available in Midjourney v6 and later — users on earlier versions must rely on the seed workflow, which is less reliable - This skill produces prompt sets; the user must manage the actual generation workflow (noting seeds, uploading reference images for cref) in the Midjourney interface
- For photorealistic character consistency, other tools (Stable Diffusion with ControlNet, or dedicated avatar generation tools) may outperform Midjourney in the current generation of models