Verified against Midjourney · 2026-07-24
Keep one character looking the same across comic panels with --cref
A character-reference workflow using Midjourney's --cref and --cw parameters to hold a specific character's face and build consistent across multiple panels of a comic or storyboard, instead of re-describing them from scratch each time and getting a different-looking person in every panel.
The prompt
Ready to copy — highlighted parts are example details you can swap.
REFERENCE IMAGE Start from one Midjourney-generated (or uploaded) image of the character that clearly shows their face and overall build — this becomes the anchor every later panel points back to. Its URL goes in https://cdn.midjourney.com/abc123-anchor-character.png. CHARACTER DESCRIPTION FOR THIS PANEL a young woman with short copper hair, a scar above her left eyebrow, wearing a weathered brown leather jacket WHAT IS DIFFERENT IN THIS PANEL crouching behind a rusted shipping container, peering around the edge in a rain-soaked shipping yard at night, sodium lights glowing overhead. CHARACTER WEIGHT --cw 100 — this controls how much of the reference image --cref pulls from. --cw 100 (the default) matches face, hair, and clothing from the reference. Lower it toward --cw 0-50 if this panel needs the same face but a different outfit or pose the reference image doesn't show, since --cw 100 will otherwise try to carry the reference's clothing into a scene where it doesn't belong. STYLE AND LINE CONSISTENCY gritty ink-and-wash graphic novel style, heavy shadow, muted desaturated color — state this the same way in every panel's prompt, not just the character reference, since --cref locks the character's likeness but does nothing to keep the surrounding rendering style consistent panel to panel; that is a separate, ordinary text instruction you have to repeat every time. WHAT TO AVOID --no text, speech bubbles, watermark, extra fingers PARAMETERS --cref https://cdn.midjourney.com/abc123-anchor-character.png --cw 100 --ar 16:9 --v 7 OUTPUT a young woman with short copper hair, a scar above her left eyebrow, wearing a weathered brown leather jacket, crouching behind a rusted shipping container, peering around the edge, in a rain-soaked shipping yard at night, sodium lights glowing overhead, gritty ink-and-wash graphic novel style, heavy shadow, muted desaturated color --cref https://cdn.midjourney.com/abc123-anchor-character.png --cw 100 --no text, speech bubbles, watermark, extra fingers --ar 16:9 --v 7 REPEAT FOR EVERY PANEL Reuse the exact same https://cdn.midjourney.com/abc123-anchor-character.png value across every panel in the sequence — swapping it out for a new anchor image partway through a comic is the single most common cause of a character's face subtly drifting halfway through a story. Only the action, setting, and, when the outfit genuinely changes, the --cw value should differ between panels; the reference URL and the art-style line should not.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
--cref conditions the generation on the actual visual features of a reference image — face structure, hair, general build — rather than on the text description of that character alone, which is the direct fix for the most common failure in multi-panel work: describing "a young woman with short copper hair and a scar" in five separate prompts and getting five plausibly-matching but visibly different faces, because Midjourney has no memory between generations and text descriptions alone are never specific enough to pin down a unique face across independent runs. Anchoring every panel to the identical reference image URL, rather than to the previous panel's output, matters because chaining reference images panel-to-panel compounds small drifts — each new image is a slightly different interpretation of the last one, and by panel five the character has visibly drifted from panel one. A single fixed anchor means every panel drifts from the same source by roughly the same small amount, instead of drifting cumulatively. --cw is the parameter that resolves the single biggest practical complaint about character reference: it "sticks" the reference outfit onto every panel, even panels where the story calls for a costume change. --cw 100 matches face, hair, and clothing together, because Midjourney has no separate control for "match the face but not the shirt" — the actual lever is to lower --cw toward 0-50, which biases the match toward facial features and away from the specific clothing in the reference, letting the panel-specific outfit description in the prompt text take over instead of getting silently overridden by the reference image's outfit. The instruction to restate the art-style line identically in every panel, rather than assuming --cref handles it, targets a real gap in what the parameter actually does: --cref is scoped to the character's likeness, not the surrounding rendering style, line weight, or color grading of the panel. Two panels can use the identical --cref value and --cw setting and still look like they were drawn by two different artists if the accompanying style text drifts even slightly between prompts — consistency of rendering style is achieved the ordinary way, by repeating the same words, not by any parameter.
What you get back
A sequence of panels where the same character — same face, same scar, same build — appears in different poses and settings, with clothing that follows the panel-specific description rather than defaulting back to the reference image once --cw is lowered for a costume change.
Verified against
Midjourney v7 · 2026-07-24
Changelog
- 2026-07-24 — Initial publish, verified against Midjourney v7 --cref/--cw for multi-panel character consistency.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
