jordangaston Consistent Characters
A three-step prompt workflow for generating the same character across many images. The core move: create the character once, lock their identity into a multi-view character sheet, and reference that sheet in every future generation instead of re-describing the character in words (which drifts).
Works with any image model that accepts reference images alongside a text prompt.
Conventions
- 🔒 marks a locked section. Copy it verbatim — the exact wording is what keeps results consistent. Unlocked sections are yours to fill in or adapt.
- Reference images are attached to the prompt. Refer to them in the prompt text exactly the way your model expects (some models use tags like
@img1, others accept plain phrases like "the first attached image"). The templates below use plain phrases — swap in your model's syntax if it has one. - Generate the character image in 9:16 portrait so the full body fits the frame.
Step 1: Create the character (pure text)
Generate the master character image from a text description alone.
🔒 [IMAGE TYPE: Full-length fashion photo]
[SUBJECT: enter your subject details here]
🔒 [POSE: Neutral standing pose, front view. Direct gaze at camera — expression neutral, composed, slightly guarded.]
[OUTFIT: enter your outfit details here]
🔒 [ENVIRONMENT & LIGHTING: pure white seamless background. High-key lighting, minimal cast shadow on the floor.]
🔒 [CAMERA & FILM STOCK: shot on Hasselblad H6D, 85mm lens, f/8, sharp focus, Kodak Portra 160 film stock, fine grain, accurate skin tones, commercial fashion photography style]
Step 1 (alt): Create the character from a headshot or clipped shot
If you already have a face — a headshot, a partial photo, or a crop — attach it as the reference image and let it carry the subject:
🔒 [IMAGE TYPE: Full-length fashion photo]
[SUBJECT: the attached reference image]
🔒 [POSE: Neutral standing pose, front view.]
[OUTFIT: add your outfit description here]
🔒 [ENVIRONMENT & LIGHTING: pure white seamless background. High-key lighting, minimal cast shadow on the floor.]
🔒 [CAMERA & FILM STOCK: shot on Hasselblad H6D, 85mm lens, f/8, sharp focus, Kodak Portra 160 film stock, fine grain, accurate skin tones, commercial fashion photography style]
Either way, the output is a single full-length, front-view character image on white. This is your character image.
Step 2: Build the character sheet (composition prompt)
Attach two reference images:
- The composition template — a five-panel model-sheet layout. Use the bundled one at
assets/character-sheet-template.png: ECU (head & shoulders), right side profile (upper body), left side profile (upper body), full shot (front), full shot (back). It controls layout only. - Your character image from Step 1. It controls identity only — face, skin, body, hair, makeup, expression, outfit.
🔒 [IMAGE TYPE: Fashion photo model sheet collage in **exactly** the same 5-shot layout and composition as the attached template image.]
🔒 [SUBJECT: Use the attached character photo as the ONLY reference for the character's exact face, skin details, body type, hair, makeup, expression, and full outfit/accessories — apply them identically and with perfect consistency across all five views.]
🔒 [ENVIRONMENT & LIGHTING: pure white seamless background. High-key lighting, minimal cast shadow on the floor.]
🔒 [CAMERA & FILM STOCK: shot on Hasselblad H6D, 85mm lens, f/8, sharp focus, Kodak Portra 160 film stock, fine grain, accurate skin tones, commercial fashion photography style]
🔒 [DETAILS: Perfect anatomical and clothing consistency across every angle, high-end editorial look, no text, no logos, no graphics.]
The output is your character sheet: one image showing the character from five angles.
Step 3: Place the character in scenes (scene prompt)
Attach your character image from Step 1 as the reference and describe the scene. When the scene needs an angle beyond front view (back shots, profiles), attach the Step 2 character sheet instead — it gives the model every angle to draw from.
[IMAGE TYPE: e.g., cinematic film still, dramatic photograph, moody portrait, etc.]
🔒 [CHARACTER REFERENCE: the attached character image]
[SUBJECT ACTION: e.g., standing still with hand on hip, walking slowly through grass, turning to look over the shoulder, etc.]
[SHOT TYPE: e.g., full-length, medium shot, wide establishing shot, close-up portrait, three-quarter length, etc.]
[CAMERA ANGLE: e.g., eye-level front view]
[LENS: e.g., 35mm wide-angle f/4 deep depth of field]
[LIGHTING: describe direction, quality & mood]
[ENVIRONMENT: full background description]
[COLOR GRADE: overall color treatment]
[MOOD & ATMOSPHERE: emotional feeling]
[TECHNICAL SPECS: camera, film emulation & effects]
Why this works
Text descriptions of a character drift: every generation reinterprets "red-haired woman in a cream dress." Reference images don't. The bracket structure separates concerns — layout, identity, environment, camera — so the model doesn't blend them, and the locked sections keep the parts that anchor consistency identical across every generation.
Pitfalls
- Don't re-describe the character in scene prompts. The moment you add "a woman with auburn hair," you compete with the reference image and reintroduce drift. Let the reference carry identity.
- One character per sheet. Multiple people in the character image confuses the identity binding.
- If a view drifts (usually the back view), regenerate rather than accepting it — every future image inherits the sheet's errors.
- Outfit changes come later. Lock the base character first; generate outfit variants in follow-up prompts using the sheet as reference.