Keeping a Character Consistent Across Twelve Shots: A Guide to Image Generation

Why character identity drifts between images

A character can look convincing in a single generated image and still become difficult to recognise across a sequence. Small changes accumulate: the hair becomes shorter, clothing changes colour, facial proportions shift or an accessory disappears. Each image may look acceptable in isolation, but the differences become obvious when the pictures are placed consecutively in a vertical video.

The underlying production problem is that an image prompt describes a requested picture, not necessarily a persistent character identity. A tool may interpret the same descriptive language differently each time. Reference-image inputs, image editing, seed controls and other consistency features can help, but their availability and behaviour depend on the particular product and its current capabilities.

The solution begins before image generation. Establish which details make the character recognisable, decide which details can change and organise the twelve shots around a single visual specification. This gives the image workflow a defined continuity target.

Build a character specification before creating scenes

Write a character description that is precise enough to distinguish the subject from a generic person. Include approximate age, face shape, skin tone, hairstyle, hair colour, eye characteristics, clothing, accessories and distinctive features. Avoid relying on subjective descriptions such as attractive, friendly or interesting. They offer little control over visual identity.

For example, a fictional presenter might have a rounded face, short dark curly hair, a small beauty mark beneath the left eye, a mustard-yellow jacket and a narrow black watch on the right wrist. These details form an identity specification. They should remain stable throughout the sequence unless a scene explicitly requires a change.

Separate permanent characteristics from scene-specific details. Hair colour and facial structure may be permanent. Expression, posture, camera distance and background are variable. Clothing may remain fixed for a single scene or production, while a change of outfit should be documented as a deliberate continuity decision.

• Permanent identity: facial structure, hair, skin tone and distinguishing marks.
• Wardrobe: colours, garment shapes, materials and accessories.
• Variable performance: expression, pose, gesture and direction of gaze.
• Environment: location, lighting, background and camera position.

Create a character sheet that works as a visual reference

A character sheet translates the written specification into a visual standard. Aim to establish a clear front view, a three-quarter view, a side profile and at least two facial expressions. Include a full-body image when clothing, height or body proportions matter to the production. A consistent neutral background makes it easier to inspect the subject without environmental distractions.

The character sheet should establish the relationship between facial features and the rest of the design. A profile can reveal the nose shape and hairline, while a three-quarter view can clarify the jaw and cheek structure. Full-body references help identify changes in garment length, footwear and proportions that may not be visible in portrait images.

Keep one approved reference set in a dedicated project folder. Use descriptive filenames such as character-front, character-profile and character-full-body. If a new reference replaces an earlier design, record the change and update the production materials. Multiple competing reference sheets can create ambiguity during later image generation.

Plan the twelve shots as a continuity sequence

A twelve-shot sequence needs a visual plan. For example, a one-minute fictional introduction could contain an opening portrait, a wide establishing view, a medium shot, a close-up, two action frames, a reaction, a profile, an environmental shot, a second close-up, a concluding medium shot and a final portrait. This structure provides variety without requiring the character to change identity.

Record the intended camera distance and subject position for each shot. A close-up needs different reference information from a full-body image. The character's face may occupy most of the frame in one image and only a small part of another. Establishing these differences in advance helps identify where a reference image needs to be supplemented by another angle.

Also record continuity between adjacent shots. If the character holds a cup in the fourth image, decide whether the cup remains present in the fifth. If the character looks towards the right edge of one frame, the next frame should respect that direction unless the edit intentionally changes the viewpoint. Small physical details make a sequence feel connected.

• Shots 1-3: establish identity, setting and initial composition.
• Shots 4-6: introduce close-ups and purposeful action.
• Shots 7-9: vary expression, angle and environment.
• Shots 10-12: return to recognisable framing and resolve the sequence.

Control prompts by separating identity from action

Use a stable identity description in every image request. Add a separate scene description that specifies the action, composition, lighting and camera position. This makes the prompt easier to maintain and revise. The identity description should not be rewritten casually for every shot, because even small wording changes can introduce unnecessary variation.

For example, the fixed character description can identify the same mustard jacket, short dark curls and beauty mark. The scene instruction can then request a close-up with a concerned expression in a softly lit kitchen. The next image can use the same identity description with a full-body composition in a garden. Only the scene variables change.

Where supported, supply the approved character reference image with each request. Some tools provide reference-strength settings, image guidance or character-specific controls. Their labels and effects vary, so consult the product documentation before relying on a particular setting. Strong reference guidance may improve identity retention but can also reduce flexibility in pose or composition.

Inspect visual continuity before moving to the edit

Review all twelve images together in shot order. This is more useful than assessing each output independently. Compare the face, hairline, wardrobe, body proportions, accessories and lighting. Look for details that change without an editorial reason. A missing watch or a suddenly different jacket can be more distracting than a modest difference in background colour.

A practical continuity table can record the shot number, approved reference, identity match, wardrobe match and correction required. Use simple descriptions such as retain, replace or revise. Avoid changing an entire image when only one detail needs attention. An editing tool may be able to correct a local area while preserving the rest of the composition.

Some corrections are best handled through image editing; others require a new generation request. If the face is substantially different, repeated small edits may be inefficient. If the only problem is an accessory, a targeted correction can be sufficient. Keep the approved images separate from rejected alternatives so that the editor cannot accidentally assemble an outdated version.

• Check facial structure and distinctive marks.
• Compare clothing and accessories across adjacent frames.
• Confirm that changes in pose and expression are intentional.
• Check whether lighting changes create apparent identity differences.
• Replace images with substantial continuity problems before editing.

Prepare the consistent images for vertical video

The final image sequence needs to work at the actual delivery dimensions. A portrait composition may place the character close to the top of the frame, leaving insufficient space for captions. A full-body composition may become too small when viewed on a mobile screen. Check each image inside a 9:16 editing canvas before committing to the final sequence.

Use a consistent visual treatment across the twelve shots. This includes colour grading, contrast, grain, background treatment and any artificial camera movement added in the editor. Even consistent character images can look unrelated when one is heavily sharpened, another is unusually warm and a third has a different contrast profile.

Keep the original image files and the final cropped versions. If the edit changes later, the original files provide more flexibility for reframing. Avoid applying irreversible crops to the only available copy of an image. A simple asset folder containing references, approved images, alternates and final crops will make subsequent revisions easier.

Questions readers ask