AI Short-Drama Image Roles: Character, Setting, Composition & Style

Why “Use This Image as a Reference” Can Go Wrong

One image may contain a face, clothing, pose, background, text and lighting. If you only say “follow this image,” the next person or tool cannot tell which features must remain as identity anchors and which are incidental to that particular shot. The result may preserve the face but also carry over an unwanted coat, background sign or composition.

For every reference, state: what or who it depicts, the role it serves this time, which visible details it may control, and which details must not carry over. A file’s existence does not mean it has been inspected. Without authorization to inspect it, mark watermark, text and cropping risks as unverified.

Nine Common Reference-Image Roles

If a character’s look stays the same throughout this episode, it can be recorded with identity. A changing state, such as a different outfit, wetness or damage, should be bound separately so an old state does not carry into later shots.

A Reusable Reference-Image Role Card

Reference name / location: ___
Role this time: identity / appearance state / geography / composition / scale / effect / opening frame / closing frame / style
May control: ___
Must not control: ___
Inspection status: authorized and observed: ___ / creator’s description only: ___ / unverified
If references conflict, preserve first: ___

If the same image provides both identity and composition, write two role cards: the identity card is only for recognizing the character; the composition card is only for the frame layout. Separating them lets you change the composition while keeping the character, or reuse a suitable layout with a different character.

The card above is an author worksheet. A skill handoff separately records a stable slot, explicit order, image locator, and name for each role; the worksheet is not the complete native record. Supplied real images use REF slots. Images the creator will mount at generation time, currently identified through an established image-prompt or shot entry, use PLAN slots. References with an actual project image path use REF records. IMG identifies a prompt, not proof of a generated image. Replacing an image updates the locator and boundaries in its existing slot; inserting another must not silently change the old slot's meaning.

Original Example: Three References at a Coastal Weather Station

Shot: New observer Xu Lan discovers that one corner of the wind-record sheet has been torn away in the night-shift room, then walks to the window to check the anemometer tower outside.

Describe the opening frame separately: “The record sheet is still on the desk, its corner is already torn, and Xu Lan has not yet walked to the window.” If the style sample contains another person wearing a raincoat, do not bring them into the weather station just because you are referencing its style. The bright background on the character sheet must not override the night-shift room’s lighting.

This example registers the character, geography, and style roles in that order; having three images alone does not establish completed inspection. The badge shape is included in the identity slot only because it stays unchanged throughout this example's episode, not as a permanent bodily feature. Its text follows the separately established text policy.

Filled example: separate binding notes from the image prompt

Use Xu Lan’s night-shift room again. For this filled example, define the location layout: door screen left, record desk center, window right and tower outside it; Xu stands by the desk. The paper corner is already torn and she has not reached the window. This paper plan needs real images; it is not evidence that three media inputs were attached.

Binding notes for the creator

The character image controls Xu Lan’s short wavy hair, brow shape, build and episode-stable badge form, without its daylight background, pose or studio illumination. The location image controls door left, desk center, window right and tower outside, without temporary people, torn paper or plot action. The night-sea sample supplies warm/cool hierarchy, damp materials and shadow edges, without its raincoat figure, signs or coastal layout. Record character, location and style in that order, then bind each through the target’s supported method.

These are author notes. A real project also needs actual image locations, names, inspection status and stable records. If a target cannot consume all these roles together, the text is not a universal parameter. A supported image stage may first produce the shot’s opening image, which can be inspected and used through the video target’s real input method.

Readable opening-image prompt

Medium view of the night-shift room. Xu Lan stands beside the central record desk, with the established short wavy hair, brow shape, build and current badge form. She looks down at the wind-record sheet on the desk; its corner is already torn and the sheet stays on the surface. The door is screen left, the window right and the tower outside. Xu has not yet walked to the window. Use the established night-sea warm/cool hierarchy, damp materials and shadow edges, while retaining this scene’s actual light sources rather than the character board’s daylight studio light. Make the torn paper corner the visual focus. Keep out additional people and the sample’s signs.

The prompt freezes the beginning. Walking to the window belongs to later motion rather than an attractive but premature start image. Filenames, inspection records, reference order and previous-failure notes stay outside the text describing what should be drawn.

This example provides neither wind values nor verbatim badge text. Follow the project’s established text policy rather than inventing values to fill the worksheet. If viewers must read a number, obtain the exact wording and readable/post-production choice before adding its surface requirement. Do not erase that duty with a global no-text instruction.

Inspect the actual image by role. Correct identity with a reversed door suggests checking geography and composition; correct layout with an extra raincoat figure suggests checking whether the style sample was copied wholesale. Also verify the submitted inputs. The result alone does not establish a unique cause.

How to Resolve Conflicts Between References

These causes are diagnostic leads rather than conclusions established by one output. Compare role notes, submitted inputs and the revised candidate before choosing a reference, composition or other setting change.

FAQ

Can one image be used for both a character and composition reference?

Yes, but state both roles separately and list what each may and must not control. Do not use “reference everything” to let the background, clothing, text and pose carry over beyond their assigned roles.

Can a negative prompt remove a watermark from a reference image?

You cannot rely on that. An exclusion phrase does not automatically erase pixels already present in the reference. First confirm the asset’s condition through authorized inspection, and use a clean, properly sourced reference.

Related reading

• AI short drama character consistency across shots→
• Image-to-Video Prompts: Reference Subtraction + 8 Motion Problems→
• How to write a storyboard script: shot list template for short drama→
• Drama Skills→