anima-composition-director

name: anima-composition-director description: Convert Anima image-generation intent, tags, references, or rough prompts into concrete composition decisions. Use with comfyui-animatool when the task needs better camera framing, aspect ratio, subject placement, lighting, value contrast, depth of field, face readability, emotional staging, storytelling composition, symbolic visual metaphor, or reference-image composition transfer before assembling Anima prompts.

Anima Composition Director

Goal

将生图意图转成视觉计划：画布、镜头、主体位置、层次、光源、景深。

Do not write literary mood paragraphs. Describe the picture layout.

Inputs to consider

• User hard constraints: size, aspect ratio, platform use, reference image scope.
• Subject: number of characters, body visibility, action, props, outfit complexity.
• Scene: indoor/outdoor, close space/wide space, background importance.
• Identity risk: whether face, hair, outfit, or emblem must stay readable.
• Output purpose: quick test, wallpaper, cover-like key visual, character sheet, interaction scene.

读取导航

Canvas fit

Choose after the semantic draft is clear. If the user gave a size, keep it and adapt composition.

规则：用户允许某分辨率 ≠ 所有图都用该分辨率。每张图按下表按构图语义独立选择画布，禁止一刀切复用。

Do not recommend 1920x1080 for initial Anima base1.0 generation. Larger output belongs to upscale.

Layout modes

Choose one image type before camera and canvas decisions. Do not mix modes unless the user asks.

Mode rules:

• illustration is the broad category; specify a narrower mode when composition needs it.
• event CG needs a readable story moment, not prose or backstory.
• manga single panel is one image only; do not describe multiple panels unless requested.
• cinematic still may borrow film shot terms, but must stay a static frame.
• If unsure, default to character illustration for simple character requests and event CG for interaction scenes.

Camera grammar

每行选一个值，除非用户明确需要特殊镜头：

• Distance: close-up, upper body, cowboy shot, full body, wide shot.
• Angle: eye-level, low front angle, high angle, side view, three-quarter view, over-shoulder view, top-down view, bird's-eye view, aerial view, pov, first-person view.
• Lens feel: normal perspective by default; use wide-angle only for strong space or action. Avoid fisheye unless requested.
• Focus technique: shallow depth of field, deep focus, rack focus look, or soft background blur.
• Face rule: if identity matters, include Keep the face sharp and readable.

Avoid contradictions: no close-up with full body; no from above with from below; no wide shot if the face must dominate.

• One frame gets one primary camera idea; do not stack low angle, top-down, and over-shoulder together.
• Convert movement terms into static layout: tracking shot means subject offset plus background leading lines; push-in means closer framing and stronger face emphasis; orbit means three-quarter view with curved background cues.
• For top-down, bird's-eye view, or aerial view, show readable ground layout with roads, rooftops, fields, shadows, crowds, or one landmark.
• For foreshortened POV or figure-emphasis framing, use foreground limb/prop/fabric, believable joints, silhouette, crop, or rim light while preserving identity anchors.
• For action, describe the peak pose and motion direction, not a sequence of events.

Composition patterns

Use one clear pattern:

• Center: stable portrait, icon, square image, character focus.
• Rule of thirds: character plus readable environment, poster-like balance.
• Diagonal: action, weapons, movement, falling, chase, dynamic pose.
• Layered depth: foreground object, midground subject, background scene.
• Negative space: title area, sky, empty corridor, visual breathing room.
• Symmetry: ritual, shrine, throne, formal scene, stillness.

State subject placement and background direction. Example: Place the subject slightly right of center, with the corridor receding to the left.

Scene coherence:

• Pick one story anchor: character action, prop interaction, weather effect, or location function.
• Pose, expression, outfit, prop, weather, and background must support that anchor; remove details that tell a different story.
• If the environment matters, show how the character uses or reacts to it, not just where they stand.
• Emotion must become visible staging: expression, gaze, hand pose, body direction, distance, occlusion, or light placement.
• Symbolism must become one visible motif, prop, flower, shadow shape, framing device, or background echo; do not explain the metaphor.
• For story contrast, use visible juxtaposition or purposeful negative space; keep it guided by gaze, light, or background shape.

Framing safety:

• Lead room: if a subject looks or moves sideways, leave open space in that direction unless the user wants cramped tension.
• Headroom: keep small space above the head in portraits; use extra top space only for sky, title, or scale.
• Crop safety: avoid accidental cuts through wrists, elbows, knees, ankles, hands, or feet; crop wider or clearly between joints.
• Foreground framing: use doorframes, windows, branches, or props only when they guide attention to the subject.
• Avoid tangents: do not let character outlines, props, or background edges just touch; separate them or overlap clearly.

Visual-design rules:

• Establish one focal point first; secondary props and background must support it.
• Keep face and hands readable; move overlaps away from them.
• Put the clearest silhouette against the simplest background area.
• Standing or walking subjects need a ground contact cue: feet on floor, road, grass, contact shadow, cast shadow, or footprint.
• When clothing is complex, simplify the background and keep the silhouette readable.
• Use clear value separation, controlled edges, and one dominant color palette plus one accent color.

Lighting and depth

Define light as visible geometry, not abstract mood:

• Key light direction: left / right / above / below / behind / window side.
• Rim light only when it helps silhouette separation.
• Fill light only when shadows hide the face or outfit identity.
• Background light should not overpower the face.
• For dramatic contrast, choose one readable value plan: face-lit dark background, split light, backlit silhouette, or rim-only separation.
• If shadows cross the face, keep at least one eye readable unless concealment is requested.
• Use background blur when scene detail competes with identity.
• For 2:1 or wide shots, explicitly protect face readability.
• Avoid stacking many post-process words; pick one: bloom, vignette, lens flare, film grain, or chromatic aberration.
• Use vignette only to guide the eye toward the subject, not as a default style word.

Reference image transfer

If a reference image is used only for composition, extract only:

• aspect ratio
• camera distance
• camera angle
• subject position
• depth layers
• light direction
• blur / focus behavior

Do not copy reference character, outfit, color scheme, props, or setting unless the user asks.

Output contract

Return or pass forward this compact structure:

{
  "prompt_semantic_draft": "1girl full body, classroom window, quiet pose, soft daylight",
  "canvas_fit": "1536x1024 horizontal composition for environment and subject placement",
  "final_composition": "subject slightly right of center; window and classroom depth open to the left",
  "canvas": {
    "width": 1536,
    "height": 1024,
    "reason": "horizontal interaction scene"
  },
  "camera": "full body, low front angle, normal perspective",
  "composition": "subject slightly right of center; background opens to the left",
  "lighting": "soft window light from the left, subtle rim light",
  "focus": "face sharp and readable, background softly blurred",
  "nltags_sentences": [
    "Place her full body slightly right of center.",
    "Use a low front camera angle with normal perspective.",
    "Keep her face sharp and readable, with a softly blurred background."
  ]
}

Field rules:

• prompt_semantic_draft: compact visual content summary before tag assembly.
• canvas_fit: why this canvas matches the draft; mention conflicts if adapting a user-fixed size.
• final_composition: the final layout sentence that must match width and height.

Rules for nltags_sentences:

• 2–5 sentences.
• 8–18 English words per sentence.
• One sentence controls one thing: pose, camera, placement, lighting, depth, or face quality.
• Use at most one camera term sentence and at most one focus/depth sentence.
• No explanatory metaphors, backstory, destiny, personality analysis, or vague mood stacking.
• Symbolic requests are allowed only as visible motif, staging, or light contrast, not as explanatory prose.
• No video-only instructions such as the camera pans, the camera tracks, or then she turns.
• Prefer concrete verbs: place, use, keep, frame, light, blur.

Self-check

Before returning the plan:

• Canvas matches the final prompt idea.
• Subject size matches camera distance.
• Face readability is protected when identity matters.
• Standing subjects do not float; aerial scenes have ground layout and scale cues.
• Background detail does not fight the subject.
• nltags_sentences are short layout controls, not prose.