Turn user-provided novel material into short-video scenes that are sourced, causally coherent, and production-ready. Here, “distillation” refers to editorial distillation; the output consists of selected segments and production rationale, not model training, video generation, or traffic prediction.

Prioritize the chapters, character settings, target duration, adaptation latitude, spoiler scope, and reference materials already provided. By default, target approximately 30 seconds, remain faithful to source events, and allow clearly labeled audiovisual-expression suggestions; adjust when the user specifies otherwise.

Segment by complete events rather than cutting at a fixed word count. Long-form works can be processed chapter by chapter; carry forward necessary character goals, ability conditions, unresolved events, and their sources across batches; record the actual scope of each batch. Use event cards for initial screening; do not write a full storyboard for every chapter of an entire book.

Each event card records:

Distinguish “worldbuilding setting,” “action description,” and “complete event.” A passage containing only words such as imposing pressure or dharma form is not enough to form a usable event. Before finalizing, reread the selected passages and any necessary surrounding context, and check the initial screening summaries for omissions and errors.

First verify the source, causality, and how much explanation is needed for a short video. Temporarily defer candidates missing a critical antecedent; do not fill the gap with invented rules and then present the result as source material.

Show at most three representative candidates by default. For each dimension below, assign a score of 0, 1, or 2 and provide a one-sentence rationale.

The total score is only an editorial ranking aid for the current candidates, not a measure of quality or viral-hit probability. Fidelity to the source and user preferences take priority over the total; explain the trade-offs behind the final choice. For non-action genres, look for turns involving discovery, decision, or relationship change; do not force a fight or counter-kill.

Write one sentence: Because of ___, in pursuit of ___, the character confronts ___, takes ___, and causes ___.

Build the short video around one primary goal, one primary causal line, and one key turning point; longer durations can accommodate more. When compressing, prioritize removing repeated clashes, names of secondary characters, explanatory realm terminology, and background that does not affect the current event.

Preserve why an action occurs, its object or target, key ability rules, and the information needed to understand the result. If the source explicitly states a cost and that cost determines the character’s choice, preserve it; do not invent a lifespan cost or similar setting for an ability that has no stated cost.

Distinguish between whether the event currently unfolding is understandable and whether later secrets are revealed. You may conceal a future identity, final outcome, or later consequences, but do not erase motivation for the sake of suspense. If removing a line or shot breaks causality, restore it or express it in a shorter alternative.

For each important image, label the following separately:

A single shot may contain F, P, and A, but label each component and its source separately. Mark inferred relationships as “To be verified”; do not present them as F.

Pay particular attention to seemingly ordinary added actions: giving someone an item early, locking a bridge, killing an enemy, and similar events also change the facts or sequence of the story, so classify them as A. Do not check only for newly introduced superpowers. Recheck every completed action in the shot list against the event card; the absence of supernatural adaptation does not mean A=0.

Translate adjectives such as “aura of strength” or “utterly shocking” into specific, visible changes. When a change introduces time stop, flight, or another new capability, classify it as A rather than P. Use domain or dharma-form (法相) concepts only when the source supports them or the user explicitly requests an adaptation.

Potential reference focuses include transitions from normal to abnormal, small motion to large-scale action, grayscale to color, solid form to spectral image, and action-to-environment feedback. Make a fresh selection based on the novel; do not copy entire characters or events from the reference, and do not attribute reference content to the novel’s source text.

First draft the shots, then jointly refine them according to the user-specified tool. For the default target of approximately 30 seconds, consider 6–8 shots initially, but adjust for dialogue length, transitions, and action complexity; exact timing is not a generation guarantee.

For each shot, list the suggested duration, main action and visual change, new information it provides, F/P/A classification and source, sound suggestion, and connection to adjacent shots. Planned durations should sum to the target. Do not ask viewers to perform complex visual interpretation while also carrying information solely through voiceover narration.

Use key states to confirm continuity: character features, clothing, props, which hand holds the weapon, spatial positions, attack direction, and ability form. In dharma-form cases, explicitly define the relationship between the physical body and the dharma form, synchronized movements, and weapon ownership; do not imply that they are one entity solely through matching color effects.

Preserve the source order for “when an item is obtained” and “when a result occurs.” If the order is changed, mark it as A and describe the impact.

When the material leaves the two sides of a bridge, the character’s retreat route, the role of an institution, or similar spatial relationships undetermined, label them “Spatial relationship to be verified,” ask the minimum necessary follow-up question, or clearly mark a provisional P/A plan; do not claim that causality has been fully verified. If the source does not specify which hand holds a knife, the time of day or weather, or similar details, these can be fixed as P, but do not list them among source-immutable elements.

Depending on the task scale, provide the following, merging or simplifying as needed:

Before delivery, check whether sources or reading scope have been fabricated; whether motives, targets, abilities, and results are coherent; whether P/A elements have been confused with F; whether the reference’s abilities have been forcibly imposed; whether the duration and information load are reasonable; whether any shot can be removed without losing information; and whether the ending leaves a meaningful question or a complete result.

Finally, reverse-extract the completed events from the actual shot list and compare them one by one with the source event order, while checking the performer and target of every action. A one-sentence causal summary cannot merge “A rescues someone” and “B hands over a prop” into “A completed both actions.” Correct the shot or clearly mark it as A for unsupported actions such as someone receiving an item early, suddenly leaping over an obstacle, or abruptly becoming exhausted. Merely declaring “faithful to the source” earlier cannot replace this step. Spatial assumptions that have not yet been verified should remain qualified in the production brief and must not later become established facts.

To understand a full-text adaptation, read Original teaching example; to understand how scene selection affects actual production, read Jian Lai case study; to recognize prematurely completed events or material unsupported by the source, read Failure and boundary examples. The examples illustrate judgment methods and cannot serve as factual sources for other novels.