Comic Drama Generate

Use this skill to produce short comic-drama episodes through a deterministic, character-consistent pipeline.

Required prerequisite skills

Require these skills before using this skill:

• - INLINECODE0
• INLINECODE1
• INLINECODE2

If any are missing, install them first. Use this install pattern:

CODEBLOCK0

Before starting production, verify:

• - all three prerequisite skill folders exist
• required API credentials are configured for the prerequisite skills
• INLINECODE3 is available locally
• there is enough credit/quota for script, image, and video generation

Credential note:

• - comic-drama-generate orchestrates other skills and does not introduce its own separate API key.
• The prerequisite skills use their own environment-variable contract, typically NANOPHOTO_API_KEY.
• Prefer normal sub-skill execution that relies on the prerequisite skill's configured env.
• If direct script execution is necessary, first preserve the same env contract by ensuring NANOPHOTO_API_KEY is present in the shell.
• Use explicit --api-key only as a fallback when the shell cannot inherit the expected env.
• Do not treat direct reads from config files such as openclaw.json as the primary workflow; reserve them for debugging or recovery paths.

Mandatory production order

Always follow this exact order.

1. 1. Ask the user for style first

- Ask for visual style before generating anything. - Also confirm tone, audience, episode count, target platform, aspect ratio, and approximate episode duration. - Lock the series bible first: premise, character roster, style keywords, and production constraints.

1. 2. Use video-prompt-generator to generate the script foundation

- Generate the story premise, dramatic arc, episode beats, and scene-level writing foundation. - Make the episode structure suitable for short-form serialized漫剧. - Keep the pacing compatible with downstream 15-second shot generation.

1. 3. Use nano-banana-pro to generate character turnarounds

- Generate character three-view sheets for all core recurring characters. - Treat the three-view sheets as the canonical identity source for face, hair, silhouette, wardrobe, and palette. - Do not skip this step when character consistency matters.

1. 4. Use nano-banana-pro to generate keyframes

- Generate keyframes for each planned shot using the turnaround image URLs as reference inputs. - For multi-character shots, include all relevant turnaround URLs. - Use keyframes to lock composition, costume continuity, environment cues, and emotional staging.

1. 5. Use video-prompt-generator to write the storyboard / shot script

- Convert each episode into shot-level prompts and a shot list. - Write for single-shot generation with clean handoff into video prompts. - Leave a hook at the end of every episode.

1. 6. Use sora-2-generate to create 15-second single-shot videos

- Default to 15-second shots unless the user explicitly wants a different supported duration. - Prefer imageToVideo for character shots. - Pass keyframe URLs as image inputs. - Use textToVideo only for shots where character consistency is unimportant, such as atmosphere inserts or environment transitions.

1. 7. Use local ffmpeg to trim, arrange, and merge clips

- Download and save each finished shot locally. - Trim or normalize clips locally if needed. - Merge shots in deterministic order with ffmpeg. - Export final episode cuts under a dedicated project directory.

Non-negotiable working rules

• - Prefer this consistency chain at all times:

- character turnarounds → keyframes → image-to-video

• - Use public URLs, not local file paths, when passing assets between:

- character turnarounds → keyframes - keyframes → video generation

• - Do not pass local files into nano-banana-pro or sora-2-generate APIs for these handoff steps.
• Do not skip directly from text prompt to final video for main character shots unless the user explicitly accepts weaker consistency.
• Keep exact turnaround and keyframe URLs recorded in project docs because downstream generation depends on them.
• Rewrite storyboard timing before generation if model limits, quota, or rate limits make the original plan unrealistic.

Story and pacing guidance

• - Design shots around the active video model's supported duration.
• For this workflow, prefer a shot grammar that fits 15-second single-shot clips.
• If an episode needs more time, split it into more shots rather than overloading one prompt.
• Ensure each episode ends with a hook, reveal, reversal, question, or emotional cliffhanger.

Suggested project structure

CODEBLOCK1

Execution notes

• - Use video-prompt-generator twice when needed:

- once for premise / episode writing - once for shot-level prompt refinement

• - Use nano-banana-pro for both:

- character turnaround generation - keyframe generation

• - Use sora-2-generate mainly in imageToVideo mode for character-led shots.
• Use local ffmpeg for final editorial assembly, not remote editing tools.

Expected deliverables

Produce organized outputs that make the project easy to resume:

• - docs/series-bible.md — premise, style, characters, world rules
• INLINECODE26 — episode summaries and hook endings
• INLINECODE27 — shot-by-shot structure for each episode
• INLINECODE28 — turnaround URLs, keyframe URLs, and video URLs
• local shot files under INLINECODE29
• local edit outputs under INLINECODE30
• final episode exports under INLINECODE31

Read these references when needed

• - references/workflow.md — end-to-end production workflow
• INLINECODE33 — prerequisite skill verification and install steps
• INLINECODE34 — character turnaround, keyframe, and image-to-video consistency rules