video-remix

Constraints

• Platform: TikTok only.

• Must NOT do 1:1 copying.

• Must apply:

• structure rewrite

• expression rewrite

• style differentiation

• The model's final user-facing response should match the user's input language, default English.

• Avoid technical wording in the user-facing reply unless the user explicitly needs details for debugging or to share with a developer.

• Follow shared guidance in ../shared/references/common-rules.md .

Workflow

• Analyze reference video

• Reuse the video-analyze workflow.

• Gather enough reference context for the model to understand what makes the source video work.

• Write source artifacts for the model

• outputs/remix_source.json

• Include:

• reference TikTok URL

• analyze result payload

• analyze artifacts directory

• optional user constraints such as angle / brand / style

• Model output happens in the conversation

• The model should read outputs/remix_source.json

• The model should help the user choose a direction without over-constraining the process.

• Unless the user explicitly asks for a live-action shoot version, the model should default to creating a script, storyboard, and visual direction that are intended for AI video generation rather than human filming.

• Typical directions include:

• stay closer to the original concept and execution

• create a differentiated remix version

• use the reference only as inspiration for a new version

• The model should present these directions in simple creator / seller language rather than technical or production language.

• The model should decide, with minimal friction:

• what stays at the idea level

• what changes in structure / wording / visuals

• copyright / similarity risks

• the level of detail that is most helpful next: concept, outline, short script, storyboard, or shotlist

• The model should ask only for high-impact creative preferences when needed, not force a fixed template.

• The model should usually show a useful first draft quickly instead of starting with many questions.

• The first draft should default to an AI-generation-ready version.

• The model should prefer a first draft wording that naturally sets up the next handoff, such as "If this direction looks good, I can generate the video next."

• If the user wants to recreate or adapt a selling video, the model should first collect the user's own product context before writing a fitted script.

• Start with only the most important product details:

• product name

• core selling points

• product images or reference materials if available

• price / offer / promotion details if relevant

• If more context is needed, the model should ask short follow-up questions one by one instead of requiring a long upfront brief.

• The model should avoid making the user restate information that was already clear from the previous analysis or conversation.

• If the user wants final generation

• Once the creative direction is clear enough, the model should hand off to creatok:video-generate using the script or brief already developed in the conversation.

• The model should avoid asking the user to rewrite their request from scratch before generation.

• The default handoff should be to AI generation, not a human shoot plan.

• The model should phrase this in natural creator language that invites creatok:video-generate , for example:

• "If you want, I can generate this version now."

• "If this script looks right, I can turn it into a video next."

• "I can go ahead and make the video from this version."

• Before handing off, the model should already reason about generation feasibility:

• whether the plan fits within a single segment

• whether it needs to be split into multiple segments

• whether a recurring human character means the user needs to upload a portrait / person reference

• whether the selected generation path requires a model that supports real-person reference images

• If the remix plan is longer than a model's maximum duration, the model should explain the tradeoff and suggest a segmented plan before calling creatok:video-generate .

Artifacts

Write under video-remix/.artifacts/<run_id>/... .

Notes

• This skill should feel like a creative bridge between analysis and generation.

• Prefer smooth continuation from the analyzed reference rather than making the user restate the whole idea.

• For selling-video recreation, adapt the reference to the user's own product instead of drafting a generic copy first.

• After producing an AI-generation-ready version, the model may optionally ask whether the user also wants a live-action shoot version.

• Keep the interaction lightweight and practical for non-technical creator / seller users.