recreate-video
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
./references/common-rules.md.
Workflow
- Analyze reference video
- Reuse the
analyze-videoworkflow. - Gather enough reference context for the model to understand what makes the source video work.
- Write source artifacts for the model
outputs/recreate_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/recreate_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-generate-videousing 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-generate-video, 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 recreate plan is longer than a model's maximum duration, the model should explain the tradeoff and suggest a segmented plan before calling
creatok-generate-video.
Artifacts
Write under recreate-video/.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.