AI Tool YouTube Review Workflow · публичный навык Open Skill
AI Tool YouTube Review Workflow
AI Tool YouTube Review Workflow — публичный Open Skill на Digen. Откройте, чтобы увидеть, что создаёт этот процесс.
Other
2 2 3
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4144395/2026-09-24/6cn8yt6b.png
# AI Tool YouTube Review Workflow Produce a polished, sectioned YouTube review video for an AI software/tool. Defaults: **landscape 16:9**, **~3–5 minutes**, **clear technology-review tone**, and **real screen recordings as the primary demonstration footage**. Never invent or claim features that cannot be verified. ## When to use - The user wants a YouTube review/demo video for an AI tool (from a one-liner, a product name, a script, or uploaded screen recordings / screenshots). - The user wants to revise one stage of an existing review video (rewrite the script, regenerate narration, restyle a shot, re-mix the cut). ## Reference files - `references/video-structure.md`: The 7-section structure (hook, intro, features, demo, narration, sections/transitions, summary+CTA) and the narration/visual conventions. Read it before writing the script. ## Steps 1. **Clarify** the AI tool, the input type (product name / idea / script / uploaded screen recordings or screenshots), and any specific features or angle the user wants covered. Default format is fixed: `orientation='landscape'` (16:9), target length **3–5 minutes**, **technology-review tone**. 2. **Research** the tool with `search_web` (features, pricing, use cases, limitations) so feature explanations are accurate. If the user provided screen recordings, **analyze them before writing the script**: use `recognize_video(video_url=<asset URL>, prompt=<describe the UI flow, steps shown, and key features visible>)` to understand the visuals, and `transcribe_audio(audio_url=<asset URL>)` for any existing narration — never guess what the footage shows. 3. **Write the script** as a document with `create_document(content=<full script>, name="<tool> review script", entity_type="scripts")`, following `references/video-structure.md`. Structure it into the 7 sections, each with a narration line and a visual instruction. Confirm the script with the user before generating. 4. **Generate narration** per section with `generate_tts(model='tts.design', voice_instruction=<clear, confident, engaging technology-reviewer voice>, text=<section narration>, name="<tool> VO <n>", entity_type="voiceovers")`. Keep each line concise (one idea per line). 5. **Generate visuals**: - Product / hero / thumbnail shots: `generate_image(model='t2i', prompt=<product visual>, orientation='landscape', name="<tool> hero", entity_type="storyboard_images")`. - Motion b-roll from a still (only when screen recordings are insufficient): `generate_video(model='i2v', input_urls=[<hero image>], prompt=<subtle camera motion>, orientation='landscape', name="<tool> b-roll", entity_type="storyboard_videos")`. - **Screen-recording demo**: use the user's uploaded footage directly (pass its asset URL into the mix) as the primary demonstration footage. If none exists, generate an illustrative UI shot with `generate_image` and animate it with `i2v`, and clearly tell the user this is illustrative, not a real recording. 6. **Generate background music** with `generate_music(prompt=<understated, tech-review background track>, name="<tool> music", entity_type="music")`. 7. **Assemble** with `mix_clips`: concatenate the clips in script order, lay the narration and music tracks over them (music at low volume, e.g. `volume: 0.3`). Pass `name="<tool> review video"`, `entity_type="final_videos"`, and `orientation='landscape'` matching the clips. 8. **Summarize** in natural language: the sections produced, the narration, and the final cut. Offer to revise any stage. ## Notes - Use `write_todos` to list the steps and update as you progress. - Keep narration concise and one idea per line; prioritize visual continuity (consistent product look, matching landscape orientation across all shots, consistent terminology). - **Accuracy rule**: do not invent or claim features that cannot be verified. If a feature is unverified, say so and suggest verifying it. Screen recordings are the primary demo source; generated visuals are supporting B-roll o