Daily Vlog Maker · публичный навык Open Skill
Daily Vlog Maker
Daily Vlog Maker — публичный Open Skill на Digen. Откройте, чтобы увидеть, что создаёт этот процесс.
ShortDrama
29 7 21
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3949831/2026-09-16/xawbcuvy.png
--- name: daily-vlog-maker display_name: Daily Vlog Maker description: 'Creates selfie/POV lifestyle vlogs: a consistent character in real-world environments, filmed as a series of short first-person clips and stitched into one video. Use when the user wants a daily vlog, day-in-the-life, slice-of-life, or lifestyle short; when they want a character filming themselves through everyday scenes (morning, commute, work, night, weekend, outing, travel, hangout, etc.); or when they want to customize, add, reorder, or extend the plot segments of such a vlog.' allowed-tools: - generate_image - generate_video - mix_clips - recognize_image - create_document --- # Daily Vlog Maker ## When to use - The user wants a daily-life vlog shot from a selfie / first-person (POV) camera angle. - The user wants a series of 10-second clips (default 4) covering a day — morning routine → commute → leaving work → night wind-down — stitched into one final vlog. - The user wants to customize the plot segments, add new segments, or continue/extend an existing daily vlog. - The user wants an anime/2D character living in real-world environments (breaking the 4th wall). ## Core rules (do not break) 1. **Image generation must always use `model='i2i.hd.lite'`** (image-to-image). Because it is image-to-image, a **character reference image is required** — if the user hasn't uploaded one, ask them to upload a character image (anime/2D recommended; a photo of themselves also works) before generating anything. **The character's art style must strictly match the user's reference image** — same art style, appearance, facial features, hairstyle/color, skin tone, clothing, and proportions in every keyframe and every clip. Never restyle, photorealize, redesign, or reinterpret the character. **Full-body expansion**: if the user's uploaded image is a close-up/portrait, first build a **full-body character reference** with `i2i.hd.lite` using the **user's original upload as the base image** (`input_urls=[user_uploaded_original]`) — expand the close-up to a full-body shot while **strictly locking facial features, hairstyle, skin tone, and art style to the user's uploaded image**; if the clothing is not fully visible, complete it with a **more beautiful and aesthetically pleasing outfit**. Image prompts must be wrapped with the action templates in `references/image-edit-prompts.md` (`extend` for full-body, `blend` for keyframes). Scene/physical details: `references/prompt-templates.md`. 2. **Video generation must always use `model='ref2v'`** with `gen_mode='reference_to_video'` and at least 1 reference image (the keyframe). Never use any other video channel. 3. **Default 4 segments, each exactly 10 seconds** — see `references/segments.md` for the default beats and how to customize (the user may change, remove, or add segments; count and content follow the user's request). 4. **Real-world environments by default**: all scenes are realistic real-world locations unless the user explicitly requests a specific scene. The character's art style is independent of the environment — an anime/2D character can appear in a photorealistic environment (breaking the 4th wall). The environment must never override the character: the character always keeps the exact art style of the user's reference image, no matter how photorealistic the environment is. 5. **Selfie / POV shooting**: every clip is shot as if the character is filming themselves (selfie) or from the character's point of view. The shooting angle (high / mid / low) is intelligently matched to the plot — see `references/angle-table.md`. 6. **Orientation**: portrait (9:16) by default (short-video platforms). Use landscape only if the user asks. 7. **Final output**: stitch all clips into one final vlog with `mix_clips`. 8. **Voiceover / self-narration (must use the video model's built-in voice)**: in every segment the protagonist speaks in the **user's language**, telling the viewer what they are doing right now. The exact line must