A companion-type boyfriend · সর্বজনীন Open Skill
A companion-type boyfriend
A companion-type boyfriend Digen-এ একটি সর্বজনীন Open Skill। খুলে দেখুন এই ওয়ার্কফ্লো কী তৈরি করে।
Manhua
3 4 4
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3947739/2026-08-24/ravezj29.png
# Female-Oriented · First-Person Boyfriend Daily Companion Video Turn one uploaded photo of the male lead into a daily companion slice-of-life long-take short video where "camera = the user's eyes": the male lead accompanies the user while working, shopping, traveling, sleeping, eating, exercising... The male lead does not speak, no dialogue, no dialogue box, no TTS; the final output is picture + BGM. Content is **clean and sweet** (wholesome and cozy, no adult content of any kind). The video is driven by **start/end frames**: first generate the start frame and end frame of the same scene, letting the male lead transition naturally from the "interaction starting point" to a "closer-to-camera interaction high point", which creates a stronger interactive feel; interactivity and sweetness are the core experience, and movements must be natural and logical. > **Mandatory reference file**: You MUST read `references/hard-requirements.md` before executing. All hard requirements in it (image-generation model uniqueness, character/body type/art style reproduction, on-screen character uniqueness, start/end-frame coherence, interaction intensity & sweetness, sleeping-scene special case, natural and logical movements, handing-over logic, user checkpoints, no dialogue/no open mouth, language consistency, content & compliance, honest reporting) are mandatory and take priority over anything else in this document; the body below only keeps the execution workflow and parameters and does not repeat the hard-requirement details. ## When to use - The user uploads one photo of a male lead and wants a "female-oriented first-person daily companion life" short video (working/shopping/traveling/sleeping/eating/exercising/staying home, etc.). - The user wants to redo only one stage: start/end frame stills, long-take video, BGM, or final mix. - The user only describes a scene (e.g., "keep me company while I watch dramas at home", "sleep with me") and uploads a photo, expecting a finished video directly. ## Steps ### 1. Collect input and confirm requirements Confirm with the user up front (ask everything at once; do not keep asking): - **Male lead photo**: the user must upload exactly 1 photo; it is the only appearance reference (see the reference file for character/body type/art style reproduction). - **Scene**: free-form description from the user, or choose from templates: ① Working (accompanying the user at the office/overtime) ② Shopping ③ Traveling ④ Sleeping/resting (lulling to sleep/sharing the bed; see the reference file for the interaction special case) ⑤ Eating (cooking together/dining/feeding) ⑥ Working out (spotting/handing water) ⑦ Staying home (watching dramas/gaming/cuddling) ⑧ Commuting. - **Aspect ratio**: default vertical 9:16 (portrait); for scenes like traveling the user may request horizontal 16:9 (landscape). - **Duration**: long take defaults to 10 seconds; the user may specify 10-12 seconds (final length auto-clamped by model limits). - **BGM style**: default gentle piano / Lo-fi warm tone; user can customize (e.g., upbeat, romantic, healing). - **Language**: communicate with the user throughout in the SAME language as the user's input, and make sure the language of all generated outputs (prompts, asset naming, BGM style descriptions, etc.) matches the user's input language; never mix in other languages (hard requirement, see the reference file). ### 2. Generate start/end frame stills (first-person view) Use `generate_image`, with the uploaded male lead photo as the only appearance reference, to generate two first-person **start frame** and **end frame** stills of the same scene (hard requirements such as image-generation model uniqueness, character/body type/art style reproduction, on-screen character uniqueness, start/end-frame coherence, and user checkpoints are in the reference file): - Use for both: `model='i2i.hd.lite'`, `orientation='portrait'` (or the user-requested landscape), `resolution='720P'`. - **Start frame**: