Mr.K Cover Studio

A fixed visual cover system for Douyin creators who publish AI, business, judgment, and field-note content.

中文 | English | Live Site | GitHub

Mr.K Cover Studio turns a Douyin link, copied share text, screenshot, or content idea into an on-brand cover. It is not a general poster editor. It is a repeatable content asset system for one account style: fixed identity, fixed categories, flexible titles and backgrounds.

Batch Agent

Paste one Douyin link or share snippet per line, then run the Batch Agent. The studio processes the queue one by one, creates a content-matched background, renders the fixed visual system, saves each result to the material library, and lets you load or download each generated platform pack from the same page.

The default batch limit is 12 works per run so browser export and model calls stay predictable.

Two Ways To Use It

Delivery	Best For	Command
Agent Skill	Send a Douyin link to an Agent, answer a few questions, generate a cover brief and asset	scripts/install_agent_skill.sh
Local Web Studio	Preview, tune, save to material library, export PNG/M4V locally	scripts/run_local_studio.sh

Conversational Agent Workflow

The Agent does not generate first. It asks for context, then turns the answer into a production brief:

Agent integration guide: download the Skill or copy the prompt.

1. Ask for Douyin ID, profile link, or profile screenshot. For Mr.K, default to KevPH2026.
2. Ask for the task: single-work cover, motion intro/outro, DNA decode, batch cover refresh, collection cover, or profile background.
3. Choose the category: prefer AI下半场, 强者恒强, or 在路上; use a custom category only when needed.
4. Collect material: one link/caption/screenshot for a single work; image/video material for a motion intro/outro when available; for DNA decode, one public Douyin work link is enough to start because the crawler will try to discover the creator profile, avatar, ID, and more public works.
5. For a new creator or client account, crawl account_identity and at least 10 public samples before generating an applicable style_profile.
6. Confirm the brief: Chinese title, English subtitle, background direction, ratio/duration, and code. Then generate and save to the material library.

Starter prompt:

My Douyin ID is KevPH2026. I want a single-work cover. The content link is ...

Why It Exists

Most creator homepages fail quietly. The content may be sharp, but the grid looks accidental: different crops, different fonts, different visual logic, no memory hook.

This project fixes that problem for MR.K 在路上.

• One visual language across single posts, 1-2 second motion intros/outros, collections, and profile background
• AI-assisted title, summary, category, and background direction
• Title-driven generation: enter one title, auto-classify, translate, code, subtitle, generate Image2 background, then edit text live
• Batch Agent queue: paste multiple Douyin links, process them sequentially, save each generated cover, and reload any result into the current preview
• DNA decode from at least 10 content samples before proposing a design direction
• One-link account discovery: from a single public Douyin work link, try to pull creator profile, avatar, ID, and related public works
• Account Style generation that can be applied to category, accent color, image style, density, and background prompt controls
• First-run onboarding for choosing Agent/Web workflow, starter template, and content category
• Quick templates for AI news, strong quotes, field notes, motion intros/outros, collections, profile banners, and custom columns
• Custom categories with editable Chinese name, English name, subtitle, and accent color
• Local-only configuration for OpenAI-compatible LLM and image models
• Anonymous users can try 3 AI generation actions before email-code login is required
• Email verification login with auto-create account on first verified email
• In-product feedback entry for bugs, confusing flows, output issues, and feature requests; KV-backed in production with a GitHub Issue draft fallback
• Realistic editorial backgrounds instead of generic AI wallpaper
• Local material management for generated covers and motion frames
• A deployable Skill for Agent workflows

Brand DNA

• Account: MR.K 在路上
• Douyin ID: KevPH2026
• WeChat ID: kevph2026
• Domain: dna.superk.ai
• Categories: AI下半场, 强者恒强, 在路上

Core Workflow

1. Paste a Douyin share snippet, short link, screenshot, or topic. For bulk work, paste one link or share snippet per line into Batch Agent.
2. The configured LLM extracts the content angle and proposes title/category/summary.
3. The configured image model creates a background that matches the copy.
4. The fixed Mr.K overlay system renders the final cover.
5. Save to material library or export PNG/M4V.

DNA Decode

For a new creator or account, paste one public Douyin profile/work link. The crawler first tries to discover the author profile, avatar, ID, and related public works. If Douyin blocks the public page, paste titles, cover text, captions, or screenshot OCR as fallback. The Agent then extracts:

• crawled account identity: nickname, Douyin/profile ID, avatar URL, and profile URL when available
• account positioning, audience, voice, and content pillars
• style signals and visual taboos
• an applicable style_profile: category, column labels, accent color, image style, density, background direction, and cover rules
• 2-3 design directions for covers, collections, and profile background
• operating rules for future cover generation

If fewer than 10 usable samples are found, the Agent returns a sample-gap request instead of inventing a DNA.

Outputs

• Single work cover: 1080x1920
• Motion intro/outro overlay: 1080x1920, 1.2-2s, M4V-first export from the browser
• Collection cover: 1080x1080
• Profile background: 1920x640
• Material library: local by default, KV-backed in production

Local Development

python3 cover_studio_server.py

Open:

http://127.0.0.1:8765/mrk-cover-studio.html

With image generation:

OPENAI_API_KEY=sk-... python3 cover_studio_server.py

Local Model Configuration

The Web Studio includes a local-only model panel for analysis, copywriting, and background generation.

• LLM: OpenAI-compatible Chat Completions endpoint, model name, optional API key
• Image: OpenAI-compatible Images endpoint, model name, optional API key
• Storage: browser localStorage only, under mrk.localModelConfig
• Server behavior: the config is used only for the current API request and is not persisted

Leave these fields empty to use environment variables such as DEEPSEEK_API_KEY and OPENAI_API_KEY.

Auth And Trial Limit

Logged-out visitors can use 3 AI production actions: Agent cover/motion generation, DNA decode, or manual AI background generation. After the third action, the page opens the account menu and requires email-code login.

Email-code login supports:

• Resend: RESEND_API_KEY, optional AUTH_EMAIL_FROM, AUTH_EMAIL_REPLY_TO
• SMTP: SMTP_HOST, SMTP_PORT, SMTP_USER, SMTP_PASSWORD, SMTP_FROM
• Local/dev fallback: when KV or email delivery is not configured, the browser uses a local debug code flow

Production

Vercel serves the static studio and Python serverless APIs:

/api/cover-agent
/api/dna-agent
/api/generate-image
/api/transcode-m4v
/api/auth
/api/assets
/api/feedback

Required environment variables:

DEEPSEEK_API_KEY
OPENAI_API_KEY
KV_REST_API_URL
KV_REST_API_TOKEN

Optional:

DEEPSEEK_MODEL=deepseek-chat
OPENAI_IMAGE_MODEL=gpt-image-2
OPENAI_IMAGE_QUALITY=medium
OPENAI_IMAGE_INPUT_FIDELITY=high
ACCOUNT_KV_PREFIX=mrkcover
RESEND_API_KEY=re_...
AUTH_EMAIL_FROM="Mr.K Cover Studio <login@your-domain.com>"
AUTH_EMAIL_REPLY_TO=kevph2026@example.com
SMTP_HOST=smtp.example.com
SMTP_PORT=587
SMTP_USER=...
SMTP_PASSWORD=...
SMTP_FROM=login@your-domain.com

When KV is not configured, the browser falls back to local-only login and local asset storage.