Source: raw/How_to_Automate_Content_Creation_With_an_AI_Design_Agent_Full_Tutorial.md (https://www.youtube.com/watch?v=BXG4h1KLUTs), YouTube auto-captions fetched 2026-10-30.
Creator: AI Master (youtube.com/@iamAImaster)
Platform: YouTube
Published: not recorded in the raw file
Sponsored product demo, beta feature
“This video is sponsored by them.” Everything runs inside the sponsor’s product, and the agent is labelled beta (“some button names and behavior may change”). Name note: the captions spell the product “Love Art 2.0” and the URL “loveart.ai”; spoken, both match Lovart (lovart.ai), which this article uses. The agent’s name appears as “Arty”, “Artie” and “ARD”; this article uses Arty. Neither spelling was checked against the product.
Arty is a personal design agent inside Lovart 2.0 that keeps a memory of how you work, holds a brand kit, does its own design research, checks the canvas for problems, turns reviewer comments into a revision plan, and runs scheduled tasks. The creator teaches it his channel’s style, has it build a multi-format brand and social kit, then schedules a daily task that finds the top AI news stories and produces branded X and Instagram posts each morning. He treats each morning’s output as a draft that still needs a fact check.
Key Takeaways
- Two ways to teach the agent. A short onboarding conversation (name, work type, four multiple-choice questions on industry, content, audience and when you want several style directions, plus a free-text working rule), and “learn from past projects,” which reads your existing project library in about a minute and writes notes such as “prefers airy layouts” into memory. Auto-update memory keeps learning after that.
- Set the brand once. Assets → Brand kit → Create brand kit, from an uploaded file or a URL. From his website it pulled the logo, hex colours, font and a reference image. A design-guidance field holds the rules (his: clean high-contrast layouts, bold sans-serif headlines, one accent colour, white space, no stock-photo clichés, no emojis).
- Design research asks clarifying questions. Asked for visual-identity research on AI news channels, it asked three questions itself, searched the web, returned a reference grid to pick from, and wrote a report with two directions (“dark and glowing” and “light and honest”), each covering colour, logo, typography, shape, layout and a do/don’t list.
- Pick references that resemble each other. “Pick eight or nine references that look like each other because a coherent set gives the agent a clear direction and a random mix gives it mush.”
- One prompt, whole kit. With the brand kit and research report attached via
@, one request produced an avatar, banner, 16:9 thumbnail templates, post templates at 1:1, 4:5 and 9:16, and a poster, in one palette and font. A/menu offers prebuilt skills; skill names can differ per account. - Scheduled “delegated tasks.” A daily 9:00 task with its own instructions, a deliver-to target (a new project per run) and a per-run credit cap. Test-run it once before enabling.
- Comments become a plan you approve. Share with comments on; Arty summarises reviewers’ notes into one line per change with the asset each touches, and applies it only after you confirm.
- Self-evolving skills are not visible on day one. After you repeat a workflow a few times, it offers to save it as a skill. At “level one” the section was empty.
The Daily News-Post Task
His instructions, as read in the video (paraphrased only where the captions garble): every morning, find the two or three biggest AI news stories of the last 24 hours; for each, make one X post (short text plus a square visual) and one Instagram post (4:5 visual plus a short caption); keep logo, colours and fonts intact; use only brand assets and palette; keep copy dry and short; no emojis and no invented claims; no third-party logos or screenshots.
- That last line is deliberate: “News visuals are exactly where an AI is tempted to drop in someone else’s logo or a quote nobody said.”
- He left out font sizes, layouts and colour placement because the brand kit and memory cover them; “the instructions can stay about the content.”
- Settings: runs daily at 9:00; deliver to a new project; a modest credit cap. “Start low and see what one run actually costs. Then raise it from there.”
- Templates also exist (“daily visual inspiration radar”, “daily marketing moment prep”).
Other Agent Features
- Canvas checkup (runs once a day): flags off-brand drift or missing elements; click resolve to fix.
- Edit text changes words on a layer directly; touch edit lets you mark a spot and describe the change.
Try It
- Write your own working rule in the onboarding step (his: “Give me three different directions first, then refine the one I pick”).
- Build the brand kit from your live website, then verify every pulled hex code against the site.
- For any scheduled content agent, put the risky exclusions in the instructions: no third-party logos, no screenshots, no invented claims or quotes.
- Run one test before scheduling, set a low credit cap, and check the cost of a single run.
- Treat every scheduled news batch as a draft: check the facts in each headline before posting. “Checking a finished post is a much smaller job than building one from a blank canvas.”
Implementation
Tool/Service: Lovart 2.0 with the Arty personal design agent (beta). Setup: Left menu → the beta agent section; complete “tell me about yourself” and “learn from past projects”; create a brand kit under Assets. Cost: Not stated. Each scheduled run spends credits; a per-task credit limit is available. Integration notes:
- Attach assets and reports with
@; start from skills with/. - Delegated tasks deliver to a new project per run.
- Sharing with comments enables the feedback-revision flow.
Open Questions
- Pricing, credit cost per run, and plan limits are not given.
- The product and agent names are inferred from captions (see the warning callout).
- How the research step chooses and licenses web references is not covered.
- How well the news task avoids factual errors over many days is untested; the video shows one test run.
Related
- Building a Creator Media Kit With an AI Infographic Generator — the same channel’s brand-kit-plus-editor workflow for documents.
- Claude Design Marketing Skills (Grace Leung) — a Claude-native route to branded social assets.
- Cowork Content System (Grace Leung) — scheduled content production in Claude Cowork.
- Postiz — open-source scheduling for the posts once approved.
- Google AI Content Engine (Grace Leung) — a Google-stack content engine for comparison.
- Crafft AI — another hosted creative studio with brand assets and agent skills.