Claude Design for Beginners: The Complete Guide
A screen-recorded walkthrough of Anthropic's Claude Design tool, built around one argument: skip the design system and every output stays generic.
July 15thA start-to-finish walkthrough of building an Awwwards-level portfolio site with Claude Design 2.0's new canvas-editing tools, then handing it off to Claude Code and shipping it live on Vercel.
Claude Design 2.0 replaces the old prompt-and-wait loop with direct canvas editing, so making a portfolio site look intentional is now about feeding it strong visual references and a precise prompt, then hand-tuning layout and effects live instead of re-generating from scratch.
The core lesson is that AI design output is only as good as its direction: pulling three to five annotated references from sites like Mobbin, Awwwards, and Savee before prompting is what keeps a Claude Design output from looking like a generic PowerPoint slide. The new update's headline feature is direct canvas editing — moving, deleting, and repositioning elements instantly instead of leaving a comment and waiting for a full regeneration — plus finer layout and alignment controls for individual elements. The demonstrated pipeline runs reference gathering, a role-and-clarifying-questions prompt, iterative canvas edits (removing elements, centering text, adding a cursor-trail effect, rebuilding the work section into an asymmetric grid from a pasted mockup, and layering in a Three.js liquid-scroll effect), then a one-click handoff to Claude Code, a GitHub commit, and a Vercel deploy to get a public live link.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →
Creator claims the update makes Claude Design 90% better and previews the finished site with its image cursor-trail and liquid-scroll effects.

Shows the target portfolio site's hero cursor-trail effect and Three.js liquid scrolling before explaining the process.

Argues against prompting cold; pulls 3-5 screenshots from Mobbin, Awwwards, and Savee, drops them into Figma with notes on what's liked about each.

Shares the exact prompt: assigns Claude a 'principal product designer and award-winning creative developer' role and instructs it to ask clarifying questions until 99.9% confident before generating.

Claude Design asks clarifying questions, then returns a first-pass portfolio concept.

Uses the new direct canvas edit mode to delete/move elements instantly, center text via layout controls, add a mouse-trail effect to the hero via a text prompt, rebuild the work section into a custom asymmetric grid by pasting a mocked-up reference image, and layer in a Three.js liquid-scroll effect.

Pauses to mention the sponsor-style plug for a Framer portfolio template (Lofi) as a simpler alternative to the full custom build.

Uses the share button to send the design to Claude Code web, which rebuilds the design in code; meanwhile creates a new private GitHub repo to receive it.

Tells Claude Code to commit the rebuilt project to the newly created GitHub repo by pasting the repo URL into the chat.

Imports the GitHub repo into Vercel and deploys; the site goes live with a public preview link.

Frames the real takeaway as Claude Design and Claude Code merging into one workflow from idea to live site, then teases a follow-up video testing Claude's newest model on one-shot design tests.
An AI design tool only produces a distinctive result when it's given concrete visual direction up front and then hand-tuned through fast, targeted edits rather than repeated full regenerations.
“Claw design just got a major update and there's new things that make it 90% better than before.”
“Don't just start prompting and hope it looks good. Spoiler, it won't.”
“Now Claude isn't guessing in the dark.”
“Claude design and Claude code stopped being two separate tools and turned into one workflow. You can go from idea to live site without ever leaving that loop.”
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
The pitch is a workflow upgrade, not a new trick: Claude Design 2.0 adds direct canvas editing and a two-way handoff into Claude Code, and this video builds one portfolio site start to finish to show exactly how the new pieces fit together.
Gathering and annotating real shipped-design references before prompting so the AI has concrete direction instead of guessing.
Ending a design brief with an explicit instruction for the AI to keep asking clarifying questions until it reaches near-total confidence, rather than letting it assume intent.
“That's exactly why I built LoFi... If you wanna check it out, the link's in the description below.”
Soft mid-video plug for the creator's own Framer template as a simpler alternative, framed as an aside rather than a hard sell, with the affiliate/commission links placed in the description.
00:00
00:07
00:10
00:17
00:23
00:26
00:30
00:39
00:41
00:45
00:51
00:55
01:02
01:07
01:12
01:17
01:22
01:28
01:32
01:37
01:43
01:47
01:52
01:57
02:02
02:07
02:12
02:17
02:22
02:27
02:30
02:36
02:42
02:47
02:52
02:57
03:02
03:07
03:12
03:17
03:22
03:27
03:32
03:37
03:44
03:47
03:52
03:58
04:03
04:06
04:13
04:17
04:21
04:29
04:32
04:37
04:42
04:47
04:52
04:57
05:02
05:07
05:12
05:17
05:22
05:27
05:32
05:37
05:42
05:47
05:52
05:57
06:02
06:08
06:13
06:17
06:23
06:28
06:32
06:35A screen-recorded walkthrough of Anthropic's Claude Design tool, built around one argument: skip the design system and every output stays generic.
July 15thA folder system, not another prompt list — how to make Claude or ChatGPT write in your voice by giving it a home base of context instead of a one-off prompt.
July 13thA walkthrough of wiring Claude Opus 5 to live SEO data through DataForSEO to rebuild Semrush and Ahrefs' core reports for the price of API credits.
July 27thA screen-recorded walkthrough of a Claude Code skill that writes, designs, and quality-checks an entire Instagram carousel — hook slide included — before a human touches a single layer.
July 25thA blind, three-build test of Anthropic's newest coding models — and the first time this reviewer picked a model other than Fable as the winner.
July 25thA live, blind seven-model benchmark pits personality against performance — and crowns the model its own host can't stand talking to.
July 24th