Test log
Magic Patterns test data
Raw results for the Magic Patterns review. The verdict is in the review.
- Tested
- 5 November 2026
- Tester
- Miles Foster
- Account
- Free demo account, signed up by RankDude, nothing paid to the brand
- Tests
- 9
- Screenshots
- 8
- Method
- Published methodology
- Environment
- Google Chrome on desktop
Results
| ID | Test | Key measurement | Score |
|---|---|---|---|
| T01 | Brand Kit & Consistency | – | n/a |
| T02 | Sharing & Collaboration | – | n/a |
| T03 | Brief to First Draft | Brief to draft: 1 min 48 s | 5/5 |
| T04 | Brief Adherence | Required elements right: 59 of 60 | 2/5 |
| T05 | First Draft Quality | Design faults: 12 | 2/5 |
| T06 | Editing Freedom | Edits that worked: 10 of 12 | 4/5 |
| T07 | Export & Handoff | Export points met: 5 of 10 | 2/5 |
| T08 | Rights & Commercial Use | Owner of the output: You | 3/5 |
| T09 | Pricing & Value | Starter, paid yearly: $17 a month | 3/5 |
| Overall (in the review) | 3.6/5 | ||
Brand Kit & Consistency
n/aNot Tested- Method
- Upload logo, fonts and colours as a brand kit, then generate ten designs and check them.
- Remark
- Not graded; left out of the average.
Raw notes
Upload logo, fonts and colours as a brand kit, then generate ten designs and check them.
Not run: the three briefs cost 15, 28 and 19 credits, so ten more designs would need well over the 24 credits left, and the review stops at 85 of 100.
- Needs a paid plan or a second month of credits.
Sharing & Collaboration
n/aNot Tested- Method
- Eight points tried with a second account.
- Remark
- Not graded; left out of the average.
Raw notes
Eight points tried with a second account.
Not run: no second account was made.
- Needs a second test account.
Brief to First Draft
5/5Tested- Method
- Each brief was typed into a new design with the supplied files attached.
- Brief to draft
- 1 min 48 s
- Remark
- Scale 5: median under 5 minutes.
Raw notes
Each brief was typed into a new design with the supplied files attached. The clock ran from pressing Enter to the moment the Stop button disappeared and the assistant posted its last message, read from a script that watched the page and a screenshot every ten seconds. The browser was recorded for every run.
The landing page finished in 1 minute 32 seconds, the bakery set in 1 minute 48 seconds and the booking app in 2 minutes 43 seconds. The median is 108 seconds.
The first file was written 41 seconds in for the landing page, 51 seconds for the booking app and 73 seconds for the bakery set. The result is a live page built from editable components, not a picture.
- Times are read at five second steps, so each is accurate to about five seconds.
Brief Adherence
2/5Tested- Method
- Twenty required elements per brief were checked in the first draft before any editing.
- Required elements right
- 59 of 60
- Invented items
- 17
- Remark
- Scale 2: invented claims count against the draft, even when 58 or more elements are right.
Raw notes
Twenty required elements per brief were checked in the first draft before any editing. Text, fonts, colours, sizes and image sources were read from the rendered page's DOM, and the rest by eye at 2x.
The bakery set had 20 of 20: three pieces at 1080 by 1080, 1080 by 1920 and 820 by 312 pixels, every line of text word for word, Fraunces and Inter, the three colours, the logo top left on each and the supplied loaf and pastry photos.
The landing page had 20 of 20 at a 1440 pixel layout. The booking app had 19 of 20: its confirmation text read "We will email you a calendar invite within one hour at" plus the address I typed, not the exact sentence. All three used the supplied logos and photos. 59 of 60 in all.
Invented items: 17, counted by kind, so the three yarn photos are one item. Landing page, 8: a line under each yarn, "Add to basket" labels, a "New on the shelf" heading, a "See all yarn" link, a "Save a seat" button, a copyright line, "Find us" and "Open" headings in the footer, and three generated yarn photos. Booking app, 7: a welcome line, a claim that all sessions include an online gallery of edited photos, a description under each session (one says up to six people), "Continue" buttons, an intro line under the title on screens 3 and 4 with an "Available times" heading, a booking summary (a line on screen 4 and a details table on screen 5) and a Done button. Bakery set, 2: a heading and a note above the three pieces.
- The supplied photos are simple generated pictures, so only their use is checked, not how they look.
- The invented-item count groups items of one kind, such as the three yarn lines, as one item.
First Draft Quality
2/5Tested- Method
- The first drafts were opened at their required sizes and the landing page at 1440, 768 and 360 pixels wide.
- Design faults
- 12
- Remark
- Scale 2: 11 to 20 faults.
Raw notes
The first drafts were opened at their required sizes and the landing page at 1440, 768 and 360 pixels wide. Faults were counted: contrast calculated from the hex values (WCAG AA: 4.5, large text 3), overlaps, text cut off, tap targets under 44 pixels and uneven spacing.
12 faults. Bakery 4: brown text on the orange button, on the headline highlight and on the $18 highlight all measure 2.98 to 1, just under the 3 to 1 that large text needs, and in the story the "Order now" button covers the line "Pre-order by Friday 6 p.m.".
Booking app 3: the calendar day buttons are 40 pixels, the month arrows 36 pixels and one button 40 pixels, all under 44.
Landing page 5: the Shop, Classes and Visit links are 19 pixels tall and "See all yarn" is 23, at 360 pixels the logo text touches the Shop link, the "Merino DK" and "Cotton Aran" titles wrap to two lines, so the text under them sits lower than on the third card, and the copyright line is 3.58 to 1 on its background.
- The booking app is a phone-sized prototype, so it was not checked at other widths.
Editing Freedom
4/5Tested- Method
- Twelve fixed edits on the landing page, made with the controls a person uses: the visual editor, the chat and the version list.
- Edits that worked
- 10 of 12
- Remark
- Scale 4: 10 or 11.
Raw notes
Twelve fixed edits on the landing page, made with the controls a person uses: the visual editor, the chat and the version list. An edit counts when it did what was asked without support. Chat edits cost credits, and each was checked afterwards in the page's DOM.
Ten worked. The headline font changed everywhere from one chat line (8 seconds, 6 credits). One brand colour changed everywhere from one line, with no trace of the old value left (3 credits). A button dragged to a new place, a text box resized from 448 to 300 pixels, the hero photo swapped by address and its crop set from Cover to Contain, one word typed in place, a new Gift cards section added by chat (3 credits), a single yarn line rewritten from a written instruction with the other lines untouched (2 credits), the preview width changed from 1440 to 768 to 360 pixels with the layout re-flowing, and three steps undone by restoring an older version.
Two did not: there is no lock control, and aligning three elements with guides is not offered; selecting three titles with Shift kept only the last one. The drag moves an element within the page flow, not freely, and Ctrl+Z did not undo chat changes: the version list did.
- The canvas size edit was scored on the preview width control, which changes the view, not the design.
Export & Handoff
2/5Tested- Method
- Ten export points were tried on the finished bakery set.
- Export points met
- 5 of 10
- Remark
- Scale 2: 3 to 5 points.
Raw notes
Ten export points were tried on the finished bakery set. The Export menu lists Export to Figma, Sync with GitHub, Download Code Zip, Copy Code as Prompt and Handoff to Eng. I downloaded the zip, unpacked it in an empty folder, installed it and built it.
Five points were met. The zip is a Vite and React project that built into plain static files in under a minute, so it runs on any web host. Fonts are named: the CSS loads Fraunces and Inter from Google Fonts. The whole set of three pieces comes in one zip. The three supplied files (logo and two photos) are in its public folder. The zip and the editor preview carry no watermark.
Five were not: there is no PNG, no SVG and no print PDF, no raster export at full resolution, and the Figma export, which copies the design and asks you to paste it into a Figma file, was not tried because no Figma account was used.
- The shared preview was not opened, so a watermark there is not ruled out.
Rights & Commercial Use
3/5Tested- Method
- I read the Terms of Service and the Privacy page, reached from the site footer, and looked for a training setting in the account.
- Owner of the output
- You
- Training on the free plan
- Yes
- Indemnity for output
- None
- Remark
- Scale 3: opt-out only by paying.
Raw notes
I read the Terms of Service and the Privacy page, reached from the site footer, and looked for a training setting in the account.
Ownership is clear: "all content, including but not limited to designs, code, and user interfaces, generated through the use of the Services ... shall be the sole and exclusive property of the user who created it."
On the free plan the terms let Magic Patterns "use user provided inputs and content and User Generated Content to train and improve the Services", show your designs in public galleries, and add a note such as "Built with Magic Patterns" to public designs. Paid plans are promised no training. I found no setting to opt out on free; moving to a paid plan is the way. The indemnity in the terms runs from the user to Magic Patterns, and there is none from the vendor.
- The terms do not say whether commercial use differs by plan.
Pricing & Value
3/5Tested- Method
- Plans were read in the product's plan chooser, in both billing views, and on the public pricing page.
- Starter, paid yearly
- $17 a month
- Starter, monthly
- $20
- Remark
- Scale 3: $200 to $299 a year.
Raw notes
Plans were read in the product's plan chooser, in both billing views, and on the public pricing page.
The cheapest plan that removes the preview watermark and promises no training is Starter: $17 a month per seat paid yearly, or $20 paid monthly. Yearly that is in the $200 to $299 band. Business is $85 a month paid yearly, or $100 monthly.
Free gives 100 credits a month. The three briefs used 62 of them, so they fit the free allowance once. Starter gives 1,000 credits a month, with rollovers and on-demand usage.
- No plan was bought; renewal prices were not seen.
Other features checked
Coverage
Tested
- The three fixed briefs typed into the design prompt, with the supplied logo and photos attached
- Twenty required elements per brief read from the rendered page, with a contrast calculation
- Twelve fixed edits with the visual editor, the chat and the version list
- The code zip, installed and built
- Terms of Service, Privacy page and the plan chooser read
Not tested or limited
- Brand kit was not graded: ten new designs would not fit the credits left after the briefs and the edits, and the review stops at 85 of the 100 free credits
- Collaboration was not graded: every point needs a second account
- Export to Figma was not tried end to end: it copies the design and asks for a paste into a Figma file, and no Figma account was used
- The booking app is one clickable prototype, so its five screens were read by clicking through it, not from five separate frames
- Edit 12, lock an element, and edit 8, align three elements with guides, have no control in the editor and count as not done
- The shared preview link was not opened, so a watermark on shared designs is not ruled out; the terms say free plans may carry a Built with Magic Patterns note on public designs
Still to test
- Brand kit and ten designs on a paid month
- Collaboration with a second account
- The Figma paste
- The shared preview for a watermark