How the reference library works, and where it goes next
A plain guide to everything behind the library: what we built in the pilot, where each file lives, what Vercel does, why originals and compression matter, how a cloud version would work, and what it all costs.
What this is: a private mymind for images and video
Yes, it is heading towards mymind, narrowed to visuals, organised around Auspicious's eight spaces, and linked back to Figma.
mymind is a private place to save anything you see: notes, links, articles, images. It sorts things for you with AI, so you never file or tag by hand, and you search by colour, brand or word. Its makers say it suits people with ADHD because there are no folders, rules or set-up to fail at.
Our library shares that idea, and it is narrower and more specific:
| mymind | Reference library | |
|---|---|---|
| What goes in | Notes, links, articles, images, video (on Mastermind) | Images and video only |
| How it is organised | AI tags, general topics | A fixed brand vocabulary: the eight spaces, type, light, material, mood, colour |
| Where things come from | Anything you save from the web or phone | Your Figma boards; later, anything you add |
| Link back | To the original web page | To the exact layer in Figma |
| Who sees it | Only you | You and your team, through the portal |
| Who owns the data | Stored by mymind | Stored by you: your folder, your cloud account |
| Price | $79 to $129 a year | See section 09 |
Questions
Should I just use mymind instead?
For collecting from the web day to day, mymind is excellent, and the two can live side by side. It cannot organise by your eight spaces, link to Figma layers, or be shared with your team as a brand tool. The library does those three things.
Will the library also handle video?
It can. Video is catalogued from a still frame today. Storing and streaming the video itself is possible in the cloud plan, but video is far heavier than images, so it is a separate decision with its own cost (see section 09).
Does it work without me tagging anything?
Yes. The tagging is done by AI from your vocabulary. Your only job is a short review of the Space tag, which is the one judgement call.
Where we are today
One Figma page is catalogued, tagged and live on the portal; nothing has been reviewed by a person yet.
| Figma page | Visual Ref – Packaging, in the inspiration file |
| Images | 607 found · 601 tagged · 48 duplicates linked · 64 named sources |
| Time | 35 minutes end to end, about 3.5 seconds per image |
| Figma calls | 141, all read-only, no AI credits |
| Library online | studio-library-three.vercel.app (open, hidden from search engines) |
| Private copy | claude.ai artifact |
| Library folder | ~/Documents/Auspicious Reference Library |
| Code and tags | experiments/reference-catalogue in the repo |
Images on the page today are Figma screenshots at up to 1,200 pixels, not the original files. Section 05 explains why that matters.
An image's journey today
Figma board → screenshot → check → tag → library folder → your Mac builds the site → Vercel serves its copy.
Questions
If the files are on my Mac, how does the website show them?
Vercel never reads your Mac. When we deploy, your Mac builds the site, copies the images into it and uploads the lot to Vercel. Vercel then serves its own copy, so the site works with your Mac switched off.
If I change something in the folder, does the site update?
No. Each deploy is a snapshot. Changes appear online only after the next deploy.
Why did some images need re-rendering?
Large Figma screenshots sometimes drop images, leaving black boxes, or show a layer that sits on top. In the pilot 107 of 607 needed a second, single-image render. Automatic checks catch these now.
Where everything lives
Each place has one job, and images only ever sit in the places marked for them.
- Your boards and the high-quality originals you uploaded
- Never changed by the library
- Images, one file per Figma layer
- Tags and the built page
- Rules, process and status docs
- Not backed up by git
- Scripts, vocabulary, tags, manifests
- No images, ever: git keeps every file forever
- The portal and a snapshot of the library
- Serves pages to anyone with the link
- Not a storage drive; it only holds what each deploy sends
- The same library, visible only to you
- Size cap of about 2,000 images per page
Questions
Why can't images go in the repo?
Git remembers every version of every file forever. A gigabyte of images added once stays in the repo's history even after deletion, slowing every clone and backup. The repo is already 2.2 GB, and 370 MB of its history is images.
What does Vercel actually do?
Vercel is the web host. It takes the folder of finished pages your Mac uploads and serves it fast around the world. It also hosts the live auspicious.living website, which is why portal deploys use special settings that leave the live domains alone.
Could someone else deploy and break the library?
A deploy from a computer without the library folder would publish a placeholder in place of the images. Today, library deploys should come from your Mac. The cloud plan in section 07 removes this limit.
Screenshots or originals
Download each original once from Figma, keep it as the master, and make every smaller version from it.
When you place a photo in Figma, Figma keeps the full file you uploaded. The pilot took screenshots of the boards, at the size they appear, capped at 1,200 pixels. That was quick, but it threw quality away: a 4,000-pixel photo became a 900-pixel copy.
| Screenshot (pilot) | Original from Figma | |
|---|---|---|
| Quality | As small as it looks on the board | The full file you uploaded |
| Typical size | 60 KB, up to 1,200 px | 2 to 5 MB, often 3,000+ px (estimate) |
| Covered or black images | Happens; needs checks | Cannot happen |
| Duplicates | Found by comparing pixels | Found exactly by Figma's image ID |
| Do it again later? | Yes, to get better quality | No, done once |
Questions
How much space would 10,000 originals take?
Roughly 20 to 50 GB, assuming 2 to 5 MB per image. The real figure will come from the Packaging page test before anything bigger.
Where should originals live: the repo or the library folder?
Never the repo. They go to the cloud store as the master, with a copy in the library folder or on an external drive as a backup.
Does downloading originals touch Figma?
No. It reads the file Figma already holds. Nothing in your Figma file changes, and it uses no AI credits.
Compression, without losing the master
Keep one perfect original; serve small, compressed copies made from it on request.
Nobody browsing a grid needs a 5 MB photo. An image service can take the original and hand back exactly the size the screen needs, in a modern format, at a sensible quality. The original is never altered.
Sizes are typical estimates for WebP at medium quality; the test will give real figures.
Questions
Why does compression matter so much?
Bandwidth is what hosts charge for and what makes pages slow. Browsing 100 images in the grid costs about 3 MB with compressed copies, or 400 MB with originals.
Do we make and store all those sizes ourselves?
Not in the cloud plan. The image service makes each size the first time it is asked for, keeps it cached and serves it from then on. We store only originals.
What do we do today?
Today the build makes two sizes, 480 px and 1,200 px, and packs them into the website. It works at 600 images, but every deploy re-sends them, which does not scale.
The cloud plan
Originals live once in a cloud image store; the website carries only the index; you reach it from anywhere.
studio-library, shared by the whole studio. Keeps each master once, with a grid and a viewer size made from it.Figma → importer → image store + index → library page → you
| Today | Cloud plan | |
|---|---|---|
| Each deploy sends | 59 MB now, about 1 GB at 10,000 images | About 1 to 10 MB, whatever the size |
| Images uploaded | All again, every deploy | Each one once; repeats are skipped by file fingerprint |
| Deploy from | Your Mac only | Any computer |
| Image quality | Screenshots | Masters, with two served sizes made once |
Questions
Why Cloudflare R2?
Vick chose it on 26 October 2026. R2 is the cheapest place to keep files: the first 10 GB a month are free, then $0.015 per GB, and sending images to viewers is free. It belongs to the studio, not to Auspicious, so the planned studio library uses the same store without uploading anything again.
Who makes the smaller sizes?
The upload script, once, on the Mac: a 480 px grid copy and a 1,400 px viewer copy in WebP, stored next to the master. A small Cloudflare Worker serves them, cached for a year. Cloudflare Images, which resizes on request (5,000 free a month, then $0.50 per 1,000), is not needed for two fixed sizes.
Is the cloud version private?
As private as today: images can be opened by anyone who has their exact address, and search engines are told to stay away. A password gate on the portal is possible later if you want it.
What happens to images already uploaded when we deploy again?
Nothing. They stay in the store. Each image is filed under a fingerprint of its file, so the importer sees it is already there and skips it, and the same picture placed on two boards is stored once. Deploys send only the page and the index.
The life of one image
One image costs one Figma download, a few seconds of AI, one stored original and a tiny line in the index.
- FoundListed from the Figma page with its section, layer and image ID.shared call
- Downloaded onceThe original file comes from Figma. Skipped if its image ID is already stored.1 Figma call · ≈ 4 MB
- StoredUploaded to the image store; a copy goes to the backup drive.≈ 4 MB cloud + 4 MB drive
- TaggedSeen on a contact sheet of 20; gets space, type, light, colour, caption, brand.≈ 3 s of AI
- IndexedOne record of tags and links is added to the index.≈ 1 KB
- ReviewedYou confirm or correct its Space. Your answer is kept forever.≈ 4 s of you
- ViewedEach look loads a small copy: the grid, then the viewer if opened.≈ 30 KB / 250 KB
| Images | Stored originals | Index | AI tagging time | Your review (Space) |
|---|---|---|---|---|
| 1 | ≈ 4 MB | ≈ 1 KB | ≈ 3 s | ≈ 4 s |
| 1,000 | ≈ 4 GB | ≈ 1 MB | ≈ 20 min | ≈ 70 min |
| 10,000 | ≈ 20–50 GB | ≈ 10 MB | ≈ 3.5 hours, over several days | ≈ 11 hours, in short sittings |
Estimates from the pilot's rate. AI tagging runs on your Claude subscription; at about 9% of a 5-hour window per 600 images, 10,000 images are spread over several days.
What it costs
Free up to about 10 GB of images; a few dollars a year beyond that; plus a drive once.
| Item | Today (601) | About 5,000 | About 10,000 |
|---|---|---|---|
| Image store (Cloudflare R2) | Free: 655 MB for 3,905 images | Free: about 0.5 GB with two sizes | Free up to 10 GB; originals at 20–50 GB cost $0.15–0.60 a month |
| Website (Vercel) | Your current plan | No change; deploys stay small | No change |
| AI tagging (Claude) | Within subscription | Within subscription, paced | Within subscription, over several days |
| Figma | No credits | No credits | No credits |
| Backup drive | — | One-off, 1–2 TB portable SSD | Same drive |
| Yearly running cost | $0 extra | $0 extra | $0–7 a year |
Questions
What about video?
Video files are 50 to 500 MB each, and streaming them smoothly needs a video service with its own fees. Storing a few dozen short clips fits the plans above; a large video archive deserves its own costing first.
How does this compare with mymind?
mymind costs $79 a year (Student of Life) or $129 a year (Mastermind, with video up to 500 MB). Ours is free up to about 10 GB of images and a few dollars a year beyond, with your own vocabulary, Figma links and team access.
Could it be cheaper?
Hardly: this is the cheapest of the options compared. Making the two sizes once on the Mac avoids paying for resizing on request.
Are these prices certain?
They are the published prices read on 22 October 2026 (sources at the end). Prices change, and your own plan may differ, so we confirm them before moving anything.
Your Mac and an external drive
Yes, get a fast portable SSD, but as the backup, with the cloud as the working copy.
- Why not the drive alone? Drives get lost, dropped or fail, and a drive is not reachable from your phone or the studio.
- What size? 1 TB covers 10,000 images many times over; 2 TB leaves room for video.
- What type? A portable SSD (solid-state): fast, light and quiet. Keep it out of the repo folder and plug it in when the importer runs.
- Your Mac keeps only the code and, if you like, the small index. The library folder can move to the drive.
Rules and safety
Figma is only read, images never enter git, the live website is never touched, and every fix sticks.
- No residue in Figma. Every Figma action is a read. Nothing is created, moved or changed, and no AI credits are used.
- No residue on your Mac. Temporary renders and build copies are deleted after each run; this session cleaned up about 2 GB of its own.
- No images in git, ever. Code, tags and the index only.
- The live website stays separate. Portal deploys never reach auspicious.living; going live there is its own conversation.
- Your corrections win. A tag you fix is stored apart from the AI's guess, and re-runs never overwrite it.
- One page at a time. Whoever works on a page claims it in STATUS.md first, so two agents never collide.
- Tags come from your word list. A check stops any run that invents a word.
The full rules are in docs/RULES.md, with the step-by-step method in docs/PROCESS.md, both in this folder and in the repo.
The phases
Prove the cloud set-up on one page, then let it run through the rest of Figma.
Words, plainly
- Repo
- The project's code folder, tracked by git and stored on GitHub.
- Git
- The tool that remembers every version of every file, forever.
- Deploy
- Sending a finished copy of the website to Vercel so people can visit it.
- Vercel
- The company that hosts the portal and your live website.
- Original
- The full image file you uploaded to Figma.
- Render or screenshot
- A picture of how the image looks on a Figma board.
- Compression
- Making an image file smaller, with a little or no visible quality loss.
- CDN
- A network of servers that delivers files quickly from somewhere near the viewer.
- Image CDN
- A CDN that also resizes and compresses images on request.
- Cloudflare R2
- Cloudflare's file store, where the studio's images live. A Worker is a small program on Cloudflare that hands them to the page.
- Index or manifest
- The list of every image with its tags, colours, caption and Figma link.
- Image ID
- Figma's own name for an uploaded file. The same photo placed ten times shares one.
- Bandwidth
- The amount of data sent to viewers; what hosts charge for.
- SSD
- A fast, silent storage drive with no moving parts.
Prices read on 22 October 2026 from the providers' own pages:
Size and time figures marked as estimates come from the pilot and will be replaced by measured ones after the cloud test.