A day collecting, rating, and assembling direction for a character redesign

The flow is not linear. A concept artist's reference workflow overlaps four parallel jobs: collecting the raw material, filtering it for the direction you're arguing for, assembling it into a moodboard that communicates intent, and keeping the archive searchable for the next project phase or the next character that needs the same kind of energy.

A reference system that is not built around iteration collapses as direction changes. Coii Ref handles this because references live in folders on your drive the whole time, tags are metadata you layer on top rather than a structure that requires reorganising, and moodboards are plain files you can move anywhere — into a presentation, into revision history, or into a locked archive at day's end.

This is what a day looks like when it works.

9:00 am — Open the project workspace and scan what you have

The project folder is /Volumes/ProjectDrive/Character-Direction/ and it's the root of your workspace. Two subfolders hold the real work: References/ and Boards/. The workspace also writes three Coii things beside them: the database, thumbnails, and canvas-resolution proxies.

Open the app and start with a view you saved yesterday:

tag:characterA rating:>=3 -is:favourite

This pulls all the images rated as keepers for character A that are not yet flagged as final. Forty-two images. Scroll through; none of them feel right for the new direction the director sketched this morning. They were good for the last pass, and now they're a starting point for what comes next.

9:30 am — Search the archive for the silhouette the director described

She said "Miyazaki's Nausicaä, but grounded in our world." That's three things to hunt: organic curves instead of angular geometry, a defined waistline, movement implied by fabric. None of it was in yesterday's collection.

Search is one line:

type:image tag:any:silhouette,anatomy in:/Volumes/Archive rating:>=2

This hits the archive drive, pulls silhouettes and anatomy studies rated at least acceptable from previous projects, and brings up one hundred and fourteen images. Start a new canvas—name it "Direction A"—and begin dragging the images that match the sketch onto it. A canvas is the fastest way to audition a direction before committing anything to the final tag.

Not every image lands on the board. Some hit the source of inspiration but don't quite fit the brief. Rate those up to 4, tag them direction-a-adjacent, and move on. They're not wasted; they're options for "if the director changes their mind in the next round."

10:30 am — Build out the board with detail passes

The silhouettes are solid. Now add fabric studies, hand gestures, and expression options. For each, search something specific:

type:image tag:fabric in:/Volumes/Archive width:>2000

type:image tag:hands rating:>=3 added:>2026-08-01

The width filter finds detail enough to paint from. The date filter finds your most recent hand studies, because old ones get progressively less relevant as your style evolves. Drag the best matches onto the board.

Build the canvas as a 9-plate sequence: silhouette pass, fabric flow, hand options, expression options, color studies, material treatments, final composition, turnaround reference, scale comparison. Each plate is a single image from the library, so a plate can be a full study sheet or a close-up; the board holds them all at canvas resolution for scrolling and comparing without leaving the app.

12:00 pm — Export the board and show the director

Save the canvas to the project folder and export it—the JSON file shows up in Finder and goes straight into Slack. Two minutes later the director has the link and can open it in a text editor to read the file names, or open the project folder to see the library and browse it in context.

She says: "I like the silhouette, but the fabric is too soft. More sculptural. And forget the expression options—I want her more stoic."

That is a new direction, not a tweak. Start over.

1:00 pm — Create the Direction B space and cross-tag

Create a new canvas called "Direction B" and a new tag: direction-b. Go back to the silhouettes that matched Direction A, search them all again:

tag:direction-a rating:>=3 in:/Volumes/Archive

View them in the app and re-tag each one. Those silhouettes work for Direction B too—they're directional enough. Just add the direction-b tag to the same image. An image can hold as many tags as it earns; you're not moving anything, just layering metadata on top.

The fabric studies are different. Pull up:

type:image tag:fabric in:/Volumes/Archive

But this time search for structured folds instead—a different query line is faster than re-tagging three hundred images:

type:image tag:sculpture,structure in:/Volumes/Archive

The archive is thin on structured fabric studies—you've never had a project that needed that exact thing before. Note this as a gap: tomorrow's reference collecting has a new target.

2:30 pm — Build the Direction B board

Eight plates this time, because the director is more decisive on the second pass. Silhouette, fabric structure, gesture, detail focus, color palette, reference comparison, final pose, scale.

Drag each image onto the new canvas. This is faster than Direction A because you already know the archive. You know where the fabric studies are, which year had the good anatomy references, that the best gesture images are tagged movement not dance.

3:30 pm — Rate and finalize Direction A as archive

Direction A is no longer active. Move it to history. Search the direction-a tag and rate all of them up by one point—they were useful for exploring an option, and that's worth recording. Drop the direction-a tag and replace it with archived-direction. The images don't move; the metadata just shifts, and any view built on the old tag stops working the moment you change it, which is exactly when you want to know about it.

tag:archived-direction rating:>=3

Five images got promoted into keepers. They're going into the active direction-b set.

4:00 pm — Set up the ongoing reference view

Searches you run more than once are worth saving. Direction B is active for the next week, so save this view:

tag:direction-b rating:>=3

Save another:

tag:characterA rating:>=4 -is:favourite in:/Volumes/ProjectDrive

This is your "approved directions from this character, not yet final," and it becomes the query line you check every morning while the character is in production.

4:30 pm — Collect new references from research

Three hours into production, your modeller sends a Slack image of a reference they found: a musician's vest with sculptural pleating. Download it to References/Vendor/ with a descriptive name: Vogue-sculptural-vest-2024.jpg

The watcher sees the new file immediately and adds it to the index. Search all new images from today:

added:2026-09-11 type:image

This pulls that one file plus anything else that arrived today. Tag it direction-b, rate it 5, and add a note: "Pleating reference for shoulders." The image never leaves Vendor/; you tagged it, rated it, and noted it while it lives exactly where it landed.

Pull it onto the active Direction B canvas as plate eight: reference comparison.

5:00 pm — End-of-day check: consistency and gaps

Before you finish, run a sanity check. Search for every image tagged characterA:

tag:characterA

Sixty-eight images. One of them is rated 0—it means "I looked at it, not a match." A few are rated 2—"maybe if we revise." Twenty-two are rated 3 or higher. That's the shape you expect: most images are trying-ons, a few are directions, and only one set are keepers for the current phase.

Search for images you added today:

added:2026-09-11

One image. Good—you're not drowning in raw material yet.

Last search—the gap you identified this morning:

tag:sculpture type:image in:/Volumes/Archive

Twenty images total. For a character at the sculptural detail phase, that's thin. Add to your notes: "Need to build structured fabric library for next project."

Close the app. The canvas, tagged and rated, sits in the project folder as JSON. The image files sit in their original folders. Nothing got copied, nothing got reorganised, and nothing got locked into a library structure that depends on the app still existing to open it.

A week later: Direction B is approved

The director approves Direction B. Search direction-b and mark every image as 5-star.

tag:direction-b rating:>=4

Now it's the canonical reference set. Create one final canvas—"Approved Direction"—and add every image rated 5. Save the JSON to the project folder as the reference archive. Tag every image with approved:characterA so a future phase can pull them back out.

Close those temporary canvases. They're still in Boards/ if you need to reference what you considered, but the active set is the one that proved out.

Where this fits with building a library over years

Every project adds images. Every direction adds tags. Over three years of character work, your References folder holds eight thousand images across four drives. The app still searches them all with the same one-line query.

Old projects are tagged archived:2024, archived:2025. New projects inherit tags from adjacent work—a rigging job adds to tag:anatomy, an environment adds to tag:structure. Saved views become templates: when a new character starts, you open the view from the last character and run it again against the new project folder. The query is the same; the folder is different; the results are what you collected for the last similar job.

When a vendor image gets used in production, you add a tag used:projectName. Search used:* to count how much of the archive actually made it to final art. The ones that did not are still there, still tagged, still rated—and they become the starting point for the next project that needs the same reference type.

Moodboards are version control. Save Direction A, Direction B, and Approved Direction as separate canvases in the project folder. Three weeks from now, the director asks "what happened to those sculptural pleats we looked at?" Run the search, pull up Direction A, and the answer is right there. The boards are not screenshots or exports—they're plain files in revision history, which means you can recover a decision you made and forgot.

How to organize references by project covers the folder structure that makes this work at scale. How to rate and filter digs into the tagging and rating system. Building a moodboard is the technical step. What is a reference manager answers whether you even need one. And the reference manager for Mac is the full feature list if you want to know what is not in this workflow—the things that don't come up in a day like this one.

Managing revisions: when Direction B becomes Direction C

This workflow does not require a perfect forecast of the direction. The point is that you can pivot fast without losing what you learned.

On day four, the director sends feedback: "Direction B is great for the silhouette, but the fabric is now too stiff. Let's go sculptural but with movement." This is a direction shift, not a direction kill. You keep everything from B that worked, but pivot the fabric studies.

Search and rate all images tagged direction-b and flagged as keepers. Create a new canvas: "Direction C." Copy the silhouettes from B as your first six plates. For the remaining three plates, pull new fabric studies from the archive—this time prioritising images you ranked lower in B because they had movement in the drape.

The old Direction B canvas stays in the project folder. Do not delete it. If the director changes their mind again in a week, Direction B is not lost—it is a branch you can return to. Git for moodboards.

Each canvas is a separate JSON file, which means each one is version-controlled if your project uses git, backed up if you back up the project folder, and recoverable if you moved the files somewhere else six months ago and forgot where.

Collaborating with modellers and riggers

The system scales to teams because it is built on the folder structure and metadata, not on a proprietary library database.

A character concept flows from concept art to modelling. When you hand off, pass the project folder to the modeller with all of it: your original images, your canvases, and the tags and ratings you applied. The modeller opens the workspace and does not need to import or re-tag anything. They can immediately search for what you found useful—tag:characterA rating:>=4—and see the final direction without scrolling through iterations.

If the modeller finds new reference material during their pass—sculpts, reference photos of how a fabric actually drapes when a figure moves—those go into References/ with a clear naming convention. You immediately see them on your own machine through the watcher. Tag them, rate them, and integrate them into the conversation: does this reference suggest we need to adjust Direction C? Pull the new image onto the Direction C canvas and ask the team.

Riggers work from similar reference. They care about joints, balance, weight distribution and how specific materials constrain movement. A query like tag:movement,skeleton rating:>=3 pulls every reference about anatomy and motion you collected. That is what they need. You do not need a separate "rigging reference library." You need one library and tags specific enough that a query can slice into just the part of it another discipline needs.

Rate as you go, not in a second pass

This is the most important habit to establish early. Rating is not cleanup. It is the only thing that separates a searchable library from a folder.

The instinct is to leave rating for later—to get all your references in first and rate them when you have time for a big review. But future-you has nine other projects and will never get to that big review. Instead, rate as you drag an image onto a canvas, or ten seconds after you add a new reference to the folder.

tag:direction-a images as you pull them onto the board. Rating takes no longer than an emoji reaction and it is the only thing that will make the board searchable when you want to find it again in six months.

One more month in, you will have a library where the images you actually use are flagged as keepers, the ones you tried and rejected are marked down, and the images you never looked at are left at default. A search for any of the direction tags brings up the best versions first. That is the difference between a library and chaos.

The archive that becomes next month's reference

At the end of this character's production, you have eight thousand images, seventeen canvases, and five direction iterations documented in the project folder as separate JSON files. This character is done. Next character starts tomorrow.

Do not delete yesterday's archive. Move it to an external drive or a locked folder in your main project directory, tag everything in it as archived:character-name, and leave it searchable.

Three months from now you will remember something about this character's skeleton that worked. Search tag:characterA rating:>=4 in:/Volumes/Archive and there it is—not in the active project, but available whenever you need to reference what you built last time.

This is where a reference library stops being a project artifact and starts being institutional knowledge. The next character that needs "organic curves with a defined waistline" will search tag:silhouette:organic rating:>=4 added:>2026-03-01 and what you learned about that silhouette shape from this character is the foundation for the next one.

Or just download it and try this exact workflow on a real project folder—thirty days, every feature, no card. The rate-and-filter habit is the difference between a reference library and a folder you scroll through forever.