How To

Getting started

The app runs entirely in your browser — nothing to install. Sign in with your work email on the login page; accounts are invite-only, and new signups need a one-time email verification (check your inbox for the link). Forgot your password? Use the reset link on the login page — it is fully self-serve.

Once signed in, the top navigation is the whole app: Generate builds creatives, Library holds every past run — open one to zoom, MagicFix, and download, Image Search finds source photos, Meta Assets shows everything pushed to Meta, and Changelog lists what is new.

The Generate page

Generate is the production hub. Five tabs across the top pick a way of working — Light, Ultra, Bulk Upload, AG Guide, and Creative Agent — each documented below. Two things live above the tabs:

• Recent runs — your latest generate runs, one click to reopen any of them in the Library viewer.
• Visual treatment — applies to Light and Ultra renders. Photo — bright (the default) puts copy straight on the photo with a dark outline so white type stays readable. Scrim text draws a rounded black panel behind each text block. Dark hero dims the whole canvas and cuts a feathered spotlight around the equipment.

Light Generate tab

Build branded banners in every approved size from your own photos and copy. Best when you already know what you want to say.

1. Output prefix. Names the run and every file in it. Letters, digits, hyphen, underscore; up to 40 characters (e.g. edmonton_q3).
2. Campaign type. Leave on auto-detect (it flips to Event when any proof point looks like a date) or set Event / Evergreen / Awareness explicitly. This drives Meta routing.
3. Images. Upload up to 20 photos. Each is classified on upload so you can filter to single-equipment shots — single-unit hero photos have historically out-performed fleet shots about 2× on the same audience.
4. Copy. One line per variant, and the columns pair up by line number: headline 1 goes with subheadline 1, proof point 1, badge 1, and CTA 1. A proof point (a date, dollar figure, or named auction) flips that line's layout — the headline demotes to a small all-caps label and the proof point becomes the dominant line. Badges are 8 characters max (e.g. SOLD), CTA pills 16 (e.g. View listings).
5. Dimensions. Tick the sizes you need, or add a custom width × height (32–12,000 px per axis).
6. Generate. Watch progress on the page; finished runs open straight in the Library viewer.

Ultra Generate tab · Meta-driven

Start from what is already winning. Ultra reads your live Meta performance and generates new creatives that imitate your best ads.

1. Pick an ad account, and optionally type to drill into one campaign. Leave the campaign blank to rank winners across the whole account.
2. Refresh inspiration. Pulls active campaigns and the last 30 days of performance from Meta. Results are cached for a day per account; the button re-runs the pipeline.
3. Read "What's working." A KPI row and a pattern callout summarize what your spend data says is performing.
4. Pick 1–4 winners to imitate. Their visual style and copy patterns seed the generator. Hover a card to see which sizes Meta is actually spending on for it.
5. Copy instructions (optional). A free-form brief for the Claude copywriter, e.g. "big online auction May 11 in Edmonton — short punchy copy with CTA." Brand constraints still apply.
6. Set the fan-out. Heroes per winner × copy pairs per winner controls how many creatives you get. Sizes default to the dimensions Meta is currently spending on; override with the checkboxes.
7. Prefix and Generate. The run lands in the Library like any other.

Bulk Upload Generate tab · Name-routed

Take a folder of finished creatives, match each file to the right live Meta campaign and ad set by its filename, and push the whole batch in one go — every ad created PAUSED.

1. Pick the ad account.
2. Drop your files (JPG/PNG) on the dropzone — drop or pick as many times as you like; new files add to the staged set.
3. Check the preview table. Filenames are matched loosely: the tool scans for any mix of location, industry, language, month, year, quarter, size, and brand markers. Event files need at least 4 of those labels, brand files at least 2. A red BELOW-THRESHOLD chip lists what is missing from the filename; a yellow REVIEW chip means it just barely matched — double-check the matched campaigns before pushing. "Show naming conventions" on the tab documents the standard patterns, and Meta's _105 re-upload suffix is stripped automatically.
4. Ad copy (optional). Leave everything blank to inherit each matched ad's existing copy, or open the panel to set up to 5 body/title variants for A/B testing, a call-to-action, URL tracking parameters, a link description, and a label suffix for ad names. Destination URLs come from the shared event schedule sheet automatically on every push — each ad is matched by the event code and month in its filename; upload a CSV only to override a single push.
5. Push. The button tells you exactly how many ads are about to go up, always as PAUSED. The results table shows per-file status with a direct "open in Ads Manager" link for each new ad.

AG Guide Generate tab · Print PDF

Build the press-ready Ag Auction Guide page — photos, auction details, Territory Manager, and consignors — and download a print-ready PDF.

1. Pick a template, then create or select a build. Builds are shared with the whole team: anyone can open one, restore a checkpoint, and save new checkpoints onto it (each checkpoint shows who saved it); only the creator can delete a build. Save named checkpoints as you work ("before reordering photos") and roll back to any of them — checkpoints live on the server, so they survive a reload and follow you to any device. Search all saved work by sale name, town, people on the sale, auction date, checkpoint name, save date, or teammate; the line under the search box says how many saved pages it covers, including the ones teammates saved. The matching checkpoints show underneath; select one to open its build, then choose Restore when you are ready.
2. Add photos. Use "Add photos in bulk" or fill slots one by one. Each box is the shape of the printed image (with its one-line printed name) and shows the crop the page will get. Framing is automatic: the tool finds the equipment and zooms toward it as far as it can without cutting anything off, and the small readout beside each box's zoom pill shows the factor it expects to print at (worked out from a reduced copy of the photo, so it can differ from the page by a hair). To take over, press + or −: the thumbnail changes as you click, the percentage prints exactly as shown, and clicking the percentage returns to Auto. Photos get the same levels lift the designers apply (dull skies to white, mid-tones up) and are separated to press CMYK like a designer's page; uncheck "Auto-brighten" to print a photo as shot. Ordering is automatic by equipment category, or switch to custom and drag to arrange. Photo codes from the SOP sort automatically, and printed captions are generated for each photo.
3. Paste the sale write-up from the Word doc or Wrike and click Auto-fill fields from text — the auction fields below fill themselves, and bulleted equipment lines become the printed Equipment Highlights. Always review the filled fields before generating.
4. Complete the auction details — date, auction type, location, seller, directions, equipment-viewing dates, GPS, legal land description, auction URL, and page number.
5. Territory Manager. Search the saved TM library by name to fill everything in one click, or type new details and hit "Save TM" to add them. Headshots are cropped face-aware so the head sits correctly in the portrait slot, and a pink or magenta cast on the skin is neutralised automatically (skin only; backdrop, shirt and hair print as shot). A retouched headshot in the library still gives the best result.
6. Seller contact. Up to two consignors print in the page footer; leave blank to use the seller line.
7. Hero image. Answering Yes to "Do you have a Hero Image ready to upload?" opens an upload box. The image is not printed on the guide page; it is attached to the Asana task for the auction web page build. It saves with your checkpoints, and switching the answer back to No clears it.
8. Generate Press PDF. Check the warnings list, inspect the zoomable preview, then download. The status line shows the upload as it goes; photos restored from a saved build or already saved in a checkpoint are not uploaded again (the page sends a reference), so a re-render takes seconds. A dropped connection is named as such and retried once. Photos come through with their true color, so the equipment on the page looks like the photos you uploaded. Type and brand color are set for the press, in the guide's Inter typeface at the exact 8.25 × 10.75 in trim with bleed, so pages match the rest of the printed guide.
9. Finalize to Asana. Pick the guide's Asana project, answer the Asana-only questions, and click Finalize: the task is created with the sale details, the press PDF, the photos and the blueprint subtasks. Finalizing also saves the sale as a build (a "Finalized: ..." checkpoint appears under Save progress), and that build is what links the sale to its task: finalizing the same build again updates the task in place, replacing its notes, PDF and photos, instead of creating a second one. After a reload, open the sale's build from the list first so the button reads "Update Asana task". Later edits to key details also post a change notice on the task whenever you save a checkpoint.

Creative Agent Generate tab · Chat

Set up Meta campaigns by talking, powered by Claude. Describe what you want in plain language and the agent does the campaign and ad-set legwork with you.

• Chat naturally. "Set up a retargeting campaign for the June Edmonton auction" is a fine place to start. The agent asks for whatever it is missing.
• Attach photos to the chat and rendered previews appear inline, so you confirm each creative before anything is pushed.
• Start from scratch. No reference creative needed — give the agent the subject, objective, facts, and your visual or copy direction in your own words. The agent does not choose a template, lens, palette, crop, medium, lighting setup, or copy formula for you. It defaults to one concept and only creates multiple directions when you request them. You can also explicitly ask it to interpret the brief freely. The authentic RB logo, kept legible, is the only fixed visual brand decision. State the total number of final images you want; the confirmation card locks that exact count before generation. If you need to halt an active run, use Stop generation on its progress card. Completed files are kept and queued work is cancelled.
• Your conversations persist. Threads are saved to your login and reload when you come back.
• Nothing ships silently. Every push to Meta asks for your explicit confirmation first, and every ad is created PAUSED.

Maintenance Agent Generate tab · Chat

Ask for a change to Bulk Upload in plain English, or paste the Asana task. The agent makes the change, proves it with the automated checks, and tells you how to see it. Available to a short allowlist; the tab is greyed out otherwise.

• What it maintains. Two kinds of thing. The words Bulk Upload recognises in image file names (industries and their short forms, event location codes, the call-to-action list), and the behaviour behind the tab: which campaigns and ad groups a file lands in, what the preview shows, how the paused ads are built, the URL and date automation, and the tab itself.
• How to ask. "Recognise Mining as an industry in image file names", "The Montreal campaign says there is no live ad group but there is", "Keep the files from my first batch when I add a second", or paste the Asana task link (screenshots on the task are read too). Give the exact word or the exact campaign and ad group name, and what should happen instead.
• What happens. The agent reads the code, asks an independent strategist for the plan, makes the change, and then the checks run: the automated tests, a check of the real account (every existing file name and ad group is classified before and after, and anything that would route differently is listed), and two independent reviewers who must say in one sentence what the change does for you. A word-list change goes live on its own once everything clears, in roughly 20 to 30 minutes. A behaviour change runs the automated checks first (about 10 minutes), then waits for Ishti to confirm it on a card, then goes live; if the checks do not pass, nothing changes and you can ask the agent to try once more. The agent tells you which kind yours is. You get plain-English verify steps and a note on the Asana task.
• Short forms. Everyday words such as "for" and one- or two-letter forms are not added on their own, because they would match ordinary file and ad group names; the agent can show you how often a word already appears in the account's real names, uses the full word, and Ishti decides the exception.
• When it declines. The tool's login, the Meta connection itself, the other tabs and anything that would turn ads on are out of bounds; those go to Ishti with a note. That is expected, not an error.
• Fresh chat. The tab opens a new conversation each time; earlier ones are in the picker at the top.

Library

Every past generate run, newest first — and the place to open one: see every creative at a glance, fix the fixable, and download the set.

• Open a run by clicking its card, or follow the "View run" link from Generate's results, the Recent runs list, or a Creative Agent run card. All the run's creatives render as one grid.
• Filter by size or image name. Creatives whose copy lacks specificity (no date, dollar figure, or place) carry a "no proof point" chip right on the grid.
• Click a creative to zoom. Step through the run with ← / →, close with Esc.
• MagicFix. In the zoomed view, describe what to change — "brighten the machine", "remove the second excavator" — and the image model re-renders that creative in place.
• Download as zip grabs every image in the run, named and foldered exactly as Bulk Upload expects.

Image Search

Natural-language search across the Ritchie Bros. reference photo library. Describe the shot you need; get the closest photos back.

• Type a plain description — "yellow bulldozer at an auction yard", "aerial view of a full yard", "close-up excavator bucket". No tags or exact keywords needed.
• Read the relevance badges. Results are scored perfect / strong / partial / weak so you can tell an exact hit from a near miss at a glance.
• Use it to feed Generate. Found the right source photo? Download it and upload it on the Light tab, or reference it in a Creative Agent chat.

Meta Assets

A gallery of every creative this app has pushed to Meta's image library, grouped by run. The source of truth is Meta itself — thumbnails are re-fetched fresh on every page load, so what you see is what Meta has. Use Upload a run to send a finished generate run's images to the Meta library, and the filter box to find a file by name, run, or hash.

Usage

A live feed of who used which feature and when — "Venice used AG Guide at 12:30pm PDT" — grouped by day, newest first, with all times in US Pacific. It refreshes itself every 30 seconds. Repeated activity within ten minutes (say, several searches in a row) is recorded as one use, so the feed stays readable.