Library — every draft and every image, never lost

The hardest part of a creative workflow is trust. If you spend two minutes generating an image, change your mind, and the image disappears, you stop generating. You start hedging — you generate less, you commit less, you write less. The Library exists so that never happens. Every post you save lands here. Every image the AI produces lands here. Every video you generate or upload lands here. The moment the AI gives you something, the Library remembers it on your behalf, so you can come back tomorrow, next week, or next month and pick up where you left off.

What’s on the page

Three tabs at the top:

Drafts & posts — every social post the system has saved for this venue, newest first. That includes drafts you saved and never scheduled, posts pending approval, posts scheduled to go out, posts that already went live, posts that failed to publish, and posts you cancelled. Each card shows the channel, the status (Draft / Scheduled / Published / etc.), the caption preview, and the image thumbnail. Click Copy to put the caption and hashtags on your clipboard, Download to save the image, Trash to delete the row entirely. Filter pills above the grid let you narrow to just one status — handy when you want “show me only the published posts from last month” or “show me everything that’s still a draft.”

Generated images — every image the AI ever produced for you (plus anything you uploaded), in a tidy grid. Each tile is the image with a small badge in the corner: Generated, Edited (cropped or recoloured), Uploaded, or AI Edit (described below). Every tile carries one clear button — Edit with AI — and a ⋯ menu for everything else (crop, download, copy link, remove). Tap the photo itself to open the full-screen viewer, where you can move between images with the arrow keys. These controls are always visible and finger-sized, so the page works the same on a phone as on a laptop.

Generated videos — every clip you generated with the AI or uploaded yourself. Hover any tile for a silent autoplay preview; click any tile to open the full-screen viewer with playback controls and audio. The viewer is the same one images use, minus the editing affordances (no video editor yet).

The rule

You paid for it, you keep it. Every credit burn leaves an artefact you can find later.

This isn’t an empty promise — it’s a literal one. The Composer’s Generate button calls the AI, the AI returns an image, and the system writes a Library row before the image even reaches your screen. If you close the tab without saving the draft, the image is still in the Library. If you come back next month and want to repost that truffle pizza shot, it’s there. The credits you spent didn’t evaporate.

How to use it

Open Growth → Library. The page opens on the Drafts & posts tab by default — the place most operators are looking when they ask “where did my work go?”

Finding a draft you forgot about

The cards are sorted newest first. The most common case is “I started a post on Tuesday and I want to finish it now.” Scan the first row of cards; the Tuesday draft will have a grey Draft pill in the corner. Click anywhere on the card (or the small Pencil icon in the action row) to re-open it in the Composer for editing. Every field — caption, hashtags, alt text, image variants, the active aspect ratio — comes back exactly as you left it. The Composer’s header changes to read “Edit social post” with an “Editing existing draft” pill so you know you’re in resume mode.

Edits save in place via the “Update post” button (no duplicate draft created). You can iterate as many times as you want — the button stays unlocked the whole time. Drafts, cancelled posts, and posts that came back with “Changes requested” all support resume-editing. Posts that are already pending approval, scheduled, publishing, or live can’t be edited — for those you’d duplicate to a new draft.

Filtering by status

Hit the Scheduled filter pill if you want a tighter view than the Calendar offers — the Calendar shows posts on a date grid, the Library shows them as a flat list. Hit Failed to find any post that the auto-publisher couldn’t push to Instagram (the failure reason is on the card; common cause is an expired Meta token, fixable from Connections).

Deleting a post

Every post card has a Trash button, and it works on a post in any state — a draft, one waiting for approval, a failed one, a cancelled one, even one that already went live. Click it, confirm, and the card disappears. The one moment you can’t delete is the few seconds while a post is actually being published: if you catch it mid-flight you’ll see a calm amber note asking you to wait a beat and try again once it lands as Published or Failed.

One thing worth knowing: deleting a post that already went live only removes it from your library here — it does not take the post down from Instagram or Facebook. The card’s confirm message reminds you of this. If you genuinely want it off the platform, remove it on Instagram or Facebook directly, then delete the library card to tidy up. (Until recently the Trash button quietly errored on anything that wasn’t a plain draft — that’s fixed; it now works on every state except the mid-publish moment.)

Reusing a generated image

Switch to Generated images. Find the image you want. Open its ⋯ menu and choose Copy link. Now back in the Composer, paste that link into the Image field — the system uses it as-is instead of generating a new one. Zero new credits charged. This is the trick for reusing an image across multiple channel-formatted posts: generate it once on Instagram, then reuse the same link for the Facebook draft.

Downloading an image for off-platform use

Generated images → open the tile’s ⋯ menu → Download. You get the full-resolution file the AI produced — not the smaller display version. This is what you’d hand to a designer for a print job, paste into Canva at any size, or use as a reference photo for a re-shoot. The smaller version you see in the grid is just a thumbnail for fast browsing.

Editing an image directly from the Library

Open the tile’s ⋯ menu → Crop & adjust. The full-screen image editor opens with the full-resolution original loaded into the canvas — every crop, recolour, and annotation happens at native resolution, not at the smaller display size. When you hit Save, the edit lands in the Library as a new tile (an Edited badge on the corner tells you it’s a re-edit) and the original stays untouched. You can edit the same source ten different ways and keep every variant. (This is the manual editor — distinct from Edit with AI, which restyles the photo from a written instruction.)

Uploading your own photo or video

Inside the Composer sidebar, alongside the Generate button, you’ll find an Upload your own option. It’s the answer to “the AI’s good but I’ve already got the perfect shot on my phone.” Pick a file from your computer — JPEG, PNG, HEIC, or WebP for photos; MP4 or WebM for videos, up to 50 MB. Within a few seconds the file appears in your draft just like a generated image would: as a thumbnail in the Composer preview, with the channel-specific aspect ratio already applied.

What happens behind the scenes differs between images and videos. Photos get the same three-variant treatment as generated images — a thumbnail for fast browsing, a 1000-pixel-wide version for live posts, and the full-resolution original kept safe for downloads and edits. HEIC photos from iPhone are converted to JPEG automatically. Videos keep their original encoding (no re-transcoding) and pick up a poster frame so the Library can show them in the grid before they play. The Library tab badge for uploaded items is sky-blue — Uploaded — so it’s instantly clear which tiles you brought in versus which the AI made.

Uploaded media goes through the same Library route as everything else, meaning all the per-tile actions work: download the original, copy the URL, attach to multiple posts, run an AI prompt edit on top of an uploaded photo. There’s no difference in how the Composer or scheduler treats an uploaded image versus a generated one — once it lives in the Library it’s just an asset.

The Generated videos tab

The third tab — Generated videos — is the videos-only view of the same Library. Every clip you generated with the AI lands here automatically (with a green Generated badge); every video you uploaded yourself shows up too (with a sky-blue Uploaded badge). Sorted newest-first, like everything else.

Hover any video tile and a silent autoplay preview starts — about five seconds in, you’ll be able to tell whether you’ve found the right clip without committing to a full open. Move the mouse away and the preview pauses. Click the tile and the full-screen viewer opens with proper playback controls and audio; the viewer doesn’t autoplay so it never startles you when you click. The viewer is the same one the Images tab uses, minus the editing affordances (there’s no video editor yet — AI prompt-edits and crops only run on still images).

Use the Generated videos tab when you’re hunting for a specific clip across hundreds of generated and uploaded ones. The Images tab gets noisy fast on a venue that posts daily; the Videos tab stays manageable because videos are rarer.

Opening the full-screen viewer

Click anywhere on an image or video tile (not on the hover buttons) and the viewer opens, taking over the whole screen with a dark backdrop. The image is centered; left and right arrows let you move through the gallery, the keyboard arrow keys do the same, and a swipe on a touchscreen also works. Press Esc to close, or click the dark space outside the image. At the bottom of the viewer is a thumbnail strip of every image in the gallery — click any thumbnail to jump straight to it.

The viewer also carries a bottom toolbar with Edit with AI (described below), Crop & adjust (the crop/rotate editor), Download, Copy link, and Remove. Videos use the same viewer; the playback controls appear after you press the central Play button (the viewer doesn’t autoplay so it never startles you when you open it).

Asking the AI to edit an image with a prompt

The fastest way in is the Edit with AI button on any image tile in the grid — it opens that photo with the edit panel already waiting. (You can also open the viewer first and use Edit with AI in the bottom toolbar, or press A.) A side panel slides in with a text box and a row of quick-prompt chips — Warmer lighting, More steam, Cleaner background, Brighter, Cinematic, Less salad. Type what you want changed (or click a chip to fill the box, then refine), then hit Generate edit.

The AI processes the image — usually 10 to 30 seconds — and a brand-new variant lands at the top of your Library. The viewer auto-jumps to it. Iterate from there: another prompt edits the new variant; another edit creates yet another variant. Every version stays in the Library, so you can compare three lighting choices, pick the one you like, and bin the rest later — or keep them all as a styling reference.

The AI Edit panel includes a chip that reads “Iterating on this image — previous versions are kept in the Library.” That promise is literal: each prompt creates a fresh tile rather than overwriting what was there. The original is never lost. Each edit costs the same as generating a new image, so you control how many variants you spin up.

Good prompts are short and specific: “warmer lighting”, “add steam to the pasta”, “replace the wooden board with a marble surface”. Long, hedged prompts confuse the model — pick one thing per generation. If the result isn’t what you wanted, try a different wording rather than adding qualifiers (“actually colder lighting, more blue” often works better than “a bit less warm, lean toward cool”).

Showing the AI one of your own dishes (reference image)

There’s a difference between describing a change and showing it. If you type “change the pasta to our agnolotti del plin”, the AI only has the words “agnolotti del plin” to go on — so it invents a generic plate of stuffed pasta that may look nothing like yours. The fix is to hand it a real photo to copy.

The rule: words tell the AI what to change; a reference image tells it what it should look like.

In the AI Edit panel, under the prompt box, you’ll find a Reference image (optional) row with two buttons:

Use a dish — opens a search box listing your menu. Type a few letters, pick the dish, and its photo becomes the reference. (Only dishes that already have a photo work here — if you pick one without a photo, you’ll be told to choose another or upload one.)
Upload — pick any image from your computer to use as the reference: a plating shot, a mood board, a background you want, a competitor’s dish you’re matching. Uploaded references are used for this one edit and not saved to your Library, so they never clutter it.

Once a reference is attached you’ll see a small thumbnail with its name and an ✕ to remove it. Now write the prompt as an instruction about the reference — “replace the pasta with this dish”, “plate the main like the reference”, “use this background” — and hit Generate edit. The AI keeps the rest of the scene (the table, the hands, the lighting) and changes only what you asked, matching your reference.

A reference image costs exactly the same as any other edit — it’s the same single generation, no extra charge. Occasionally the AI can’t use a reference (its safety check is cautious about some photos); when that happens it falls back to editing from your words alone and tells you so with a calm amber note, so you’re never left wondering why the result ignored your photo.

Cleaning up the Library

If the Library gets noisy after a few months — say you’ve generated 200 images, you only kept 30 — open a tile’s ⋯ menu and choose Remove (or use Remove in the full-screen viewer). That removes the entry from the Library list, but the image file itself stays in storage (in case a saved post still uses it; we’d rather keep a stray file than break a live post). Deletion is per-tile; no bulk delete today.

Worked example

Marco generated nine variants of his Pizza Tartufata image two weeks ago — five 1:1, three 4:5, one 9:16 story. He saved the best 1:1 as a draft, published it, and forgot about the rest. Today the marketing intern asks him for “a good vertical truffle pizza shot for the Facebook story we’re testing tomorrow.”

Marco opens Growth → Library → Generated images. He scans the grid, finds the 9:16 he generated two weeks ago — the badge in the corner reads 9:16, the date stamp confirms it’s the right batch. He clicks the tile, hits Copy URL, and hands it to the intern. The intern pastes it into Facebook Creator Studio. Total time: 30 seconds. Zero new generations. Zero new credits.

Without the Library, that image was lost the moment Marco closed the Composer tab two weeks ago. With it, he found it as fast as he could navigate to the page.

Worked example — uploading a phone-shot dish video

Marco’s chef Lucia just shot a 12-second clip on her phone: the moment the pici cacio e pepe gets tossed in the pan, finished with cracked pepper, ready for plating. Lucia AirDrops the MP4 to Marco’s laptop.

Marco opens Growth → Composer, picks Friday lunch on the calendar, picks Instagram as the channel. Instead of clicking Generate he clicks Upload your own and picks the MP4. Within a few seconds the clip appears in the Composer preview at 9:16 (the Instagram Reel aspect). Marco writes the caption, schedules the post for Friday at 11:30, hits Save.

The next afternoon Lucia hands Marco a photo from her phone — a tighter shot of the dish plated, much better-lit than the Composer’s AI would have generated. Marco opens Growth → Library → Generated videos to confirm the Friday clip landed (it has a sky-blue Uploaded badge in the corner), then switches to the Generated images tab to bring in the new photo. He opens the Composer for Saturday’s post, hits Upload your own on the image side, picks the photo, writes a different caption. Two posts, both built on Lucia’s actual work, neither one needing the AI to invent food.

Three weeks later Marco scrolls through Generated videos and counts — Lucia has dropped five short clips into the Composer since they started. They all live in the Library, sorted newest-first, hover-to-preview, click-to-watch. The phone has become the studio.

Worked example — swapping in the real dish with a reference

Marco has a lovely generated image: a warm trattoria table, two friends laughing, a plate of tagliatelle al ragù in the foreground. He wants the same scene but with his signature agnolotti del plin in the plate, because that’s the dish he’s promoting this week.

He opens the image in the Library viewer, clicks AI Edit, and under the prompt box clicks Use a dish. He types “agnolotti”, picks it from the list, and its photo appears as a small reference thumbnail. In the prompt box he writes “replace the pasta with this dish” and hits Generate edit. Twenty seconds later a new variant lands at the top of the Library: same table, same laughing friends, same warm light — but now it’s his actual agnolotti del plin in the plate, plated the way his kitchen plates it, not a generic stand-in the AI guessed at.

The original is untouched, the new variant sits beside it, and the whole thing cost the same five-credit edit any prompt costs. Next week, when he’s promoting a different dish, he repeats the trick with that dish’s photo.

What it doesn’t do (yet)

Bulk download — you download one image at a time today. If you need a zip of everything, ask support.
Tag or rename images — the Library shows the date, aspect ratio, and short prompt (if there was one). Free-text tags are a future addition.
Edit a published post — once a post goes live on Instagram or Facebook, the Library shows it as read-only with a small external-link icon (opens the live post on the platform). To change it you’d delete on the platform and create a new post.
Storage quota — there isn’t one. Generate as much as you want; storage is cheap. The system never auto-deletes Library rows.

Composer — where you generate the work that ends up here.
Calendar — the date-grid view of scheduled and published posts (a subset of what the Library shows).
Image editor — manual crop / rotate / colour-adjust edits made there also auto-save into the Library as new entries.
Video generation — every generated video lands in the Generated videos tab automatically.