On-brand image studio with a searchable approved library

Build a brand image studio that my marketing team opens whenever they need a new visual. It uses Black Forest Labs to generate and retouch images, Google Drive to file approved assets, and Slack to announce them. The default view is the library, not the generator, because finding an image the team already approved is the most common reason to open the app.

The library view lists every approved image as a thumbnail grid with a search box and filters for brand, campaign and creator. Each card shows the image, brand, campaign, creator and approval date. Opening a card shows the full size preview, the exact prompt text that produced it, the brand preset used including its palette and style notes, a link to the file in Google Drive, and a Rerun variation button that opens the create view prefilled with that prompt and preset so someone can change a few words instead of starting over.

Brand presets get their own management view. A preset has a name, a brand, a color palette, free text style notes, and reference images uploaded by the user. The FLUX.2 [pro] action accepts a maximum of eight reference images, so enforce that cap in the UI. Presets persist per workspace and are shared by everyone in it.

In the create view the user picks a preset, optionally tags a campaign, and describes the shot they want in plain language. Generating calls the Black Forest Labs Generate or Edit Image action on FLUX.2 [pro], sending the user's description combined with the preset's palette and style notes as the prompt, and attaching the preset's reference images so the output stays on brand. Show generation progress, and keep the composed prompt visible so the user can see what was actually sent.

A touch up panel works on the result before approval. Prompt based fixes such as "make the jacket navy" or "soften the shadow on the left" call Edit Image with FLUX.1 Kontext [pro]. Removing something calls Erase Object from Image (FLUX Tools), which requires a mask, so give the user a brush to paint over the area to erase and send that painted area as the mask. Each touch up creates a new version of the draft, and the user can step back to an earlier version.

Nothing enters the library until a reviewer approves it. Generated images sit in a pending queue showing the prompt, preset, creator and version history. Approving does three things: find or create that brand's folder in Google Drive (use List Files with a Drive query to look for a folder by name, and Create Folder only when there is no match), upload the image into it with Upload File (Multipart), and post the preview along with the brand, campaign and prompt to a Slack channel using Send a Message. Only then does the item appear in the library. Rejecting keeps it out of the library and attaches the reviewer's note. Record who approved or rejected each item and when. If the Slack preview should link to the Drive file for people who cannot already see that folder, use Create Permission to share it.

Brand presets, campaigns, drafts, approval status and full generation history persist per workspace. Two Black Forest Labs details are load bearing. First, every generation is asynchronous: the handler submits the task and then polls Get Result until the status is Ready, handling the Pending, Error, Request Moderated and Content Moderated states. Second, the returned delivery URL expires about ten minutes after generation, so the handler must download the image bytes immediately and persist them in app storage rather than saving the signed URL anywhere. Approved uploads to Drive come from those stored bytes. Google Drive multipart upload is capped at 5MB per file, so generate in a web friendly format such as JPEG or WebP to stay under the limit.