Freya turns a Google Sheet of products and their reference photos into clean, standardized shots,
scores every image against your checklist, and files everything in Drive. Keep this panel open
while you work; the app keeps running behind it.
Watch the walkthrough
Recorded on the previous version of the tool; the steps are the same, the screens look different.
1. Your API key
Click the gear icon → Paste API key .
Choose the model provider and paste that provider's key, then Save . Freya checks the key with the provider before saving; a rejected key is never stored.
A green dot on the gear means a key is active in this browser. The key travels with each run you start, is kept encrypted only while that run is active, and is discarded the moment the run finishes.
ChatGPT Chat is the recommended provider: one conversation per product keeps the whole set consistent, and re-flagged images are redone inside that same conversation. Output resolution applies to Gemini only; ChatGPT uses its native size.
2. Start a run
Prepare a Google Sheet with three columns:
FCN Code — the product code; it names the output folder.Drive Link — that product's Drive folder of reference photos.Prompt — what to generate. Say how many images you want with a line like Images: 4; a numbered scene list or one heading per image also works.
Share access: the sheet and every referenced folder must be viewable by the organisation or by anyone with the link. A row that isn't shared fails with a clear reason; the rest of the run continues.On the Google Sheet URL tab paste the link, give the run a batch name and enter your name (both show on the queue and on the Drive folders).
Optional Rules for this run apply to every image, including regenerations. Be specific: name the exact colours, text and details that must not change.
Optional Review checklist : one objective check per line, written for a non-creative intern (“Label text is legible and unchanged”, “Cap colour matches the reference”). Every generated image is scored against it. Leave it empty to skip the review.
Click Generate images . The run joins the Queue ; you can close the tab, it keeps going.
3. The queue
Queue (top right) lists every run by everyone, newest first, with status, progress (done · failed · total),
start time and an approximate finish. Filters: All , Mine (by the name you entered), Today , Needs review , Failed .
Runs by different people are served fairly, so a large batch never blocks a small one. Per run:
Open — the live progress page, or the results once finished.Cancel — waiting products stop at once; the image currently rendering finishes first (up to ~8 min). Nothing already made is lost.Retry failed — puts only the failed products back in the queue. If the run had finished, Freya asks for your key again, because it was discarded when the run ended.Drive — the run's output folder.
4. While it runs
The progress page shows how many products are done, what the workers are doing (the event list), and thumbnails as each product lands in Drive.
Stop & view results finishes the images in progress and shows what is done; Cancel run abandons it.
When the run ends, one message goes to the team's Google Chat space with the counts, the failed products and links to Drive and to the results page.
5. Review
The results page opens with a summary: images passed, needing a look, failed review, pre-flagged. The verdict filters show only those images.
Each product card shows its reference photos (the ones the run actually used) above the generated images. Click a reference to see it next to the output in the viewer.
Click any generated image to open it full size. ← → move between images, C toggles the side-by-side compare, R hides the reference rail, Esc closes.Scores come from the review checklist: 90 and above passes , below 50 fails , in between needs a look . A failing image is regenerated once automatically and the better attempt is kept. Scores are advice, not a gate: Approve or Reject records your decision with your name.
To redo an image, flag it (from the viewer, or click the image's caption on the card). A note box appears: leave it empty to redo the image fresh from the reference photos, or type an instruction to edit the current image. Then click Regenerate flagged . New versions replace the old ones in Drive; every attempt stays visible in the viewer.
Reference photos: naming (optional)
All images in a product's folder are used. To tell Freya what each one is, put one of these words in the file name:
front, back, texture, pet, side or dimension.
Photos without a hint are classified automatically; a photo of a person is treated as a model reference, never as the product.
Open from Drive
Paste a folder of finished outputs (a container of product folders, or a single product folder) to review, approve and tweak them
in the same editor without regenerating everything.
FAQs
Where do the outputs go?
Into the shared Drive: a folder for the day, then batch name_run id , then one folder per product. Re-running a sheet never overwrites an earlier run. Every card and every image links straight to Drive.
A row failed with “no images found” or a permission error.
That product's Drive folder is empty, or it isn't shared. Fix the folder, then Retry failed on the run.
My sheet was rejected entirely.
It's missing one of the three columns (Drive Link, FCN CODE, PROMPT), has no data rows, or isn't shared. The message under the form says which.
“Couldn't tell how many images this prompt wants”
Add a line Images: N to that row's prompt (or a numbered scene list) and submit again. Freya asks rather than guess.
The run stopped with “key ran out of credits” or “key rejected”.
All of your waiting products stop together so nothing is wasted; other people's runs are unaffected. Top up (or save a valid key via the gear icon), then Retry failed .
Retry failed asks for my key although I saved one.
Keys are discarded the moment a run finishes. Retry sends the key saved in this browser again; if the gear shows no green dot, paste it first.
Some images say “not judged”.
The review checklist was set but the reviewer could not score those images (provider hiccup or a key the vision model rejected). The images are fine to use; judge them yourself with Approve / Reject.
Can I stop midway without losing work?
Yes. Stop & view results keeps everything already generated and shows it; Cancel also keeps what was made and just abandons the rest. Finished images are always in Drive.
No message arrived in the Chat space.
Messages are sent only for sheet runs, once, when the run ends (regenerations don't post). If a finished run has no message, tell Divy: the run's event log shows whether Chat refused it.
