From phone photo to catalog-ready image, in 6 steps.

Palmou AI exposes two image operations chained inside a single credit:

1. Intelligent background removal — clean cutout from any background (BRIA RMBG 2.0).
2. AI scene generation on the product image — place the product in a new environment with adaptive lighting (Google Gemini 2.5 Flash Image).

Both run on Replicate, orchestrated via webhooks from a Next.js 16 backend. Below is exactly what happens when you click Generate.

1. 1

   Sign in with Google

   Open palmou.com and click 'Sign up with Google'. No password, no disposable email — Google OAuth keeps bots out and you get 1 free credit on first signup.

   ~10s

2. 2

   Upload one product photo

   Drop a JPG, PNG, or WebP up to 10 MB into the editor. Palmou holds it locally for preview so you can pick a scene before committing a credit.

   ~5s

3. 3

   Pick a scene (optional)

   Choose one of six predefined scenes — White Studio, Premium Beige, Nature, Luxury Marble, Urban, or Gradient. Skip the scene if you only want the background removed.

   ~5s

4. 4

   Click Generate

   Palmou uploads the photo to Supabase Storage in the EU, deducts 1 credit atomically, and triggers the AI pipeline on Replicate.

   ~3s

5. 5

   AI removes the background or composes the scene

   If you skipped the scene, BRIA RMBG 2.0 runs and produces a 256-level-alpha cutout at your source's native resolution. If you picked a scene, Google Gemini 2.5 Flash Image generates a new image at around 1024 px on the longest side with your product placed in the chosen environment.

   ~50s

6. 6

   Download or share

   The final WebP is stored in your gallery. Click Download to save locally, Share to publish to the community feed, or remix it with a different scene for 0.5 credits.

   ~5s

The two models in detail

BRIA RMBG 2.0

Commercial-licensed background remover. Outputs a soft alpha mask with 256 levels of transparency. This matters for any product whose silhouette is not a hard edge: hair, fur, glass, jewelry chains, fabric edges, ice, smoke, foam. Older binary-mask removers leave a halo or chew through fine detail; BRIA RMBG 2.0 keeps it.

Google Gemini 2.5 Flash Image

Image-to-image scene generator. Reads your source product photo and a scene prompt and produces a new image at around 1024 px on the longest side with the product placed in the chosen environment. It preserves product identity (logos, colors, proportions) much better than SDXL inpainting because it doesn't rely on a precomputed mask — the whole image is the conditioning.

File formats and storage

Inputs accepted: JPG, PNG, WebP, up to 10 MB and between 100 and 4096 px on each side. Internally we transcode everything to WebP at quality 92, both on upload and at the final step. Originals and processed files live in your namespace in Supabase Storage (EU region, encrypted at rest). Deleting an image from the gallery removes the row, the original, the processed file, and any intermediate masks in one transaction.

Failure modes and how we handle them

• Content moderation rejection — credit auto-refunded, error logged, user sees a clear message.
• Replicate timeout or 429 rate-limit — Palmou retries with exponential backoff up to 5 times honoring the Retry-After header. If all retries fail, the credit is refunded.
• Storage write failure — credit refunded, user notified, error captured to Sentry for postmortem.
• Webhook signature mismatch— request rejected with 401 before any work happens. Forged callbacks can't inject results.

Public API

Agency plan includes a REST API at /docs/api. Two endpoints: POST /api/v1/images to submit a job and GET /api/v1/images/:id to poll status. Auth via Bearer API key generated at /api-keys; rate-limited to 60 requests per minute per key.