Universal Image Engine

1Dataset / selection

Use incoming selection (0)Use a dataset

Dataset

0 image(s) selected.

Batch name

2Analysis & masking

Use existing analysis where availableRe-analyze all (reanalyze)Regenerate masks (regen_masks)

Visual AIQwen3-VL-8BSegmentationBiRefNetOptionalSAM2

3Routing

Allow L0 (deterministic cleanup)L0 keeps the original background. Off makes L1 the floor, so every photo gets a new backdrop — including genuine label and condition close-ups.

4Output template

What the finished images should look like. Independent of routing: the router still decides how much each image may change. The choice is frozen into the batch, so re-running it later reproduces the same output.

No templates in /config — the saved default will be used.

Background

The product's category, colour and material pick the family and the contrast colour.

CTA button text

Title on the image

Burn each photo's title onto its finished imageUses the title shown on the batch detail page — the operator's edit if there is one, otherwise the feed title. The untitled image is kept too.

5Models

Auto (router picks)

6Generation settings

Resolution

Candidates

Seed

IC-Light

Retry count

Procedural shadowQA enabled

Estimated workload: 0 images × 1 scene(s) × 1 candidate(s) = 0 generation(s).

Cancel

New Batch Job