Start a new job
Recolor one named (or auto-detected) garment consistently across a set of
images. Each input image yields num_variations outputs. The recolor target
is a reference pattern/colour image, a colour/text instruction, or both (at
least one required). Only the chosen garment is changed; everything else in
the image is preserved, and the recolor stays consistent across the whole
set. The render tier (1K/2K/4K) bills per output by tier.
token_authAuthorizationBearer <token>Token-based authentication using access tokens.
application/json- body
ai_retouch?Ai Retouchauto: once the AI reviewer has checked the outputs, the AI
retoucher fixes what it flagged on each image, once, without being asked.
Each fixed image costs the AI retouch price, charged only when an image is
fixed. Enterprise only for now, and not with review_config (images in
a review round are retouched from the review screen); otherwise ignored.
quality_summary.settled on GET /jobs//results, or the
job_quality notification, tells when it is done.
"off""off""auto"project_id*Project IdProject key where images are stored.
images*array<string>UUIDs of uploaded images, each containing the garment to recolor.
1 <= items <= 10garment_label?|Optional garment to recolor, e.g. 't-shirt', 'jacket', 'hat'. Name up to two garments to recolor together, separated by a comma, '+', '&' or 'and' (e.g. 'top, shorts' for a bikini or an activewear set). When empty, the most-probable clothing item is detected automatically.
nullreference_image?|Optional UUID of ONE pattern/colour reference image to recolor toward.
nullreference_text?|Optional textual colour/pattern reference, e.g. a hex code '#2E5A3B' or a colour name. An alternative to reference_image when the user has no swatch.
nullinstruction?|Optional free-form recolor instruction, e.g. 'make it forest green'. At least one of reference_image / reference_text / instruction must be provided.
nullnum_variations?Num VariationsVariations per input image (1-8). Output shape is len(images) x num_variations.
1model?ModelGeneration engine (auto | nano_banana_2 | nano_banana_pro | seedream | gpt_image).
"auto"processing_size?|Processing quality: low | medium | high, or the px tier 1024 | 2048 | 4096 (default 2048 = medium). The size the garment area is generated at before it is fitted back into the image; outputs keep the input's dimensions. Drives credit cost (3 / 5 / 10 credits per output); higher quality takes longer.
2048format?FormatOutput file format (png | jpg). Default png: lossless, so pixels outside the recolored garment stay byte-identical to the input. jpg gives smaller files.
"png""png""jpg"post_process?Post ProcessEnable a finishing pass that keeps the recolored garment tones consistent across the set.
trueuse_anchor?|Keep the recolored garment visually consistent across the whole set. Only applies when the job produces more than one output. Defaults on (None -> enabled) to preserve prior behaviour.
nulladd_ai_watermark?Add Ai WatermarkBake an 'AI-generated' disclosure mark into each output (irreversible).
falsereview_config?|Optional review configuration. If provided, creates a review pipeline for this job. Fields: requires_review (bool), default_reviewer_type (str), default_retouch_method (str), auto_approve_threshold (float).
nullpost_process_only?Post Process OnlySkip generation and only shift the detected garment's colour to
target_hex with image processing (admin only; ignored for other users).
falsetarget_hex?|Target colour as '#RRGGBB'. Required when post_process_only is set.
nullSuccessful Response
application/json- response
job_id*Job IdThe ID of the created garment recolor job.
status*StatusThe initial status of the job.
message*MessageA message indicating the job creation status.
total_outputs*Total OutputsNumber of outputs the job will produce (images x variations).
curl -X POST "https://example.com/garment-recolor" \ -H "Content-Type: application/json" \ -d '{ "project_id": "string", "images": [ "string" ] }'{ "job_id": "string", "status": "string", "message": "string", "total_outputs": 0}