Product Photography

Make catalog-, ad-, or hero-ready product images from a plain photo or a written brief. Each stage uses a different model, so route by stage, not by one favorite model.

Inputs to collect

• Product image, if the user has one: a public https URL or a TokenLab result URL. Skip for a from-scratch brief.
• Look: studio (seamless or white sweep), lifestyle (a real setting), or hero (dramatic, ad-grade).
• Background result: keep, cut out, or replace with a named scene.
• Format: aspect ratio per channel (1:1 marketplace, 4:5 social feed, 9:16 story, 16:9 banner) and image count.
• Hard rules: product shape, colors, and on-pack text must not change. Add no logos, awards, or claims the user did not supply.

Models

<!-- tokenlab:models:start --> <!-- Generated from the live catalog by scripts/sync_models.py. Do not edit; change preferredModels in catalog.json. -->

Operation	Model	Pick	Use
text-to-image	flux-2-pro	default	Shot from a written brief
text-to-image	seedream-5.0	alternative	Shot from a brief, alternative look (not live-tested)
text-to-image	recraft-v4.1-utility	alternative	Flat catalog mockup (not live-tested)
text-to-image	p-image	alternative	Low-cost draft
image-to-image	nano-banana-2	default	Restage an existing product photo
image-to-image	seedream-5.0	alternative	Restage, alternative look (not live-tested)
image-to-image	nano-banana-2-lite	alternative	Low-cost restage (not live-tested)
image-edit	bria-rmbg-v2-0	default	Cut out the background
image-edit	ideogram-replace-background-v3-turbo	alternative	Replace the background in place (not live-tested)
image-edit	ideogram-replace-background-v3	alternative	Replace the background, higher tier (not live-tested)

This list refreshes daily from the live catalog. Preferred models come first; otherwise models are ranked by their 24-hour success rate. Before a call, confirm the model with GET /v1/models/{model} and read its price from GET /v1/models/{model}/pricing.

<!-- tokenlab:models:end -->

How to choose, by stage:

• Shot from a brief (text-to-image): FLUX.2 or Seedream for a photographic shot, a Recraft V4.1 utility model for a flat catalog mockup, a P-Image model for a low-cost draft.
• Restage a product photo (image-to-image): Nano Banana or Seedream. The lite tier is the low-cost option.
• Cut out (image-edit): Bria background removal. It needs only the image.
• Replace the background in place (image-edit): Ideogram replace-background. The turbo tier is faster; the full tier is the higher tier.

Exact fields come from each model's contract.

Workflow

Run each call with tokenlab-run (contract → estimate → call → poll → result). Commands below use its helper, ../tokenlab-run/scripts/tokenlab_run.py, a path relative to this skill's folder. If you installed this skill without tokenlab-run, send the same requests with curl. --wait stops after 100 s; a longer task keeps running, so continue with the resume command it prints.

1. Read the contract for each model you plan to use. Note the endpoint for the operation, the size fields (aspect_ratio + resolution, size, or width / height), and the image input field (image_url, image_urls, or image).
2. Stage 1: a clean subject. If the user has a good product photo, keep it. Otherwise generate one with a text-to-image model. Brief it like a photographer (see Technique). Image responses can be sync or async; poll GET /v1/tasks/{id} when status is pending.
3. Stage 2: the scene.
   • Studio on white or transparent: send the subject to the Bria background removal model with operation: "image-edit" and image_url.
   • New background, same product: send the subject URL and a scene prompt to an Ideogram replace-background model, or restage it with a Nano Banana model image-to-image.
4. Stage 3: variants. Re-run only the weak stage, one aspect ratio per call. Keep the stage-1 URL as the anchor so the product stays the same.
5. Check the result against the Quality bar. Return the result URLs, the model per stage, and the task IDs.

Technique

• Brief order: product, surface, light, lens or angle, background, mood. Example: "matte black wireless earbuds case on raw concrete, one soft key light from upper left, 45-degree top-down view, plain warm grey background, premium tech look."
• Light is the main lever. State direction and quality: "large softbox from the left, white bounce card on the right", "low warm sun from behind", "hard single spotlight from above".
• Lens words change depth. "100mm macro, shallow depth of field" for texture close-ups; "35mm, eye level" for lifestyle context.
• Restage, do not regenerate. When the product is right but the scene is wrong, use image-to-image or background replacement with the product photo as input. A fresh generation can change the product.
• Protect the product in the prompt. In image-to-image, say what must stay: "keep the bottle shape, label, and cap exactly as in the reference; change only the background to ..."
• Ground the product. A missing or wrong-direction contact shadow is the usual tell of a pasted product. Ask for "soft contact shadow under the product, light from the left" when you restage.
• Copy on pack: do not ask a general model to write label text. Keep the original label from the photo, or route to tokenlab-text-in-image.

Quality bar

• Product shape, color, proportions, and label match the input.
• Light direction is consistent between product, shadow, and scene.
• Clean edges after a cutout: no halo, no lost thin parts (straps, cables, glass rims).
• Correct aspect ratio for each channel, and no invented text, logos, or badges.

Related skills

• tokenlab-run: the call, poll, and result mechanics.
• tokenlab-edit-image: one targeted change to an existing photo.
• tokenlab-text-in-image: packaging and poster copy.
• tokenlab-product-ad-video: turn the final still into an ad clip.
• tokenlab-prompting: prompt craft per model family.