Remove the still background for a pack shot

Cut the pack out of the photo with an image edit, then attach it. Remove the background from a photo is generate_image_from_image with an instruction, not the video keyer. A pack shot has to stay readable — label, shape, colour. A room, a watermark, a second SKU in the frame will move when you I2V. Isolate first. Composite a back you can live with. Then motion.

This is a still cutout. It is not remove-video-background for UGC talking heads, and it is not a one-click eraser that does not exist on this catalog.

Photo cutout is not the video keyer

Versely’s remove_background tool is built for video. Its own description sends stills through generate_image_from_image: an image-editing model, a prompt like “remove the background, transparent backdrop, keep the subject exactly as-is.” There is no separate deterministic photo eraser in the editor the way there is an FFmpeg colorkey for pure-black video mattes.

That honesty matters. Edge quality (hair, condensation, shrink-wrap highlights, a pour mid-stream) will vary by model and by how you word the instruction. Review at 100% before the still becomes frame one of a 70-credit clip.

Do not run the video keyer on a one-frame export of the PNG. You will pay a video path to maybe cut a still, and colorkey will only work if you first put the pack on a pure-black matte — which is the thing you did not have.

Video paths, for contrast, so you do not grab them:

• remove_video_background: FFmpeg colorkey, only solid pure-black or near-black.
• remove_background on a video_url: model path for other video backs.

Neither is this page. This page is a photo of a SKU.

Text-to-image is how you got a generated pack, maybe. Cutout is how you stop generating the kitchen behind it on every I2V.

Approve the pack, then isolate it

Do not cut out a cousin. Approve the still as the listing:

1. The actual SKU. Colourway, cap, label version.
2. Letters you can read paused at phone size. Isolation will not typeset a mush label. If type is illegal, fix or reshoot the still first. Upscale only if the file is small but already sharp.
3. Angle that matches the next verb.
4. Rights. You own the photo or the generate.

Then isolate:

1. Attach image_url to generate_image_from_image.
2. Instruct: remove the background, keep the pack, no new reflections, no extra bottles. Transparent if you will composite. A named studio back if you will not.
3. Pick an edit-capable image model. Ask estimate_cost. Image-edit credits are the model’s sticker, not a video row.
4. Review edges: cap outline, shrink band, drops, shadow. A chewed cap is a different product.
5. Reject hair-like fringing on plastic. Try a tighter prompt or a different model. Do not I2V a halo and hope motion hides it. Motion shows it.

If you needed a white back for retail, say white, not transparent-then-you’ll-flatten-later, unless you truly need alpha. Some pipelines flatten white better than checkerboard.

Do not ask the editor to “make it look like a pack shot.” A pack shot is the readable product. Pretty lifestyle with a melted label is not one.

Wan 2.7 Pro Edit is 8 credits image-edit on this catalog if that row is the one you pick; confirm the live card. Do not invent a photo-eraser slug.

Composite before I2V

A transparent pack on checkerboard is not a video parent. I2V will invent a room in the transparent holes, or treat checkerboard as texture. Composite a back you accept:

• Solid brand colour.
• A table you photographed.
• A simple gradient you will not have to explain.

Then crop to the ship ratio. 9:16 if the clip is a Reel or TikTok. I2V follows the still. A cutout sitting in a 16:9 canvas becomes a landscape generate.

Prompt I2V for what moves around the pack: a slow push, a hand that does not cover the letters, a turn that keeps the label toward camera. Do not re-describe the box. The still is the box.

Kling O3 Pro image-to-video is 70 credits, native audio, 3–15s. Wan 2.7 image-to-video is 15 credits, 1080p, audio null. Silent plus VO if the line is a contract.

If I2V warps the logo, cut back to the still. Do not “fix the SKU” in a second motion prompt. Isolation was supposed to reduce the model’s urge to invent a kitchen; it cannot enforce a still contract the row then breaks. Shorter duration leaks less.

Hands: a cutout pack plus a generated hand is how you get extra fingers. If the verb needs hands, prefer a still that already has legal hands, or live-action. Isolation of a bottle does not make a hand model careful.

Shadow is part of the pack, not part of the kitchen. A hard cutout that kills the contact shadow makes the bottle float on whatever back you composite next; I2V will then invent a second shadow that fights the first. If the listing still needs a table, keep a short, honest shadow in the image edit, or composite the pack onto a photographed table before you animate. Do not ask the video row to “add a natural shadow” on a checkerboard parent.

Batch SKUs the same way: isolate, review, composite, then I2V. Do not cut out twelve bottles, skip the edge pass, and generate twelve clips. The cheap step is the still. The expensive step is motion on a chewed cap.

Caption never belongs on checkerboard

Type after the composite and after the motion keeper, not on the alpha still.

Add captions to a video transcribes speech on the clip. A silent pack needs overlay copy you wrote — flavour, price, “new” — after picture lock. Add a text overlay is one authored line, whole clip. Do not generate the claim into the cutout. You will recrop it.

Burned-in captions on a checkerboard preview will sit on squares that will not exist in the ad. Composite, I2V, then type.

Safe area still applies: do not park a flavour line on the cap and also in TikTok’s username stack.

Article 50 (2 August 2026) still applies if the photoreal pack-plus-motion would pass as a real commercial at first exposure. A cutout does not make it “photography only” when the motion is generated. YouTube Shorts overlay on photoreal AI (27 May 2026) is more chrome. Leave room.

NY GBL § 396-b is about synthetic performers that appear human. A bottle is not that object. Do not file a pack cutout under that statute.

Isolate the pack still in Versely with an image edit, composite a back, then I2V. Do not send the PNG to the video keyer.

FAQ

Is there a one-click photo background eraser?

Not a separate deterministic one. Stills go through generate_image_from_image. Video has the keyer and the video model path. Review edges before you animate.

Can I I2V the transparent cutout directly?

You can. You will often get a invented room in the holes, or checkerboard texture. Composite a back you accept, then attach.

Is this the same as removing a talking-head’s background?

No. That is the video pipeline (colorkey on black, or a video model). This page is a still of a SKU.