WeChat shops: the pack still is the SKU across languages
Identity hold first. Dub later.
Identity hold first. Dub later. A WeChat shop listing does not become a new product because the caption is Mandarin in one cut and English in another. The SKU is the pack: label, colourway, proportion, the face of the bottle the buyer will recognise in the feed. Motion has to copy that still. Language is a second pass on a file that already held.
Image-to-video is the tool shape. Frame one is the file you sent. The prompt describes camera and physics, not a second bottle. Seedance 2.5 Image to Video is the catalog row built for that contract: one still into a continuous clip, durations from four seconds to thirty, output at 480p or 720p, native audio with the picture, and an optional second image to set the closing frame. requires_image is true. No pack photo, no clip. The still is the contract is the longer router; this page is only the shop.
Do not regenerate the bottle per locale
Text-to-video re-decides the product every take. That is how a gold cap becomes a silver cap in the Cantonese cut. Image-to-video does not get a vote on the label. If the still is wrong, every second is waste — up to thirty seconds of the misspelling, billed on a resolution-based per-second matrix (listed display 47 credits; the matrix also quotes 21 for SD and 47 for the HD/4K option, with a published range of 83 to 1392 credits across length).
A last frame is control, not a default. Use it when the shot must land on the back of pack or on a mark you already approved. A single hero still with a short orbit is the usual listing loop.
720p is the ceiling on this endpoint. Feed and Moments cuts live here. A 2K plate you intend to grade is a different row. Identity is still the I2V choice either way.
Dub is a pipeline, not a prompt adjective
Once the pack holds, dub the video. dub_video translates, clones delivery, and optionally resyncs. It is not "write the Mandarin into the Seedance prompt and hope the bottle stays." Native audio on the I2V row is atmosphere generated with the picture. It is not a localisation desk.
Two engines, one async job. elevenlabs (default) works on audio or video, supports trimming a range, and handles clips up to 30 minutes. heygen is video-only, adds lip-sync so the mouth matches the new language, caps at 8 minutes, and supports fewer languages. The source must already be Versely-hosted — a prior generate, upload, or output. External links are not a dub source. The tool returns a project_id; poll get_dub_status. Cost scales with length and engine; there is no honest flat number in the spec. Ask for estimate_cost on a long cut.
If you only needed the spoken content translated and not a cloned mouth, that is translate_audio_only on the same tool — the lighter job at translate a video. A pack orbit with no presenter does not need HeyGen lip-sync. A talking founder who already exists as footage might.
WeChat does not change the order of operations
The shop can be Moments, a Mini Program, or a live room. The file is still: lock the SKU as a still, animate that still, then dub. Regenerating a "Chinese version" of the product as a new text-to-video take is how you pay twice and ship two SKUs. Identity hold first. Dub later.
FAQ
Can I prompt Seedance in Chinese and skip dubbing?
You can write a motion prompt in any language. That does not localise a talking track you never recorded, and it does not replace dub_video when the read has to change. The still is still the SKU. Language on the label is in the photograph, not in the prompt.
Do I need lip-sync for a pack turn?
No. Lip-sync is the HeyGen engine on a face. A bottle does not have a mouth. Use the default audio dub, or skip dubbing entirely if the clip is silent motion plus on-pack type.
Why not make a new generate per market?
Because T2V re-decides the pack. The expensive failure is a drifted label, not a missed idiom. Hold the still. Dub the file you already approved.
Is 30 seconds the listing length I should buy?
Thirty seconds is the ceiling on Seedance 2.5 I2V, not a compliment. Prove the identity on a length you would actually ship. Climb only when the move is actually that long. Stitching shorter T2V clips around a product is how the label drifts; one pass from a locked frame is how it does not.