To make a product-in-hand video from a product photo, do not ask a video model to add the hand. Make a still first in which a hand already holds the product the way you want, check the fingers and the product in that still, then animate it with an image-to-video node and a small motion. On 8frame: Nano Banana 2 for the still (11 credits at 1K, up to 10 references), Kling v3 Standard or Kling 2.6 Pro for the motion (57 or 47 credits for 5 silent seconds). The thing people do not expect is how retakes are billed: a generation that fails is refunded automatically, but a take that runs and comes back with six fingers is a finished generation and costs its full price. So the cheap place to fix hands is the still, where a retake is 11 credits, not the clip.
TL;DR
| Step | Node or model on 8frame | Price in credits | Input |
|---|---|---|---|
| Clean product cut-out | Background removal | 12 per image | Your product photo |
| Hand-and-product still | Nano Banana 2 | 8 / 11 / 17 / 22 per image (0.5K / 1K / 2K / 4K), up to 4 per run | Product photo, a hand photo, a scene |
| Fix a finger or a grip | Flux Edit (FLUX.2 pro) | 7 for about 1 MP | The still |
| Motion, 720p | Kling v3 Standard | 57 for 5 s silent, 86 with sound | Start image (required) |
| Motion, 1080p | Kling 2.6 Pro | 47 for 5 s silent, 95 with sound | Start image |
| Motion ending on a set pose | Kling 2.5 Turbo Pro | 48 for 5 s | Start and end frame |
| Longer single take | Seedance 2.0 | 108 for 5 s at 720p, 4 to 15 s | Start image |
| Join short takes (silent) | Montage | 10 per run | 2 to 10 clips |
Why the still decides the hands
An image-to-video model continues the first frame it is given. If the first frame has a correct hand around a correct product, the model has to move it a little. If the first frame has only the product, the model has to invent a hand, its grip, its fingers and how they cover the label, all while moving. The second job is the one that goes wrong. Our page on the best AI model for hands explains why hands fail and why we do not rank models on hands without measuring; this page is the workflow that keeps the hard part in a still.
The board, node by node
The closest live template is PRODUCT PLACEMENT on 8frame workflows, which places a product into a scene. Open it as a starting board and add the hand steps below.
1. The product, clean
Upload your best product photo: front-on, even light, label readable. If the background is busy, run Background removal (12 credits). The product photo is the reference that must not change; keep it on the board next to every result.
2. The hand reference
Use a photo of a real hand, from a person who agreed to its use, holding something of a similar size and shape in the grip you want: a bottle held at the neck, a jar on an open palm, a phone between thumb and fingers. A real hand reference gives the image model the anatomy instead of a description of it. If you generate a hand instead, attach nothing personal and describe skin tone, nails and sleeve in words.
3. The hand-and-product still
On a Nano Banana 2 node, attach the product photo, the hand reference and, if you have one, a scene reference (your kitchen counter, a bathroom shelf). Prompt the composition, not the motion: "a hand holding [product] at the neck, label facing the camera, fingers below the label, soft window light, shallow depth of field, 9:16". Ask for four images in one run and pick the one where:
- every finger is accounted for and bends the right way;
- the fingers do not cover the label or logo;
- the product's shape, colour and text match the product photo.
If one finger is wrong in an otherwise good still, fix it with Flux Edit (7 credits for about a 1 MP input) rather than starting over.
For a top-down flat lay shot (the product on a surface, a hand entering from the edge), make the still top-down with the hand already in frame. A hand that has to enter during the clip is the hardest version of this shot.
4. Animate with small motion
Connect the approved still to an image-to-video node. Prompt one small, physical action: "the hand tilts the bottle slightly toward the camera", "the fingers rotate the jar a quarter turn", "the camera pushes in slowly". The bigger the motion, the more of the hand the model has to redraw.
Sound: for a product page you usually want none. Kling v3 Standard is 57 credits for 5 silent seconds against 86 with sound; Kling 2.6 Pro is 47 silent against 95. Etsy removes sound anyway (see below).
If the clip must end in a specific pose (the label square to the camera), use a model with an end frame on the canvas: Kling 2.5 Turbo Pro (48 for 5 seconds), Kling O3, Kling O1 Image or Seedance. Kling v3 and Kling 2.6 Pro take no end frame. Make the end frame as a second still from the same references.
5. Review the clip like a buyer
Watch at full size, frame by frame where the hand moves. Check: fingers merging, the product changing shape, the label rewriting itself, the hand passing through the product. Regenerate takes with any of these. Remember the billing rule: only a failed generation is refunded; a finished take you reject is paid.
Which model for which step
| Model | Start image | End frame | Resolution | Lengths | Price |
|---|---|---|---|---|---|
| Kling v3 Standard | required | no | 720p | 3 to 15 s | 57 silent, 86 with sound (5 s) |
| Kling 2.6 Pro | optional | no | 1080p | 5 or 10 s | 47 silent, 95 with sound (5 s) |
| Kling 2.5 Turbo Pro | yes | yes | 1080p | 5 or 10 s | 48 (5 s), 95 (10 s) |
| Kling O3 | required | yes | 720p | 3 to 15 s | 57 silent, 76 with sound (5 s) |
| Seedance 2.0 | yes | yes (no references then) | 480p or 720p | 4 to 15 s | 48 / 108 for 5 s at 480p / 720p |
| Hailuo H3 | yes | no | 2K | 5, 10, 15 s | 88 (5 s) |
We have not measured which of these keeps hands correct most often. Choose by resolution, end frame, length and price, and judge every take yourself.
Where it breaks
- The hand appears mid-clip. Avoid it. Start with the hand in frame.
- Fingers over the label. Fix in the still; the clip will not move them off.
- Big motion. A full turn of the product or a hand that lets go and grabs again gives the model the most to redraw.
- Retakes cost. Failed generations are refunded automatically; finished takes you discard are not. Iterate on stills.
- Resolution. Seedance tops out at 720p and Kling v3 Standard returns 720p. There is no video upscaler on 8frame (Topaz, Clarity and BRIA upscale images), and no video model outputs 4K. For 1080p use Kling 2.6 Pro or Kling 2.5 Turbo Pro.
- Montage drops sound. That does not matter for Etsy, which removes audio, but it does for an ad. Use the Video Editor timeline when you need sound.
Where the video goes
Etsy. Etsy's help page (checked 2026-11-09) says a listing video "may be 3-15 seconds long", will "not contain any audio" once uploaded, should ideally be at least 1080px (minimum 500px), must be under 100 MB, and you can add two videos per listing. So: generate silent, pick 1080p models for the ideal size, and join two 5-second takes with Montage (10 credits) if you want 10 seconds: Montage's lack of sound costs you nothing here.
Shopify. Product media videos can be up to 10 minutes and 1 GB, in .mp4, .mov or .webm. Length is your choice. The Shopify product video guide covers the shapes per placement.
Google Merchant Center. If the clip goes into your product feed, a single 5-second take is too short; the rules are on Google Merchant Center product video.
FAQ
Can AI make a video of a hand holding my product?
Yes, most reliably from a still in which the hand already holds the product. Make and check that still on an image model, then animate it with small motion.
Why do AI hands look wrong in product videos?
Usually because the video model had to invent the hand. Give it a first frame with a correct hand, and keep the motion small. See our page on hands for the reasons.
Do I pay for a video take with bad hands?
Yes. Failed generations are refunded automatically, but a finished take is paid even if you discard it. Fix hands in the still, where a retake is cheaper.
How long should an Etsy product video be?
Etsy's help page says 3 to 15 seconds, and the audio is removed on upload. Two videos per listing are allowed.
Which model can end on a set pose of the product?
One with an end frame on the canvas: Kling 2.5 Turbo Pro, Kling O3, Kling O1 Image or Seedance. Not Kling v3 or Kling 2.6 Pro.
Sources
- Etsy Help Center, How to Add a Listing Video (3 to 15 seconds, no audio after upload, ideal 1080px, minimum 500px, 100 MB, two videos per listing), checked 2026-11-09
- Shopify Help Center, Product media types (video up to 10 minutes, 1 GB, .mp4, .mov or .webm), checked 2026-11-09
- 8frame prices, end-frame support, refunds, defaults and limits: the 8frame app, checked 2026-11-09
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.