Correction, 2026-10-02: this page stated untested behaviour as fact: that Seedance 2.0 is the right model when motion must stay physically grounded, that references reduce drift, that a background reference controls the light, that long prompts stiffen the motion, and that Seedance holds up better than Kling v3 under close inspection. These are now advice. It also called Kling v3 the cheaper option, but at 480p Seedance 2.0 (48 credits for 5 seconds) costs less than Kling v3 Standard without sound (57), and it did not say that Kling v3 needs a start image. It left out Seedance's lengths (4 to 15 seconds on 2.0, 4 to 30 on 2.5) and the canvas input rules (a start image or reference images, not both; an end frame only with a start image and no reference images), and its note on close-up faces now says it comes from our August 2026 tests.
Product demos need the model to show your screen or your product, not one it invents. With Seedance you attach the real screenshot or product photo as a reference image, describe one action around it, and check the result against the reference label by label, because a match is not guaranteed. Here are 8 prompts with what to check in each result. On 8frame, Seedance 2.0 costs 48 credits for 5 seconds at 480p and 108 at 720p; Seedance 2.5, the version the node opens on, costs 70 and 157 and runs up to 30 seconds.
Correction, 2026-09-30: the first version of this page presented these prompts as run on the canvas, with generation times and results we cannot reproduce, and said Seedance outputs 1080p at 30fps (on 8frame it runs at 480p or 720p). The prompts and the formula stay; the rest is gone.
TL;DR
- Give Seedance your real screenshot or product photo as a reference image; the canvas takes up to 9
- Formula: product or screen reference + environment reference + action + framing cue; set the shape on the node
- On 8frame: Seedance 2.0 is 48 / 108 credits for 5 s (480p / 720p) and runs 4 to 15 s; Seedance 2.5, the default, is 70 / 157 and runs 4 to 30 s; audio is on by default at the same price
- A start image or reference images, not both; an end frame works only with a start image and no reference images
- Kling v3 Standard (57 credits for 5 s at 720p without sound) needs a start image; Seedance takes several reference images and, on 2.5, runs past 15 seconds
When to use Seedance 2.0 for product demos
Without a reference, a video model has no way to know your UI, so it makes one up: plausible labels and layouts that aren't yours. Attaching the actual screenshot gives the model your screen to work from. Check every label in the result; a reference is something to follow, not a guarantee of pixel accuracy.
For physical products, attach a clean product photo (front label, plain background). Seedance takes up to 9 reference images on the canvas, so the product, the setting and a second angle can go in together. Kling v3 Standard is the other common pick: 57 credits for 5 seconds at 720p without sound, 86 with sound (the setting the Kling node opens on), and it needs a start image to animate from. At 480p, Seedance 2.0 is the cheaper of the two (48); at 720p, Kling v3 is (Seedance 2.0 is 108 there, Seedance 2.5 157).
If a person appears in a reference image, use a waist-up or full-length shot. In our tests in August 2026, Seedance's own filter rejected reference images with a close-up face on every run, on both 2.0 and 2.5, with an E005 "flagged as sensitive" error; waist-up and full-length shots of the same person passed. Details in Seedance flagged as sensitive (E005).
The prompt formula
Reference image (UI screenshot or product) + Reference image (environment) + Action + Framing cue
For SaaS demos, attach the actual screen screenshot. For physical demos, attach the product photo. Second reference: a background for the scene, such as a desk or an office. Keep prompts short: one action and one framing cue. With fewer instructions it is easier to see which one a take ignored, and to fix just that one.
If you want your screenshot as the exact first frame rather than a look to follow, connect it as the start image instead. On the canvas a start image and reference images are either/or: if both are attached, the reference images win and the start image is left out. An end frame works only with a start image and no reference images.
8 prompts for product demo formats
1. Feature highlight close-up
Prompt: Reference: [dashboard screenshot showing analytics chart], [dark desk surface photo]. Camera slowly pushes into a laptop screen displaying a bar chart with the analytics section visible. Cursor hovers over one bar, tooltip appears. Subtle ambient office light. Slightly shallow depth of field.
What to check: Whether the chart stays readable through the push-in and the tooltip lands on the element from your screenshot. One element and one action keep the check simple.
2. Use-case scenario in context
Prompt: Reference: [project management app screenshot], [coworking space background photo]. A person sits at a table in a busy open office, types briefly, then leans back as the project board on their screen updates with new cards. Natural ambient noise implied by the setting. Handheld feel.
What to check: That the screen content stays your screen while the person and room move. If the board update doesn't show, make it a separate shot.
3. Side-by-side comparison
Prompt: Reference: [before-state UI screenshot], [after-state UI screenshot]. A split-screen view: the left half shows the old cluttered interface, the right half shows the new clean version. A thin animated divider line sweeps from left to right revealing the improvement. Clean white background.
What to check: That both screens stay true to their references and the divider sweep is smooth. With two screens in one frame, compare each half with its own screenshot.
4. UI screen-record style overlay
Prompt: Reference: [app UI screenshot], [laptop on light wood desk photo]. Camera sits at desk level looking slightly up at a laptop screen. The interface is clearly visible, stable, and occupies most of the frame. Mouse cursor moves across the screen to a button in the top right. Cursor clicks.
What to check: Cursor path, the click, and whether UI text stays legible at the upward angle.
5. Onboarding step sequence
Prompt: Reference: [sign-up flow screenshot, step 1], [neutral light background photo]. A single screen fills the frame. A hand moves into shot from the right side holding a stylus, taps the first input field, and the field activates. Progress indicator at the top of the screen shows step 1 of 4.
What to check: The progress indicator and the moment of contact between stylus and screen. One step per clip, then edit them together.
6. Problem-solution demonstration
Prompt: Reference: [cluttered spreadsheet screenshot], [modern app UI screenshot]. Two sequential scenes separated by a cut: first, a hand scrolls through a long, messy spreadsheet on a laptop. Cut to the same data displayed cleanly in a dashboard view. Both scenes use the same desk environment.
What to check: That props and light match across the implied cut. If the cut doesn't happen, generate the two scenes separately.
7. Result reveal (after-state)
Prompt: Reference: [finished report or output screenshot], [home office desk photo]. A laptop screen transitions from a loading state (spinning indicator) to a fully populated results page. The results page matches the reference image exactly. Camera stays static. Brief desk ambient sound implied.
What to check: That the results screen matches your reference without extra elements. A static camera keeps attention on the reveal. If the results page drifts from your screenshot, try the frame route instead: the loading screen as the start image, the finished page as the end frame, and no reference images.
8. Hands-on first-person demo
Prompt: Reference: [physical hardware product photo, e.g. a router or IoT device], [white desk photo]. First-person perspective, looking down at hands unboxing a compact device, plugging in a cable, then pressing the power button. LED light on device blinks on. No face visible. Slightly warm overhead light.
What to check: That the device keeps the shape of your product photo, the cable goes in cleanly, and the LED comes on after the press.
Common failures
UI text drift. Without a screen reference, the model has to make the UI up. Attach the actual screenshot, and still check every label.
Over-long prompts. A long instruction list makes it hard to tell which instruction a take ignored. Cut to action and framing.
Multi-step in one clip. Several UI states in one clip means several things to check in every take. One action per clip, then edit the clips together.
Wrong shape. Desktop demos: 16:9. Mobile app demos: 9:16. Set it on the node before generating rather than cropping later.
E005 "flagged as sensitive". If a reference image shows a face in close-up, reframe it to waist-up or full-length. Retrying the same image does not help.
Step-by-step on 8frame
-
Open the 8frame canvas, add a Seedance node and connect your UI screenshot or product photo as a reference image. The node opens on Seedance 2.5, 5 seconds, 720p (157 credits); 2.0 is in the same picker.
-
Add an environment reference: desk photo, office background, or clean surface, for the setting and light you want.
-
Write a short prompt: one action, one framing cue.
-
Set 16:9 for desktop demos or 9:16 for mobile, the resolution (480p or 720p) and the length: 4 to 15 seconds on 2.0, 4 to 30 on 2.5, billed per second.
-
Run a 480p draft first, then the keeper at 720p. There is no video upscaling on 8frame, so render the keeper at the resolution you deliver.
-
Review the clip against your screenshot, label by label. If a generation fails, its credits come back automatically.
FAQ
Why attach a screenshot for a SaaS demo?
Without it, a video model generates a UI that resembles a product, not your product. With it, the model has your real screen to work from. Still check labels in the result.
How much does a Seedance product demo clip cost?
On 8frame, Seedance 2.0 is 48 credits for 5 seconds at 480p and 108 at 720p; Seedance 2.5 is 70 and 157. Both are billed per second with audio included, so a 10-second take at 720p is 216 on 2.0 and 313 on 2.5.
How many reference images can I attach?
Up to 9 on the canvas. A start image and reference images are either/or, and an end frame works only with a start image and no reference images.
When should I use Kling v3 instead of Seedance?
When you have one start image per shot and want 720p variants for less: Kling v3 Standard is 57 credits for 5 seconds without sound and 86 with sound. When the shot needs several reference images, or runs longer than 15 seconds (Seedance 2.5 goes to 30), use Seedance.
Why does Seedance reject my reference with E005?
If the error is E005 ("flagged as sensitive"), check for a close-up face in a reference image: in our August 2026 tests those were rejected on both 2.0 and 2.5, while waist-up and full-length shots of the same person passed. Reframe the reference to waist-up or full-length. Retrying the same image does not help.
For physical product demos in an ecommerce context, how to make a Shopify product video with AI covers the full workflow.
Sources
- 8frame Seedance prices (per second, by length), lengths, resolutions, node defaults and canvas inputs (up to 9 reference images; a start image or reference images; an end frame only with a start image and no reference images), and Kling v3 Standard prices, node default and start-image requirement: the 8frame tool registry, read 2026-10-02
- Close-up face references rejected with E005, waist-up and full-length passed: 8frame tests on Seedance 2.0 and 2.5, August 2026
Your real screen in, one action per clip.
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.