Correction, 2026-10-02: this page stated as fact results we have not measured: that starting from a still is "the most controllable path", that the video model "inherits" the still's colour, label and composition, that the still makes everything downstream "predictable", that fidelity loss is "the most common failure", that the variants differ only in motion, and that a naming habit saves "30 minutes". Those are now advice. It said Seedance's Auto shape follows the image; Seedance's documentation says Auto lets the model choose the ratio from your inputs, and we have not tested how Seedance reframes a still into a different fixed shape. It also did not say that Kling v3 only animates a start image, that the chain costs assume one try per step, or that 8frame's upscalers work on stills only.
Correction, 2026-10-01: an earlier version of this page described a test run whose prices and settings do not match 8frame. It priced a Kling v3 clip at $0.38 (it is 86 credits with sound, about $0.86), said Seedance delivered 1080p (it tops out at 720p), made vertical and horizontal Kling variants from one still (Kling follows the shape of the start image), and described an export node with 4K output, which 8frame does not have. The walkthrough is replaced by costs from the 8frame tool registry and generation times measured in production.
The image to video workflow chain starts with an approved still, then adds motion to that still. The reason for that order: going straight from text to video leaves the video model to invent everything (composition, product colour, lighting, depth), and its guesses are not your brand. Starting from an approved image, the video model begins from the colour, label and composition you already signed off, and your job narrows to checking that they survive the motion.
TL;DR
- Start from a still, not from text: the still is where you approve the product, the light and the frame
- The chain on 8frame: Nano Banana Pro (master stills) → Seedance 2.0 (hero clip) → Kling v3 (variants) → download
- What it costs: two stills, one 5-second hero clip and five 5-second variants come to 580 credits with sound, about $5.80 at pack rate; 435 with silent variants (one try per step)
- How long it takes: a median of 37 seconds for a Nano Banana Pro still, about 3 minutes for a 720p Seedance 2.0 clip and about 1 minute for a Kling v3 clip, measured on 8frame
- Shape is set by the still: Kling v3 only animates a start image and follows its shape, so a vertical variant needs a vertical still
- Three pitfalls to plan around: fidelity loss at the handoff, motion drift, and scale shifts
Why start from a still
When you prompt a video model with text alone, it has to decide every visual element: the exact shade of the product, where the label sits, how the light falls, what is in the background. It makes reasonable guesses. Reasonable guesses are not your brand.
Image-to-video removes most of those decisions. You supply the approved still with the right product colour, label placement and environment, and the video model's job narrows to putting that still into motion.
This is why the product hero shot pipeline, workflow 1 in the 10 AI video workflows every brand should have, runs Nano Banana Pro into Seedance rather than going text-to-Seedance direct. The intermediate still costs 21 credits and gives you a frame to approve before you pay for motion.
One approved still can also feed several moves: a slow push-in, a rotation and a reveal, all from the same start frame. Each variant changes the motion direction and starts from the same approved product.
The 4-step image to video workflow chain
Step 1. Master stills via Nano Banana Pro
Nano Banana Pro generates the reference image: 21 credits at 1K or 2K, 42 at 4K. On 8frame its median generation time is 37 seconds (p90 77 seconds). Parameters that matter:
- Background: white or gradient studio for product isolation, or a specific environment if the scene is part of the brief.
- Lighting: 3-point product lighting for packshots; match the campaign lighting reference for lifestyle.
- Resolution: 2K is already larger than any frame the video steps return (Seedance tops out at 720p; Kling v3 Standard comes back at 720p). Pay for 4K only if the still is itself a deliverable.
- Aspect ratio: make one still per shape you need. Kling, Wan and Hailuo H3 follow the start image's shape when you start from an image, so a 9:16 variant starts from a 9:16 still. Seedance offers "Auto", which lets the model choose the ratio from your inputs, plus fixed shapes. The full list is in AI video aspect ratios by model.
What you have after step 1: an approved master still for each shape, with your product colour, label and composition decided.
Step 2. Hero clip via Seedance 2.0
Seedance 2.0 takes the still as its reference and animates it. On the canvas it accepts up to 9 reference images, runs 4 to 15 seconds, and generates sound. A 5-second clip is 108 credits at 720p and 48 at 480p; the price is the same with or without references. Its median generation time on 8frame is about 3 minutes for a clip of 6 seconds or less at 720p (197 seconds, p90 413).
Prompts for this step, one motion each:
Slow camera push-in toward the product center. Subtle ambient light.Gentle product rotation, 90 degrees clockwise. Studio white background.Camera orbits left around product at mid-height. Shallow depth of field on foreground.
Keep each prompt to one action and one framing note. The Seedance 2.0 prompts for product demos guide has more examples. Seedance 2.5, the Seedance tool's default, is 157 credits for 5 seconds at 720p; see Seedance 2.5 vs Seedance 2.0 for the differences. If you would rather use Veo for the hero clip, note that from a still Veo always makes an 8-second clip: 168 credits on Veo 3.1 Fast, the setting the Veo tool opens on.
What you have after step 2: a hero clip that starts from the master still. Compare its colour and label with the still before you run the variants.
Step 3. Variants via Kling v3
Once the hero clip clears review, Kling v3 Standard runs the variants: different motion directions from the same stills. Kling v3 only animates a start image (a text-only Kling prompt runs on Kling 2.6 Pro), and the Kling tool opens on v3 Standard with sound. A 5-second clip is 86 credits with sound and 57 silent, and it returns 720p. Its median generation time on 8frame is about 1 minute for a 5-second clip (60 seconds, p90 156). A silent Kling v3 clip costs about half a 720p Seedance 2.0 clip.
From a horizontal and a vertical still you can run, for example:
- a 9:16 push-in for Reels and TikTok (from the vertical still)
- a 16:9 slow rotation for YouTube and paid display (from the horizontal still)
- a close-up with a slow pull-back as a second cut
- a slower move for a premium feel
Every plan can run up to 10 tasks at once, so five variants run side by side rather than one after another.
What you have after step 3: several motion variants, all anchored to the same approved stills.
Step 4. Download and name
8frame does not re-encode video: the file you download is the one the model returned, H.264 in MP4 at 24 frames per second in our measurements. Nothing on the canvas outputs 4K and there is no video upscaler, so the resolution you generated at is the resolution you deliver (see AI video resolution by model). The upscale tool works on stills only: it opens on Topaz (4x, 11 credits per started 24 MP block of output) and also offers Clarity (13, always 2x) and BRIA (6, 2x or 4x).
Name files before they leave the board, for example [project]-[aspect-ratio]-[motion]-[date] (skincare-launch-16x9-push-in-20261001.mp4), so whoever receives them can tell each shape and move apart without opening the files.
What the chain costs on 8frame
Credits from the 8frame tool registry; a credit is about $0.01 at pack rate. These are first-try costs: every retry adds the price of that step again.
| Step | Model and setting | Credits |
|---|---|---|
| Two master stills (16:9 and 9:16) | Nano Banana Pro, 2K, 21 each | 42 |
| Hero clip | Seedance 2.0, 720p, 5 s with sound | 108 |
| Five variants | Kling v3 Standard, 5 s with sound, 86 each | 430 |
| Total | 580 |
With silent variants (57 each) the total is 435. If the client supplies approved product photos, drop the stills: 538 with sound, 393 silent. On the Starter plan's 1,000 monthly credits, one full chain at 580 leaves 420 for retries and drafts; for every model's per-second price, see AI video generator price per second.
Pitfalls
Fidelity loss at the still-to-motion handoff
A failure to watch for. The video model can introduce colour shift, label smear or geometric distortion when it moves the still. Check fine print and tight patterns first.
What to try: attach the reference still at full size and keep small text as large in frame as the shot allows. We have not run a controlled test of which model holds labels best, so try the hero shot on two models before you commit the variants.
Motion direction drift
The model can start moving one way and change course mid-clip: a push-in becomes a pan, a rotation gains a tilt.
What to try: keep the motion prompt to one clear direction and the clip short. If you need a change of direction (push in, hold, pull back), make two clips and cut them together.
Scale shifts
The product can appear to change size during the clip; watch rotations and orbits closely. The model is inferring a 3D object from a flat image, and the inferred geometry may not match the real proportions.
What to try: a still with neutral depth cues. Dramatic perspective, hard shadows and wide-angle distortion are more geometry for the model to guess at; flat studio light is less.
FAQ
Can I use a client's existing product photo instead of generating with Nano Banana Pro?
Yes. Step 1 is about an approved reference still. A real photo carries the actual product, so if one exists, use it; Nano Banana Pro is for when you need to create the still. Make sure you have one in each shape you will animate.
How much does an image-to-video chain cost on 8frame?
Two Nano Banana Pro stills, a 5-second Seedance 2.0 hero clip at 720p and five 5-second Kling v3 variants with sound come to 580 credits, about $5.80 at pack rate; 435 with silent variants. That is one try per step; each retry adds that step's price.
How long does the chain take?
On 8frame, the median Nano Banana Pro still takes 37 seconds, a 720p Seedance 2.0 clip about 3 minutes and a 5-second Kling v3 clip about 1 minute. Variants can run in parallel, up to 10 tasks per account.
Can I make a vertical and a horizontal variant from one still on Kling?
No. When you start from an image, Kling follows the image's shape, so generate a vertical and a horizontal still. Seedance lets you set a fixed shape that differs from the still, but we have not tested how it reframes the image, so check the result before you rely on it.
How do I keep the variants consistent when Kling generates each one independently?
Each Kling generation is independent; what the variants share is the start image. Use the same approved still for every variant, change only the motion prompt, and compare each result with the still before it goes to the client.
Sources
- Prices, lengths, reference limits, required inputs, tool defaults and upscale options: 8frame tool registry, read 2026-10-02
- Aspect ratio behaviour with a start image: 8frame tool configuration, read 2026-09-30
- Seedance's Auto setting ("adaptive" in the API): replicate.com/bytedance/seedance-2.0/api/schema, checked 2026-10-02
- Generation times (median and p90, completed tasks over 120 days) and output resolution and format: 8frame production data, read-only, measured 2026-09-30
- Concurrency (10 active tasks per account on every plan): 8frame queue code, read 2026-09-30
Ready-made boards to start from are on 8frame workflows.
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.