Can I run this from Claude instead of the canvas?

Yes. The connector's create_workflow, wire_workflow and estimate_workflow tools build the same graph, and run_workflow starts it after you have seen the price.

Why not go straight from the photo to video?

You can. The still in between gives you a controlled first frame — background, lighting and crop fixed — so the video model spends its capacity on motion instead of on cleaning up the plate.

Nano Banana Pro edit from 175 credits per image and Seedance 2.5 image-to-video from 3,030 credits for five seconds at the base resolution; the workflow's Estimate shows the total for your settings before anything runs.

From one product photo to a five-second video ad — one LUVI workflow, with sound

A phone photo of a bottle becomes a clean studio still with Nano Banana Pro, then a five-second clip with motion and sound on Seedance 2.5 — wired as one workflow, estimated before it runs, from about 3,200 credits.

This is the smallest workflow that produces something you could post: a product photograph in, a five-second clip with sound out, two models, one graph. It runs on the LUVI canvas or from Claude through the connector.

What you need

One photo of the product. A phone shot on a table is fine; the first model will fix the setting.
About 3,200 credits at the base resolution.

The graph

Three nodes carry the work:

Reference — the photo.
Nano Banana Pro · edit — turns the photo into a studio still.
Seedance 2.5 · image-to-video — animates the still and generates the sound.

Wire the reference into the edit node's image slot, the edit node's output into the video node's image slot, and both outputs into an Output node so you keep the still as well as the clip. A wrong port or an empty required slot shows as a red edge when you draw it — nothing is spent on a mis-wired graph.

Node 2: the still

Nano Banana is an edit model, so the prompt is an instruction with a preservation clause, not a description of a new picture:

Place the bottle from the image on a matte charcoal surface against a soft graphite gradient, product-photography lighting from the upper left with a gentle rim light on the right edge, shallow depth of field, the label sharp and legible. Keep the bottle, its label and its proportions exactly as in the image.

Output at 2K. From 175 credits.

Node 3: the clip

Seedance 2.5 takes the still as its first frame, so the prompt spends its words on motion and sound — not on re-describing the bottle. The guide's structure, in miniature:

Summary: a slow product reveal of the bottle on a charcoal surface, studio lighting, premium and calm. 0–2 seconds: the camera holds a medium close-up as a fine mist of condensation forms on the glass and a single droplet begins to run down the label. 2–5 seconds: the camera pushes in slowly toward the label; the rim light brightens by a hair as the droplet reaches the base. （soft, low synth pad, unhurried） <a faint hiss of condensation, one droplet tapping the surface at the 4-second mark> No subtitles.

One camera move, integer-second beats, typed audio brackets, no visual negations. Five seconds at the base resolution: from 3,030 credits.

How do I run it?

Press Estimate: the graph's total appears before anything runs. Press Run. The still lands first, then the clip; both are in the Library under the project, tagged to the workflow. A step that fails stops the chain and is not charged.

Variations that cost nothing to try

Swap node 3 for Wan 3.0 image-to-video (from 2,000 credits) when you want a longer take.
Add a Batch node in front of the video step to render three prompt variants from the same still and pick one.
Feed the clip into a lip-sync model if the ad needs a spoken line.

Can Claude build it for me?

"Build me a workflow: this photo → Nano Banana Pro edit → Seedance 2.5 image-to-video, estimate it" is enough — see how to connect LUVI to Claude and what a workflow is. Claude calls create_workflow, get_workflow_ports (the port ids are the models' real parameter keys), wire_workflow and estimate_workflow, then hands you the link to review and press Run.