How can I use Seedance from Claude or Cursor?

Install the FLORA MCP in Claude or Cursor, then ask the agent to animate an image with Seedance 2.5. The agent sends the image and your motion prompt to FLORA's Seedance 2.5 image-to-video model, waits for the clip and returns a link. A 5-second 720p clip cost $2.46 and took under 5 minutes in our run.

At a glance

What you need

• Claude (web, Desktop or Claude Code) or Cursor with the FLORA MCP installed. Claude setup is in the Claude install guide.
• For Cursor, the FLORA server in your MCP config. Setup is in the Cursor install guide.
• A FLORA account on a workspace with video generation.
• A starting image: one you made in FLORA, a public link, or (in Cursor or Claude Code) a file on your computer.

Key features

• One connector, every video model: Seedance 2.5, Kling and other video models sit behind the same FLORA MCP.
• Quotes before runs: Ask for a quote and FLORA returns the price in dollars with nothing reserved.
• Local files in Cursor and Claude Code: The agent uploads an image from your computer before animating it.
• Batches: Up to 20 clips in one call, checked together when they finish.
• Canvas history: Each clip is saved next to the still it came from, ready for another take.

Best for

• Product loops and short social clips built from an approved still
• Developers who want video generation inside the editor they already use
• Animating a whole folder of stills in one Cursor task
• Testing motion ideas before a shoot or an edit

Step by step

1. Step 1: Connect FLORA to your agent

   In Claude, add FLORA as a custom connector with the URL https://agents.flora.ai/mcp and sign in. In Cursor, add the server to ~/.cursor/mcp.json or the project's .cursor/mcp.json. Cursor opens a browser to sign you in the first time it calls a FLORA tool.

   {
     "mcpServers": {
       "flora": {
         "type": "http",
         "url": "https://agents.flora.ai/mcp"
       }
     }
   }

2. Step 2: Start from a still you like

   The Seedance 2.5 model in this guide is image-to-video: it animates a frame you give it. Get the still right first, since the clip inherits its subject, light and composition. You can make the still in the same chat with an image model and then animate it.

3. Step 3: Ask for the motion, not the scene

   The image already defines what is in the frame. Your prompt should say what moves, how fast, and what the camera does. Name Seedance 2.5 so the agent does not pick a different video model, and give the duration and resolution.

   Animate this image with Seedance 2.5 through FLORA. 5 seconds, 720p, keep the original framing. Locked-off shot, no camera move: coffee drips from the dripper into the carafe, a thin wisp of steam rises, the plant by the window sways slightly. Quote the cost before you run it.

4. Step 4: Let it run in the background

   Video takes minutes, not seconds. In Claude, the agent can keep polling while you work on something else. In Cursor, run it as an agent task. When the clip is done, the agent returns a link to the MP4 and the clip sits on your FLORA canvas next to the still it came from.

5. Step 5: Iterate on one variable at a time

   If the motion is wrong, change only the motion words and rerun. If the framing is wrong, fix the still first. Each Seedance run is billed, so ask for a quote whenever you change duration or resolution.

What happened when we ran Seedance from Claude through FLORA?

On October 15, 2026, we animated a still we had just made with GPT Image 2.5: a stone-colored ceramic pour-over dripper on a glass carafe, on a creased linen tablecloth beside a window in soft overcast light. The agent first called flora_create_generations with quote set to true. FLORA returned an estimate of $2.48 for one 5-second 720p Seedance 2.5 clip, with no usage reserved.

The agent then called flora_create_generations for real with model i2v-gengateway-seedance-2-5-i2v, the still's URL in params.image_url, duration 5, resolution 720p and aspect_ratio 16:9, plus the motion prompt. It polled flora_list_generations until the run completed. The clip took 4 minutes 46 seconds and was charged $2.457.

The result is a 5.06-second, 1280 x 720, 24 fps MP4 with an audio track. The camera holds still the whole time, as asked. A thin stream of coffee falls from the dripper into the carafe and the sprig in the vase shifts slightly; nothing else moves, and nothing warps or changes shape. Color and light match the still.

What went wrong: we asked for 16:9 from a 3:2 still, so the model widened the scene. More of the window and the city behind it came into frame and the dripper sits smaller than in the still. The fix is to set aspect_ratio to auto, or generate the still at 16:9 first, so the clip keeps the composition you approved. The steam we asked for is barely visible; ask for it explicitly and keep the background darker if you need it to read.

Model

Seedance 2.5 (image-to-video)

Model id

i2v-gengateway-seedance-2-5-i2v

Cost

$2.457 for 5 s at 720p

Time

4 min 46 s

Claude or Cursor for Seedance: pros and cons

Does it work the same in Claude and in Cursor?

The FLORA tools are the same in both, so the request, the model and the cost are the same. The differences are around the edges. Cursor and Claude Code can upload a local image file to FLORA before animating it. Claude on the web works from images that already have a public link or sit on your FLORA canvas. Cursor suits batch jobs, such as animating every image in a folder, because it can read the folder and loop.

How do I write a good Seedance prompt?

Describe motion and camera in plain, physical terms: what moves, in which direction, how fast, and whether the camera holds still, pushes in or pans. Leave the scene description to the image. Short clips of 4 to 6 seconds with one clear action hold together best. If you want a camera move, name only one.

Ask for restraint when the subject is a product. "Locked-off shot, slow, nothing changes shape" keeps logos and edges stable.

• Say what moves: coffee drips, steam rises, fabric shifts.
• Say how fast: slow, steady, a quick flick.
• Say what the camera does: locked off, slow push in, or one pan.
• Give duration, resolution and aspect ratio in the same message.
• Leave out anything already visible in the still.

Can I animate several images at once?

Yes. Ask the agent to animate a set, and it sends up to 20 generations to FLORA in one call, then checks them together. Video runs take minutes each, so start the batch and come back. Ask for a quote on the whole batch first.

Frequently asked questions