How to Make Seedance Videos in Claude Code or Codex

You can create Seedance videos from Claude Code or Codex by connecting the AiuniVid MCP server, authorizing your account and asking your assistant to quote each shot. After you approve the cost, the assistant can submit the shots and retrieve their video links. You direct the story and approve spending at each stage.

Eian’s November 5 article demonstrates a production workflow that gives planning, image preparation, animation and review separate checkpoints. We apply that approach to an original coffee-ad example with AiuniVid’s hosted MCP tools and per-shot credit quotes. The source article is linked below.

Give the assistant a production role

A useful workflow starts with clear responsibilities. Choose Claude Code or Codex as your coordinator and give it a brief, output format and budget. The assistant can then keep the storyboard and tool calls in the same conversation.

For image preparation and local editing, check which image tools and file permissions your assistant actually has. AiuniVid’s generate_image tool charges per image. Its price belongs in the production budget alongside the video quotes. A local FFmpeg installation can handle assembly when you authorize that work.

Connect Seedance to Claude Code, Codex or Claude

Create an AiuniVid account and choose your client in the MCP connection guide. The commands below come from the same configuration used in that guide. Run the relevant command once, then complete browser authorization for that client.

claude mcp add --transport http --scope user aiunivid https://aiunivid.com/api/mcp

codex mcp add aiunivid --url https://aiunivid.com/api/mcp
codex mcp login aiunivid

In the Claude app, use the remote connector settings available to your account, enter the MCP server URL below and complete authorization. Claude app and Claude Code each have their own connection setup.

https://aiunivid.com/api/mcp

Claude app connection guide

Check the AiuniVid MCP connection by calling list_models and get_credit_balance. Report the available model IDs, supported inputs, durations and resolutions. Keep this step read-only and wait for my production brief before generating media.

An optional Seedance skill can supply reusable shot-planning and review instructions. Complete the MCP connection first so those instructions have tools to call. You can also use the prompts in this article directly.

Seedance skill setup and examples

Start with a brief that the assistant can turn into shots

For a first project, use a short product teaser with three independent shots. This gives you room to choose useful clips individually. The example uses a coffee bag, beans and a cup, so the visual continuity depends on a few clearly described objects.

Plan a 15-second teaser for a small coffee brand in three separate 5-second shots. Use a 16:9 frame, warm morning light, an ivory table and a plain charcoal coffee bag. The sequence is beans catching the light, a cup filling with coffee, then the bag beside a steaming cup. Give each shot an opening frame, closing frame, one main action and one camera move. Keep faces, readable labels and dialogue out of this concept. First return the storyboard and possible continuity problems. Wait for my approval before generating images or video.

Review the storyboard as an audience member. The first image should establish the product, the middle shot should show its use, and the ending should leave space for a title added in editing. Change any detail that makes the product or action hard to recognize at phone size.

Build a storyboard before buying video seconds

Use this shot table as a starting point. Give the assistant your own product dimensions and colors when you have them. If you have product photos that you are entitled to use, ask it to inspect those before choosing a supported image-input model.

When a shot needs more precise composition, prepare first and last images and check that the chosen model supports those roles. For independent text-to-video shots, describe the endpoints in the prompt. For image-driven shots, prepare actual files before requesting the final quote.

• Check that the bag, cup, table and light direction match across the storyboard.
• Make every object needed for an action visible at the beginning of that shot.
• Check the aspect ratio and dimensions of every reference image against the selected model.
• Keep room for the pour, camera movement and any title that you will add during editing.
• List any uncertain identity, hand, reflection or lettering details before submitting a shot.

Review these three shots as if you are seeing the plan for the first time. For each shot, check whether its opening frame, action and closing frame can work together. Compare the bag, cup, light direction and object count between shots. Return specific corrections and identify anything you could not inspect. Wait for me to approve the revised storyboard.

If your assistant supports an independent reviewer, give that reviewer the storyboard and reference images. Keep the review focused on observable details. A manual review using the same checklist is also useful.

Quote the whole storyboard before approving generation

For a comparable starting budget, the table below uses three independent 5-second text-to-video tasks at 720P and 16:9, with audio disabled. Each model is priced using AiuniVid’s published quote at page-render time.

These are public text-to-video estimates. Your account quote may reflect personal pricing or a current campaign. Image generation, different inputs, reference-video durations and extra attempts belong in the budget separately. Use quote_video with the final settings and real media URLs before approving each task.

Use list_models to check the available Seedance text-to-video models. For the approved coffee storyboard, use quote_video for each of the three shots with Seedance 2.0 Mini, 5 seconds, 720P, 16:9 and audio disabled. Show the exact model ID, credit cost per shot and total. If a setting is unavailable, ask me to choose an alternative. Wait for my explicit approval before calling create_video.

Compare model costs in the public calculator

Approve one shot, then review its result

Start by approving shot 01 at its quoted price. Save its task ID and prompt together, then use get_video to follow the result. Once you have watched the clip, decide whether to reuse that setup for the remaining shots or revise the visual plan.

Generate only shot 01 using the settings and credit cost in the latest quote I approved. Return its task ID, status and video link. Keep shots 02 and 03 pending until I review shot 01. Ask again before any higher quote, new image generation or paid retry.

If a quote expires or the prompt, model or media changes, request another quote and compare it with the approved budget. For a continuation that needs the previous result’s last frame, retrieve that real frame first and quote the image-input task next.

For every paid retry, record what you changed and what it will cost. Keeping earlier clips lets you compare versions and choose the strongest usable result.

Assemble clips and check the final video

AiuniVid returns generated clips. Use an editor or authorized local tools to join the selected versions, add titles and mix audio. Give the assistant a separate editing brief with the filenames, order and intended export format.

• Watch every cut at normal speed and check the frames immediately around it.
• Compare the product’s shape and color across all selected shots.
• Listen to the complete audio mix and confirm permission for any added music.
• Check readable titles, framing, duration and export quality on a phone.
• Keep the selected clip links, prompts, quotes and final export together.

Review the assembled 15-second coffee teaser. Check shot order, product continuity, cut points, subtitles and audio using the tools you actually have. Name the files and frames you inspected and state which motion or audio checks still need my viewing or listening. Return a delivery checklist and the final export path.

For recurring characters, use a separate continuity plan. For Seedance 2.5 draft mode, include both the draft and the approved final render in the budget. Both topics have dedicated guides below.

Keep characters consistent across Seedance shots

Plan Seedance 2.5 drafts and final renders

Questions about Seedance video creation with MCP

Sources and implementation details

The production approach was informed by Eian’s article. This coffee concept, prompts and AiuniVid implementation are original additions. Client setup is documented by the tool vendors. Prices in the example come from AiuniVid’s current public quote.

• Eian · Claude Code / Codex + Seedance
• Claude Code · MCP
• OpenAI · Codex MCP