After building an 80-second animated short by hand on the Dreamina (Jimeng) canvas, Claude Code scripts the pipeline through the canvas CLI: change only the lead character, bear to penguin, and 14 images, 10 clips and the music re-run automatically
Chinese creator @servasyy_ai (huangserva) first built an 80-second animated short by hand on the new Dreamina canvas (a bear dancing in an old-town square): lock the character with a subject + turnaround sheet, cut 4- and 9-panel storyboards into single frames, hang a Seedance 2.5 video node off each frame, then add music on the timeline and export. Then he had Claude Code turn that node graph into a script via the canvas CLI with one parameter—who the lead is—and re-ran it as a penguin; he says it cost 1,750 credits and he clicked no nodes. Note: the post carries a #DreaminaPartner tag (sponsored). On Linux we installed the official dreamina-canvas CLI 1.1.1 (China build) and confirmed the quote → confirm-with-credit-ceiling → run-with-token commands, image upscale and draft-to-final promote exist; we did not log in or generate anything.
Visit site Dreamina / 即梦 AI Canvas + dreamina-canvas CLI + Claude Code ↗- Pricing
- Freemium
- Region
- Global
- Where it's from
- Chinese community
- Added
- Oct 28, 2026
- Tested on
- Oct 28, 2026
- Version we used
- dreamina-canvas CLI 1.1.1 (China public build, commit a5ca7c8, built 2026-10-23); source post 2026-10-27
- Categories
- VideoAutomation & Agents
- Original author
- huangserva @servasyy_ai ↗
- Our test
- Partly tested
- Cost / time
- The CLI is free; generation spends Dreamina credits—about 1,750 for the whole penguin re-run per the post (check current plan prices on the site). You also need your own Claude Code plan.
The task
Produce several versions of the same storyboard and prompts with different leads (account farms, series, client variants) without re-clicking dozens of canvas nodes each time.
Steps
- Build one version by hand: on the Dreamina (Jimeng) canvas, write a character description, save one image as a "subject", add a turnaround sheet for side and back views, and reference that subject in every shot so the character stays consistent.
- Storyboard to video: generate 4- and 9-panel continuous storyboards, cut them into single frames with the grid splitter, and hang a video node off each frame using Seedance 2.5 for 8–10 s clips.
- Credit saver (from the post): one 8 s shot straight at 1080p costs 512 credits; 720p then smart-upscale to 4K costs 248 in total.
- Edit on the canvas: drag ten clips plus music onto the timeline and export the film there.
- Install the canvas CLI (China build): `curl -fsSL https://jimeng.jianying.com/canvas-cli/install.sh | bash` puts `dreamina-canvas` in `~/.local/bin` and drops an agent skill in `~/.agents/skills/dreamina-canvas-cli`; check with `dreamina-canvas --help`, `version`, `schema`. International Dreamina users should start from the official install page dreamina.capcut.com/ai-tool/install.
- Log in: `dreamina-canvas auth login` (in an agent use `--non-interactive auth login`, then `auth wait`), and confirm with `auth account`.
- Have Claude Code read the CLI `schema` plus the nodes of your hand-built canvas (`node find` / `node show`) and turn the flow into a script whose only parameter is the lead character.
- Per the author's pitfalls: create an empty canvas in the web app first and let the script only write nodes into it (canvases created from the CLI wouldn't open on the web).
- Before each batch: `node quote` for the price, `node confirm --credit-ceiling <cap>` to get a credit token, then `node run --credit-token ...`; if the total exceeds the cap it stops without charging.
- Do video upscaling and the final export back in the web timeline (the author says the CLI only upscales and downloads images).
Prompt
Read `dreamina-canvas --format json schema`, then use node find / node show to inspect every node of the "bear" version on my current canvas (subject, turnaround, storyboards, grid split, 10 video nodes, music). Turn that flow into one script with a single parameter: the lead character description. Write new nodes into the empty canvas I created in the web app—do not create a canvas from the CLI. Before each batch run node quote; if the total exceeds 2,000 credits, stop and ask me instead of running.
What the author says
@servasyy_ai: the hand-made 80 s bear short used subject + turnaround for consistency; Seedance 2.5 at 8–10 s per panel; 720p + smart upscale to 4K costs less than half of native 1080p (248 vs 512 credits). After Claude Code scripted the flow via the canvas CLI, the penguin version—14 images, 10 clips, 1 music track—ran on its own for 1,750 credits total with "not one node clicked"; the video ends with a side-by-side of the hand-made bear and Claude Code's penguin. The post is tagged #DreaminaPartner and promotes a $0.29/month Basic plan.
Our test
We did not log in, generate any images or video, or drive a script from Claude Code; the 1,750-credit figure and quality comparison are unverified. We ran the official China-build installer into an isolated directory (no PATH change): dreamina-canvas 1.1.1 (commit a5ca7c8, built 2026-10-23; release notes mention new Seedance 2.5 draft generation and draft-to-final), and ran --help, version, schema, auth status and several subcommand --help pages.
What we found when we tried it:Command families: auth / canvas / model / node / operation / resource / space / team / voice / schema. Spending really is three steps: `node quote` (pre-run estimate) → `node confirm --credit-ceiling` (no token if the cap is below the server's latest quote) → `node run --credit-token`; the returned submitId can be resumed with `operation wait`. `node upscale` help says it upscales generated images, matching the author's "images only"; there's also `node promote` (draft to final). `canvas create` exists with `--use`, so the CLI can create canvases—the author's point is that such canvases wouldn't open on the web, which we couldn't verify without logging in. `schema` returns ~296 KB of JSON command contract, handy to feed an agent. Logged out, `auth status` returns loggedIn:false, region cn.
How to get in
The original post is in Chinese on X. In China this is Jimeng (jimeng.jianying.com, Douyin login); internationally it's Dreamina (dreamina.capcut.com). The CLI builds differ by region—international users should use the installer from dreamina.capcut.com/ai-tool/install.
Who it suits
Creators running account farms, animated series or multi-version client work who can already hand-build one good version on the canvas.
Pitfalls
- This is a #DreaminaPartner sponsored post; credit and quality comparisons are the author's—we generated nothing.
- Author's pitfall: canvases created from the CLI wouldn't open on the web (draft_reader_too_old); create an empty canvas on the web first and let the script add nodes.
- Author's pitfall: the CLI only upscales images (videos must be upscaled one by one on the web) and only downloads images; export the final cut from the web timeline.
- In our check, `node upscale` help literally says it upscales generated images, matching the author.
- The China build we installed only accepts `--region cn`; international Dreamina accounts need the overseas build—follow the official install page.
- Saving drafts or uploads is not approval to spend; make the script quote first and set a credit ceiling—never give an agent uncapped run rights.
- Swapping the lead only swaps the look; storyboard and prompts are reused, so watch the result to check the action and mood still fit.
More in this category
- A demo video per docs page that flags itself when the UI changesVideoScreenplayClaude CodeTool free; uses your agent's quota◐ Partly testedOverseas
- Code-drawn stick-figure explainers for about $0.54 eachVideoClaude CodeFFmpegChromeNeeds Claude plan or API; renders locally○ Not tested by usChina
- One topic in, a narrated tech explainer video out, locallyVideoClaude CodeChatterboxffmpegSkill free; needs a Claude Code plan◐ Partly testedOverseas
- Render restrained motion graphics from code, frame by frameVideoClaude CodePlaywrightffmpegNeeds a Claude plan; other tools free◐ Partly testedOverseas
- One reference image → a take-apart machine simVideoteardownClaude Codethree.jsSkill free; needs Claude Code/Opus◐ Partly testedOverseas