Skip to the usage guide
Video Generationby bytedance

Video Generation Skill.

Generate video from structured prompts in a configured DeerFlow runtime.

  • Video Generation
  • Model Selection
Source Review · video-generationSKILL.md
Task Generate video from structured prompts in a configured DeerFlow runtime. Produces A structured English JSON prompt file. Before You Start Use a configured DeerFlow runtime with the required provider access. Prepare the scene brief, actual reference images and output destination; confirm the provider before relying on particular reference or aspect-ratio behavior.
Editorial summary · source reviewed 2026-11-02 · not a generated example

01 What It Does

Generate video from structured prompts in a configured DeerFlow runtime

Prepare JSON video briefs with references and provider-aware execution, while separating successful new outputs from failed or stale jobs.

Source-based workflowNot an execution benchmark

The Working Approach

video-generation packages a video brief as structured JSON and runs it through DeerFlow's Python generation workflow. The prompt can describe the setting, characters, camera, dialogue and audio, with optional reference images. The source keeps prompt files and generated outputs in separate workspace locations.

Provider selection comes from environment configuration. The documented routes include Gemini and MiniMax, whose reference and aspect handling differ. Failure is explicit: a missing credential or failed job should produce an error, and an older file must not be presented as a newly generated result.

Expected Deliverables

  • A structured English JSON prompt file.
  • A generated video from the configured provider.
  • An optional reference image and an accurately reported generation status.
See the Example Brief

From Brief to Review

  1. 01
    Structure the brief

    Record subject, setting, style, camera and technical intent. Resolve any reference image before submitting the job.

  2. 02
    Use the configured provider

    Pass the prompt file, references and output destination to the workflow. Keep credentials in the runtime environment and check provider-specific behavior.

  3. 03
    Verify the current result

    Require successful command completion before presenting the file. Confirm the output belongs to this run and inspect the actual video against the prompt.

02 When to Use It

When to Use It

Use a configured DeerFlow runtime with the required provider access. Prepare the scene brief, actual reference images and output destination; confirm the provider before relying on particular reference or aspect-ratio behavior.

A Good Fit

  • A repeatable generation request needs structureJSON separates camera, setting and audio decisions into a reusable artifact.
  • DeerFlow already supplies the runtimeThe Skill fits an environment with the script and provider credentials configured.

Limits to Check

  • Workspace paths are environment-specificThe source uses DeerFlow's mounted directories. Another host needs an equivalent setup rather than assuming those paths exist.
  • Providers do not interpret every option identicallyThe documented MiniMax route uses the first reference image and ignores the aspect-ratio argument.
  • Generation needs external accessThe provider, credentials and any associated generation cost remain separate from the Skill file.

Bring These Inputs

  • Scene briefDescribe subject, action, location, style and camera intent.
  • Reference imagesSupply real image files with extensions matching their contents.
  • Provider and outputIdentify the configured provider, destination and authorized generation requirements.

03 Decisions That Matter

Decisions That Matter

  1. Reusable prompt structure

    The JSON file makes creative changes visible between attempts.

  2. Explicit provider routing

    Environment configuration controls which generation service receives the request.

  3. Honest failure reporting

    The source requires distinguishing a completed run from errors and stale artifacts.

04 Try a Focused Brief

Try a Focused Brief

The example brief below has not been tested. Check the skill’s setup before using it.

Review the Source and Setup

Describe subject, action, location, style and camera intent.

Upstream Source File
https://github.com/bytedance/deer-flow/blob/3f86f071f3d26cb3fcf91841523f6f7be2796914/skills/public/video-generation/SKILL.md

The reference below opens the upstream instructions. Review the repository setup before using them; it is not an installation command.

Start With One Deliverable

Adapt the scope, source material and constraints in this brief before running the skill.

Example Brief
Use video-generation in the configured DeerFlow environment to prepare a structured prompt for a quiet workshop scene. Show a ceramic cup on a wooden table as morning light moves across it, with one gentle camera push and soft room ambience. Use my supplied image as the visual reference. Confirm which provider is configured and state any reference or aspect-ratio limitation before submission. Keep credentials out of the prompt. After authorized generation, present only the new file from a successful run, inspect it against the brief, and preserve the JSON prompt for revision.

Inspect the Result

Check the actual output against these observable criteria.

  • Confirm provider selection and supported reference handling.
  • Require a successful current job rather than reuse of an old file.
  • Inspect actual output framing, motion and correspondence to the structured brief.

OpenDesign provides a local design workspace for coding agents. This listing does not mean the skill is bundled or that its external tools, credentials and software dependencies are included.

05 Source and License

Source and License

This page describes the pinned source revision linked below. Upstream behavior and requirements can change.

Source Owner
bytedance
Skill File
SKILL.md
License
MIT
Reviewed Snapshot
2026-11-02 · 3f86f071f3d2
Editorial Source Summary
Generate video from structured prompts in a configured DeerFlow runtime.

Original summary by OpenDesign, based on the linked source. Repository stars are a dated popularity signal, not skill usage or a quality score.

06 Questions Before You Start

Questions Before You Start

Can it use more than one provider?

The source documents environment-based selection between Gemini and MiniMax, with an explicit override variable.

Do reference options behave identically?

No. The source states that MiniMax uses the first reference as its first-frame image and handles aspect differently from the generic argument.

What happens when generation fails?

The command reports failure, including missing credentials. The source prohibits presenting an existing older output as a successful new result.

OpenDesign

Keep Your Brief and Design Work Together.

Keep your brief, source material and design work together in a local workspace. Check the skill’s own setup before adding it to your workflow.

Upstream Source File
https://github.com/bytedance/deer-flow/blob/3f86f071f3d26cb3fcf91841523f6f7be2796914/skills/public/video-generation/SKILL.md