A complete Claude Co-work setup for creating AI-powered motion graphics and explainer videos. This is the workflow I use for client projects like Tanishq jewelry ads and Vox-style explainers.
What you get: A ready-to-use Claude Co-work project with pre-configured skills, prompt templates, and a production workflow that cuts iteration time by half.
Who this is for: Motion designers, content creators, and video editors who want to use AI tools (Higgsfield, Kling, Runway) without spending hours on setup and prompt engineering.
2-minute payoff: Clone this, load the project in Claude Co-work, and you'll have a working motion graphics studio in under 10 minutes. No manual MCP setup, no prompt guessing.
I built this after wasting two weeks trying to get consistent results from AI video tools. The problem isn't the models — it's the workflow. Most people jump straight into generating clips without:
- A clear story breakdown
- Approved reference images
- Consistent style locks
- A system for iteration
This repo fixes that. It's the system I wish I had when I started.
- Claude Pro subscription (required for Co-work + Projects)
- Higgsfield account (free tier works, but paid gives you more generations)
- Claude desktop app (Co-work is not available in the web interface)
-
Clone this repo
git clone https://github.com/madebysaira/MotionGraphicsStudio.git cd MotionGraphicsStudio -
Open Claude Co-work
- Launch the Claude desktop app
- Go to Projects → Create New Project
- Name:
Motion Graphics Studio - Description:
AI-powered motion graphics and explainer video production - Location: Point to this cloned folder
-
Add the instruction file
- In the project setup, add
docs/project-instruction.mdas your project instruction - This tells Claude exactly how to behave for motion graphics work
- In the project setup, add
-
Connect Higgsfield MCP
- In Claude Co-work: Settings → Connectors → Add Connector
- Name:
Higgsfield - Follow the OAuth flow to sign in with your Higgsfield account
- Done. The presets are ready to use.
-
Load the skills
- In your project, go to Skills → Add Skill
- Point to the
skills/folder in this repo - You now have:
story-beats— Breaks your script into visual beatsmotion-builder— Generates reference images and video promptsstyle-lock— Maintains visual consistency across shots
- Start a new conversation in your Motion Graphics Studio project
- Paste your script or topic (e.g., "Create a 60-second explainer about how solar panels work")
- The workflow:
- Claude will first break it into story beats (wait for your approval)
- Then generate reference images for each beat (wait for approval)
- Finally generate the video prompts and clips
This approval-based workflow saves time. You catch issues early instead of generating 20 useless clips.
MotionGraphicsStudio/
├── README.md # This file
├── docs/
│ ├── project-instruction.md # Core project behavior (auto-loaded)
│ ├── workflow.md # Complete production workflow
│ └── troubleshooting.md # Common issues and fixes
├── prompts/
│ ├── story-beats.md # Story breakdown template
│ ├── reference-images.md # Image generation prompts
│ ├── video-clips.md # Video generation with motion params
│ └── style-sheets/
│ ├── vox-explainer.md # Vox-style documentary look
│ ├── ink-vector.md # Ink-style vector animation
│ └── corporate-clean.md # Clean corporate motion graphics
├── skills/
│ ├── story-beats.md # Skill: Script → Story beats
│ ├── motion-builder.md # Skill: Beats → Images + Video prompts
│ └── style-lock.md # Skill: Maintain visual consistency
├── examples/
│ ├── lebron-james-story.md # Complete example: 12-beat sports story
│ └── solar-panel-explainer.md # Complete example: 60-second explainer
└── LICENSE
Input: Your script or topic
Output: Approved story beats with timing
Example prompt:
I need a 60-second explainer video about [TOPIC].
Requirements:
- Duration: 60 seconds (approximately 12 beats at 5 seconds each)
- Style: [Reference style from style-sheets]
- Audience: [Your target audience]
Please break this into story beats. For each beat include:
1. Beat number and duration
2. What happens visually
3. Key text/numbers to display
4. Transition to next beat
Wait for my approval before moving to image generation.
Why this matters: Most people skip this and go straight to generating. You'll waste credits on clips that don't fit the story.
Input: Approved story beats
Output: Reference images for each beat
Example prompt:
Great, those beats are approved. Now generate reference images for each beat.
Use the [STYLE_NAME] style sheet from the prompts folder.
For each image:
1. Include the beat number in the filename
2. Generate at 16:9 aspect ratio
3. Focus on composition and clarity, not final quality
Generate only the reference images first. I will approve them before we move to video.
Pro tip: If you're generating a known person (athlete, celebrity), the AI will avoid showing faces due to IP restrictions. Work around this by:
- Using abstract/silhouette styles
- Focusing on actions and context (jersey numbers, settings)
- Generating "inspired by" rather than literal depictions
Input: Approved reference images
Output: Video clips with motion parameters
Example prompt:
Images approved. Now generate the video clips.
For each clip:
1. Use the approved reference image as the base
2. Add motion parameters (see prompts/video-clips.md for templates)
3. Generate at 6 seconds per clip (gives us editing room)
4. Include parallax effects where appropriate
Motion guidelines:
- Simple animations are more reliable
- Use parallax for depth on static shots
- Add camera movement (pan, rotate) on transition beats
- Keep character shots more static to avoid artifacts
Generate all clips and provide a shot plan artifact.
Input: Generated video clips
Output: Final edited video
This happens outside Claude (in your video editor), but the repo includes:
- A shot plan artifact (auto-generated by Claude)
- Transition suggestions
- Timing notes
Style sheets lock your visual consistency. Each sheet includes:
- Color palette (hex codes)
- Typography rules
- Composition guidelines
- Motion principles
- Example prompts
Vox Explainer (prompts/style-sheets/vox-explainer.md)
- Clean, documentary style
- Bold typography, high contrast
- Blueprint-style graphics
- Used for: Educational content, product explainers
Ink Vector (prompts/style-sheets/ink-vector.md)
- Hand-drawn vector aesthetic
- Limited color palette (3-4 colors)
- Organic motion, smooth transitions
- Used for: Sports stories, biographical content
Corporate Clean (prompts/style-sheets/corporate-clean.md)
- Professional, minimal
- Brand-safe colors
- Subtle motion, no flashy effects
- Used for: Client work, internal presentations
Copy an existing sheet and modify:
# [Style Name]
## Color Palette
- Primary: #XXXXXX
- Secondary: #XXXXXX
- Accent: #XXXXXX
## Typography
- Font: [Font name or description]
- Hierarchy: [How text is layered]
## Composition
- Framing: [Wide, medium, close-up preferences]
- Negative space: [How much, where]
## Motion
- Speed: [Fast, medium, slow]
- Easing: [Linear, ease-in-out, etc.]
- Transitions: [Cut, fade, slide, etc.]
## Example Prompt
[Copy-paste prompt that produces this style]Problem: You ask for a specific person (LeBron, Messi) and get generic figures.
Why: AI models have IP safeguards. They avoid generating recognizable faces of public figures.
Solutions:
- Use abstract styles (silhouettes, ink drawings, vector art)
- Focus on context (jersey numbers, settings, actions)
- Upload a reference image and ask for "inspired by" variations
- Accept that you'll need to composite real footage later
Problem: Generated clips have distorted frames, flickering, or strange motion.
Why: Complex motion + AI = unpredictable results.
Solutions:
- Simplify the motion request (start with static or slow pan)
- Generate multiple variations and pick the best
- Use shorter clips (3-4 seconds instead of 6)
- Add foreground elements for parallax (gives the AI more to work with)
Problem: Higgsfield MCP won't connect or authenticate.
Solutions:
- Check that you're using Claude desktop app (not web)
- Re-authenticate: Settings → Connectors → Remove → Re-add
- Check Higgsfield account status (free tier has limits)
- Try the CLI fallback (see
docs/troubleshooting.md)
Question: Which should I use?
Answer: Co-work for this workflow. Here's why:
| Feature | Co-work | Code |
|---|---|---|
| Setup | Built-in (Projects, Memory, Skills) | Manual config |
| Memory | Persistent per project | Requires manual setup |
| Skills | Native support | Requires MCP config |
| Best for | Creative workflows, iteration | Code generation, CLI tools |
Co-work is like WhatsApp. Code is like Slack. If you're just texting friends (creating motion graphics), WhatsApp is enough. You don't need Slack's channels and integrations.
Use Code if you need to run scripts, use ffmpeg, or integrate with CLI tools.
The skills in this repo are starting points. You can customize them or create new ones.
Want to auto-research an artist before generating? Create a skill:
# Skill: Artist Research
When the user mentions an artist or creator:
1. Search for their visual style (use web search if available)
2. Identify key visual elements (color, composition, subjects)
3. Create a style summary
4. Generate transformation prompts to apply that style
Example trigger: "Create a video in the style of [Artist Name]"Want to generate all clips in one go?
# Skill: Batch Generator
When all reference images are approved:
1. Read all approved image prompts
2. Generate video prompts for each (using video-clips.md template)
3. Send all generation requests in parallel
4. Compile results into a shot plan artifact
Warning: This uses more credits and may require more iteration.Higgsfield (as of July 2026):
- Free tier: ~20 generations/day
- Pro: $20/month, ~100 generations/day
- Studio: $50/month, unlimited
Claude Pro: $20/month (required for Co-work + Projects)
Estimated cost per project:
- 60-second video: ~30-50 Higgsfield generations
- With Pro: ~1-2 days of generation limits
- With Studio: One afternoon
- One-click video generation: There's no skill that makes a perfect video in one shot. You need iteration.
- Face generation: AI won't generate recognizable faces of real people. Plan for workarounds.
- Audio/voiceover: This workflow focuses on visuals. Use Higgsfield's built-in voiceover or add audio in post.
- Editing: Clips are generated raw. You'll need to edit them together (Premiere, DaVinci, CapCut).
This workflow was inspired by community tutorials and my own client work. The video that sparked this repo: Claude Co-work Motion Graphics Guide
I've improved on it by:
- Adding style sheets for consistency
- Creating reusable skills
- Documenting troubleshooting (the video skips the hard parts)
- Providing complete examples with prompts
MIT License. Use this for client work, modify it, share it. Just don't sell it as your own course.
If you run into issues:
- Check
docs/troubleshooting.mdfirst - Open a GitHub issue with details
- For Claude/Higgsfield-specific problems, check their docs
What I'm working on next:
- Character consistency toolkit (locking faces across shots)
- n8n automation for batch processing
- Integration with Remotion for programmatic editing
Built by Sagarika | madebysaira.me | Book a call