alvdansen

alvdansen / klippbok

Public

Video dataset curation toolkit

44
4
100% credibility
Found Mar 08, 2026 at 44 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Klippbok is a toolkit for turning raw video footage into curated datasets ready for training custom video generation AI models.

How It Works

1
📹 Gather your videos

Start by collecting your raw footage or short clips in one folder, ready to turn into a training set.

2
✂️ Split into clips

Automatically find scene changes and cut long videos into perfect short clips for training.

3
🔍 Sort by content

Use simple reference photos to automatically group clips by characters, objects, styles, or scenes.

4
💭 Add descriptions

Let AI watch each clip and write helpful descriptions tailored to what you're training.

5
🖼️ Pick key frames

Choose the clearest starting image from each clip to guide your AI during training.

6
✅ Check and organize

Review quality, fix any issues, and neatly organize everything for smooth training.

🚀 Dataset ready!

Your videos are now a high-quality training set, perfect for creating custom video AI.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 44 to 44 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is klippbok?

Klippbok is a Python CLI toolkit for curating video datasets aimed at LoRA training on diffusion models like Wan. It splits long videos into training clips via scene detection, generates captions using VLMs like Gemini or local Ollama endpoints, scores caption quality offline, and validates datasets for resolution, motion, and bucketing. Users get clean, trainer-ready folders with paired clips, captions, and references—perfect for video dataset download and prep in github video ai workflows.

Why is it gaining traction?

It streamlines the messy pipeline from raw footage to batched training data, with standout features like prompt-tuned captioning for character, style, or motion use cases, automatic auditing against fresh VLM outputs, and bucketing previews to avoid GPU waste. Unlike scattered github video enhancer or upscaler scripts, it integrates validation, organization, and even trainer configs for musubi-tuner or ai-toolkit, saving hours on video dataset for face recognition or human activity recognition.

Who should use this?

ML engineers fine-tuning video LoRAs on custom clips for object detection, safe/unsafe behaviors, or character animation. Video AI devs dealing with triage-sorted folders from embeddings, needing quick caption batches or quality gates before training.

Verdict

Worth trying for video dataset curation despite alpha status (44 stars, 1.0% credibility)—CLI is intuitive, docs cover workflows, but expect tweaks for production. Pair with stable trainers for reliable LoRA results.

(187 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.