skills.rest
sabiscore
sabiscoreCOMMUNITY
Explore all skills by this author →
GitHub

swarmx-vision-storyboard

Critiques storyboard frames for caption placement, composition, and visual continuity using a local vision model.

1|Updated Sep 3, 2026
Search Tags:#Ollama#video generation#storyboard#vision model#visual continuity#caption placement#frame critique
One-click install
npx skills add https://github.com/sabiscore/the-yap-engine --skill swarmx-vision-storyboard-sabiscore
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: swarmx-vision-storyboard
Source: https://github.com/sabiscore/the-yap-engine/tree/main/integrations/openclaw/skills/swarmx-vision-storyboard
Command: npx skills add https://github.com/sabiscore/the-yap-engine --skill swarmx-vision-storyboard-sabiscore

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Automated video pipelines can render storyboard frames with weak subjects, colliding captions, or broken visual continuity that only a human reviewer would catch. This Skill uses an approved local vision model to inspect frames and return structured visual findings before render and QC code accepts or rejects a change. ## Core Features & Use Cases - Frame-Level Visual Critique: Scores subject salience, OCR collision risk, contrast, clutter, and caption safe-zone placement per scene. - Continuity Analysis: Detects continuity and motion/camera coherence issues between adjacent frames and checks template-family fit. - Structured JSON Output: Returns per-frame findings with a pass, revise, or blocked status that deterministic render/QC code consumes. - Use Case: During short-form video generation, send the minimum required storyboard frames to the on-demand qwen3-vl:4b worker, receive structured revision guidance, and let QC code decide whether to accept the change. ## Quick Start Ask the vision worker to critique the current storyboard frames from SwarmX and return structured findings on caption placement, salience, and continuity.

Frequently Asked Questions about swarmx-vision-storyboard

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I review storyboard frames with a local vision model?▼

Request storyboard metadata from SwarmX, select only the minimum frames needed, and send them to the on-demand vision worker. The model returns structured JSON findings per frame covering salience, caption zones, OCR risk, and continuity.

What vision model is used for storyboard critique?▼

The preferred experimental worker is ollama/qwen3-vl:4b, run through Ollama. It is an on-demand worker and should not be kept resident during normal video generation to preserve memory on constrained hosts.

What visual issues does storyboard frame analysis detect?▼

The check covers subject salience, caption safe-zone placement, OCR collision risk, contrast, clutter, continuity between adjacent frames, motion and camera coherence, and template-family fit. Each frame receives a visual_failure classification and revision suggestion.

Can the vision model directly change production video output?▼

No. Vision findings are advisory only and must never silently become production state. Deterministic render and QC code decides whether a suggested change is accepted based on the structured pass, revise, or blocked status.

How does the vision worker handle limited RAM on CPU-only hosts?▼

Before loading the vision model, conflicting resident models are evicted or avoided according to the active host profile. Only the minimum required frames are sent, keeping memory pressure low on constrained hardware.

Related Skills

Explore More Design & Creative Skills→
minicooheiminicoohei
345

storyboard-generator

Generate UGC storyboards with consistent characters and scenes.json metadata.

Community
Advanced
celeroncoderceleroncoder

product-launch-video

Generates product launch and promo videos from URLs or briefs using HyperFrames.

Community
Advanced
mtsatryanmtsatryan
1

shorts-generator

Generates scripts, visuals, captions and voiceover-ready short videos for platforms like YouTube Shorts, TikTok and Instagram Reels.

Community
Advanced
OpenLAIROpenLAIR
1.0k

llava

Unify a vision encoder with a language model for multimodal image understanding.

Official
Advanced