Free ImageGen

Free ImageGen

中文 README

A no-API, no-GPU, fully local image-generation skill for content-first workflows.

If you want to create:

• Xiaohongshu covers
• knowledge cards and infographics
• article-to-image carousels
• OpenClaw-ready visual assets

this is often a better fit than diffusion models.

Why this exists

Most image generators are optimized for:

• photorealism
• atmosphere
• visual spectacle
• “one prompt, one impressive-looking image”

Free ImageGen is optimized for something else:

• ✅ No API required — no image API bills
• ✅ No hardware barrier — no GPU dependency, no local diffusion setup
• ✅ Fully local — better for privacy, reproducibility, and cost control
• ✅ High freedom — the agent can decide pagination, layout, tone, and style
• ✅ Chinese-friendly — stable Chinese typography, mixed Chinese/English copy, phone-first readability
• ✅ Xiaohongshu-friendly — designed for covers, card posts, explainers, and article image sets

If your goal is content expression, not photorealistic rendering, this workflow is usually more reliable than diffusion-style generation.

Compared with diffusion models

Diffusion models are great when you want:

• realistic illustration
• painterly or cinematic results
• visual surprise
• detail-heavy freeform artwork

Free ImageGen is better when you want:

• structured visual communication
• article pages with readable text
• explainers, comparisons, checklists, maps, and QA cards
• a full image set generated from one article
• a renderer that follows the agent instead of replacing the agent’s judgment

In one sentence:

This is a local design renderer for content workflows, not a diffusion model chasing visual spectacle.

Best use cases

1. Xiaohongshu covers

Use it for:

• bold title covers
• opinion-led covers
• tool recommendation covers
• hero-emoji covers

2. Infographics and knowledge cards

Built-in card styles include:

• checklist cards
• comparison cards
• mechanism cards
• product maps
• QA cards
• flow cards
• timeline cards

3. Turn one article into a full image set

This is one of the strongest workflows.

Good fits:

• Feishu articles
• WeChat articles
• long-form notes
• interviews and summaries
• product updates
• AI tool explainers

4. Free-form SVG creation

When you do not want a built-in layout, the agent can directly produce SVG via custom_svg.

Good fits:

• mascots
• stickers
• single-page decorative visuals
• custom illustrations where the agent wants full control

Where the creativity comes from

This skill does not invent strong ideas for you.

More accurately:

• 🧠 The creative direction comes from the human
• 🤖 The visual strategy depends on the agent
• 🛠️ The renderer makes those decisions stable and exportable

That is the point of the workflow:

• the agent decides
• the renderer executes

If the agent is strong, the output can be very strong. If the agent is weak, this skill will not magically turn weak planning into a great image set.

Quick Start

Create one cover

python3 scripts/free_image_gen.py \
  --prompt "text cover, title AI Product Design Principles, subtitle Clear hierarchy and strong mobile readability, theme: light, cover layout: hero_emoji_top, hero emoji: 💡" \
  --output /absolute/path/output/cover.png \
  --width 1080 \
  --height 1440

Create one infographic

python3 scripts/free_image_gen.py \
  --prompt "infographic mechanism card kicker: three key points title: Why AI Agents suddenly exploded subtitle: It is not just the model — the entry point, UX, and distribution all changed 1. Lower barrier 2. Wider distribution 3. Real monetization" \
  --output /absolute/path/output/infographic.png \
  --width 1080 \
  --height 1440

Give it one article and generate a full image set

python3 scripts/free_image_gen.py \
  --prompt-file /absolute/path/article.md \
  --story-output-dir /absolute/path/output/article-story \
  --story-strategy auto \
  --width 1080 \
  --height 1440

This is good for a quick first draft.

Recommended workflow: let the agent plan first

This is the workflow that best uses the skill.

Ask the agent to:

• read the full article
• decide pagination
• decide which pages should stay article-like
• decide which pages should become checklist, mechanism, comparison, map, or QA cards
• keep the tone faithful to the source
• output story-plan.json
• then call free-imagegen to render the set

That gives you a workflow where:

• the agent thinks
• the renderer executes

Output behavior

Default behavior is intentionally clean:

• PNG only by default
• no extra SVG files unless you explicitly pass --keep-svg
• cleaner default naming
• cleaner default output folders

Freedom level

This skill is not “template only.”

You can:

• use built-in layouts for speed
• let the agent plan a full image series
• let the agent directly write custom_svg

So the freedom comes from this balance:

• structured when you need stability
• open when you need control

Fallback vs main path

illustration still exists, but it is now best understood as a lightweight fallback.

Use illustration for:

• quick abstract visuals
• simple decorative single images
• lightweight placeholder-style artwork

Use custom_svg first for:

• recognizable objects
• mascots
• animals
• robots
• stickers
• any page where the agent wants direct visual control

Resources for agents

If you want to integrate this into OpenClaw or another agent workflow, start here:

• references/story-plan.schema.json
• references/story-plan.template.json
• references/story-plan.guide.md
• references/story-plan.agent-prompt.md
• references/custom-svg-best-practices.md
• references/custom-svg.story-plan.sample.json

These files are meant to protect freedom without sacrificing structure.

Summary

If you want a workflow that is:

• free
• local
• API-free
• low-barrier
• Chinese-friendly
• Xiaohongshu-friendly
• agent-driven
• article-to-image capable

then Free ImageGen is built for exactly that.