PicsArt Text to Visual
A single skill that covers every "given text, produce a matching image" workflow Picsart's gen-ai CLI supports. Use when the user has any kind of written content — a paragraph, a blog draft, a URL — and needs visuals generated from it. Replaces three narrower skills with one entry point and three mode references.
Input: text (paragraph, article, or URL). Output: one or more images matched to the text's tone, topic, and target placement.
When to Use
| Mode | Trigger phrases | Reference |
|---|---|---|
| single | "match a visual to this paragraph", "inline image for this section", "one visual for this content" | references/modes/single.md |
| article-set | "illustrate this blog post", "hero + inline visuals for an article", "full visual set for a draft" | references/modes/article-set.md |
| og | "OG image for this URL", "open graph preview", "Twitter card image", "dynamic meta image" | references/modes/og.md |
If the user wants product photos transformed, that's product-photo-studio. If they want video, that's gen-ai-use.
Prerequisites
Picsart gen-ai CLI installed and authenticated:
curl -fsSL https://picsart.com/gen-ai-cli/install.sh | bash
gen-ai login
gen-ai whoami
Per-mode setup (caching, serverless endpoints, font choices) is documented inside each mode reference.
How to Run
- Identify the mode from the user's request using the table in When to Use.
- Load the corresponding mode reference:
Readreferences/modes/<mode>.md. - Follow the procedure described there — extract text signals, build prompt, generate.
- Return to this SKILL.md only when switching modes mid-task.
Quick Reference
# Single image from a prompt
gen-ai generate --model <model> --prompt "<derived from text>"
# Estimate cost first
gen-ai pricing --model <model> --count <N>
# Browse available models
gen-ai models
Prompt-construction patterns (how to derive a prompt from a paragraph, an article, or a URL's metadata) live in the individual mode references.
Procedure
Shared outer loop:
- Extract signals — pull subject, tone, palette hints, and target dimensions from the input text.
- Build prompt — translate signals into a
gen-aiprompt; each mode has its own template. - Estimate —
gen-ai pricingbefore committing for multi-image runs. - Generate — invoke
gen-ai generate. Stream progress. - Place — drop into the right slot: PDP, blog frontmatter, OG meta tag, social variant.
Pitfalls
- Don't generate from raw text. Always extract signals first; raw paragraphs produce literal, lifeless images.
- Match aspect ratio to placement. OG = 1200×630, blog hero = 16:9, social = varies.
- Cache OG images. Don't regenerate on every page view — see
references/modes/og.md. - Mode-specific pitfalls live inside the individual mode references.
Verification
# Confirm output exists and matches expected dimensions
gen-ai inspect outputs/<run>/<image>.png
# Spot-check the visual matches the source text by re-reading both side by side
See also
product-photo-studio— transform existing product photosgen-ai-use— foundational gen-ai CLI reference