Picture It

How to Write Good Prompts

This is the difference between mediocre and professional output. Read references/prompt-library.md for a full library of tested prompts you can copy and adapt. Key rules:

For generation: Be specific about lighting ("dramatic side lighting from upper right"), camera ("shot on Canon R5 70-200mm f2.8"), and atmosphere ("dust particles visible in the light beam"). Vague prompts produce generic results.

For text-behind-subject: The key phrase is: "Add '[TEXT]' in large bold [color] letters BEHIND the [subject] — the [subject's] body overlaps and partially covers the letters." Without "BEHIND" and the occlusion instruction, the text floats on top.

For edits: Always end with "Keep everything else exactly the same" and list what to preserve. Without this, the AI changes things you didn't want changed.

For background replacement: Use realistic, specific locations ("modern upscale mall entrance during daytime, natural warm daylight"). Over-dramatic backgrounds ("city at night with neon reflections") look obviously fake.

Typography

For big titles and hero text: Use the FAL model via edit — it handles large text well and integrates it into the scene naturally. No font size math needed, just say "very large bold" in the prompt.

For precise small text (credits, URLs, badges, coverlines): Use compose or text with Satori. This is where font sizing matters — images display much smaller on phones. Quick rule: on a 1080px Instagram image, nothing under 36px is readable. Run picture-it download-fonts first if fonts aren't installed.

Hierarchy: Max 3 text sizes per image. Brand name should be larger than tagline.

Font pairing: Serif + sans-serif works best. For FAL model text, just describe the style in the prompt. For Satori, 3 fonts are bundled — drop more .ttf files into ~/.picture-it/fonts/. Run picture-it download-fonts if fonts aren't installed. See references/composition-guide.md for pairing suggestions.

Composition Techniques

Read references/composition-guide.md for detailed multi-pass workflows, product photography, magazine covers, and overlay composition.

Common Workflows

Simple: Generate an image

picture-it generate --prompt "dark cosmic background with nebula" --size 1200x630 -o bg.png

Simple: Add text to an image

picture-it text -i bg.png --title "Hello World" --font "Space Grotesk" --color white --font-size 64 -o hero.png

Medium: Blog header with AI background + text

picture-it generate --prompt "abstract dark tech background" --size 1200x630 -o bg.png
picture-it text -i bg.png --title "My Blog Post" --font "DM Serif Display" --font-size 72 -o header.png
picture-it grade -i header.png --name cinematic -o header-graded.png

Medium: Edit a photo background

picture-it edit -i photo.jpg --prompt "replace background with modern hotel entrance, keep subject identical" --model banana-pro -o edited.jpg

Advanced: Text behind subject (YouTube thumbnail style)

# 1. Generate a scene
picture-it generate --prompt "runner on mountain trail at golden hour" --model flux-dev --size 1280x720 -o runner.png

# 2. Use FAL edit to add text BEHIND the subject
picture-it edit -i runner.png --prompt "Add 'RUN FASTER' in large bold black letters BEHIND the runner — the runner's body overlaps the text" --model seedream -o thumbnail.png

Advanced: Product comparison with real photos

# 1. Remove backgrounds from product photos
picture-it remove-bg -i product-a.png --model bria -o a-cutout.png
picture-it remove-bg -i product-b.png --model bria -o b-cutout.png

# 2. Generate a background
picture-it generate --prompt "split gradient, blue left to orange right" --size 1200x630 -o bg.png

# 3. Compose cutouts onto background with text
picture-it compose -i bg.png --overlays overlays.json -o comparison.png

Advanced: Multi-pass cinematic composition

# 1. Generate base scene
picture-it generate --prompt "dark stage with green spotlight" --model flux-dev --size 2048x1080 -o stage.png

# 2. Edit scene to place objects
picture-it edit -i stage.png -i logo.png --prompt "Place Figure 2 as glowing 3D cube in the spotlight" --model seedream -o composed.png

# 3. Post-process
picture-it crop -i composed.png --size 1200x630 --position attention -o cropped.png
picture-it grade -i cropped.png --name cinematic -o graded.png
picture-it vignette -i graded.png --opacity 0.3 -o final.png

Platform Presets

Use --platform <name> with generate or crop:

Output Behavior

• stdout: only the output file path
• stderr: progress logs
• Exit 0 on success, Exit 1 on failure

Read stdout to get the file path. This is how you chain commands.

Gotchas

• Always use --model bria for remove-bg — the default birefnet leaves rectangular artifacts that cause ugly glow/shadow halos when compositing.
• The glow effect in compose mode blurs the entire rectangular buffer, not the shape. Avoid using glow on cutout images — use the background color/lighting to create the glow effect instead.
• The shadow effect has the same rectangular artifact issue. For cutout images on clean backgrounds, skip shadows entirely.
• When editing with FAL, the model may alter product details (logos, text, design elements). For product images where accuracy matters, use remove-bg + compose instead of edit to preserve the original exactly.
• SeedDream takes ~60 seconds per generation. Don't assume it failed if it's slow.
• For edit with banana-pro, don't pass resolution or limit_generations params — it auto-detects.
• Always crop to exact dimensions after FAL generation — FAL models output approximate sizes.
• Use flux-dev ($0.03) not flux-schnell ($0.003) when image quality matters (hero images, portraits). The quality difference is significant.
• Satori does NOT support: display:grid, transforms, animations, box-shadow, filters. Use flexbox only.
• When adding text behind a subject with edit, be very explicit in the prompt: "the text is BEHIND the subject — the subject's body overlaps and partially covers the letters."