Deterministic Image Ops for AI Agents: image-utils

Where this fits

You've just generated a batch of product shots with an image model. They're 1024×1024, but your e-commerce site needs 800×800 JPEGs for the grid, 400×400 for thumbnails, and a watermark on every one. Running this manually is tedious and error-prone. The image-utils skill gives your agent deterministic Pillow-based operations—resize, crop, composite, convert, watermark, adjust—so it can handle the entire pipeline without a human in the loop.

Why agents benefit

• Reliable pixel-level control: Every operation is deterministic—no hallucinated dimensions, no unexpected color shifts. Agents can trust the output.
• Batch and variant generation: Resize a directory to multiple sizes, convert formats, or generate responsive variants in a single call, reducing tool-call overhead.
• Post-processing chaining: Works seamlessly after generation APIs (like bria-ai) to crop, compress, and optimize images before delivery.
• Web-ready output: Built-in quality controls for JPEG/WEBP compression and resizing ensure fast load times without manual tuning.
• Watermarking and compositing: Add brand overlays or paste subjects onto backgrounds—common tasks that are otherwise fiddly to script mid-agent-run.

Practical scenarios

E-commerce catalog preparation

After an agent generates lifestyle images for 50 products, it uses image-utils to resize each to 800×800, convert to JPEG with quality=85, and overlay a subtle logo watermark. The whole batch completes in seconds, and the agent moves on to writing product descriptions.

Social media variant creation

A marketing agent receives a single hero image. It crops to 1:1 for Instagram, 4:5 for feed posts, and 16:9 for cover photos, then optimizes each for web delivery. No manual design tool needed.

Content moderation and cleanup

An agent processes user-uploaded images: it strips EXIF data, converts to a standard format, and resizes to a max dimension to prevent oversized uploads—all while preserving visual fidelity.

Add it to your agent workflow

Install with a single command:

npx skills add Bria-AI/bria-skill --skill image-utils

Then use it in your agent logic. Example: resize and convert a generated image for web delivery.

from PIL import Image

def optimize_for_web(input_path, output_path, max_width=800, quality=85):
    img = Image.open(input_path)
    img.thumbnail((max_width, max_width))
    img.save(output_path, "JPEG", quality=quality, optimize=True)

The skill wraps these operations into a clean callable interface your agent can invoke with simple parameters—no need to reimplement Pillow logic each time.

Compare before adopting

Before integrating, evaluate:

• Quality signals: Check the repo's issue tracker and star history (currently 65 stars). Does it handle edge cases like transparency, EXIF orientation, or color profiles?
• Maintenance freshness: Last commit date and release cadence. Pillow updates can break older APIs.
• Alternatives: Some agents use direct Pillow code or other libs like OpenCV. This skill is a lightweight wrapper—consider if you need advanced filters or GPU acceleration.
• Workflow fit: Does it integrate with your existing image generation and storage stack? It works standalone or alongside bria-ai, but verify your agent's orchestration supports the skill's input/output contracts.

Why it is worth tracking

image-utils is a focused, MIT-licensed skill with a clear scope and active maintenance from Bria AI. Its simplicity is a strength—deterministic operations are exactly what agents need for reliable pipelines. As more agents handle visual content, this pattern of wrapping common operations into a reusable skill will become standard. Keep it on your radar if you're building anything that touches images, and check the skill page for updates and community feedback.