AI-Powered Image Compression: How Mochify's Natural Language Feature Works (2026)
Type "Instagram-ready WebP, strip location data" and Mochify handles the rest. Mistral Small 3.2 parses your intent and passes the right flags to our compression engine - no dropdowns, no format knowledge required.
Published March 2026 by the Mochify Engineering Team. We've connected an LLM directly into our compression pipeline. Describe your goal in plain English - the AI does the settings layer for you.
What's in This Guide
30-second summary
You type something like "Instagram-ready WebP, strip location data" into the Mochify tool. Mistral Small 3.2 parses that, chooses the right flags (format: webp, resize to 1080x1080, EXIF strip: on), and passes them to our squish engine. Your image comes back optimized exactly as described — without you touching a single dropdown.
What is Mochify's natural language compression feature?
Instead of manually selecting formats and settings, you describe what you want and Mochify figures out the technical details for you. We've connected an LLM — specifically Mistral Small 3.2 — directly into the compression pipeline on the main tool at mochify.xyz. It reads your intent and translates it into the right squish flags before processing your image.
This sits inside the existing tool interface. No separate mode to switch to, no new account needed. Type your description, upload your image, and the AI handles the settings layer.
How the AI picks your compression settings
The core idea is simple: your plain-English description becomes a set of compression flags. Mistral Small 3.2 parses the prompt, identifies what you need — format, dimensions, metadata handling — and calls our squish engine with those exact parameters.
Here's what that looks like in practice. "Best for the web" might produce --format webp --quality 82. "Tiny for mobile" could produce --format avif --quality 65 --resize 800. "Blog-post-ready, clean upload" might add --strip-exif on top of a balanced WebP output. The LLM doesn't guess randomly — it maps your language to a defined set of flags the squish engine understands.
Format selection: WebP, AVIF, JXL, or JPEG
The AI can choose from four output formats depending on how you describe your goal. AVIF delivers roughly 45% smaller files than JPEG at equivalent visual quality — making it the right pick when file size is the priority. WebP hits the sweet spot between compression and compatibility, supported by all major browsers since 2023 (caniuse.com). JPEG XL (JXL) is the choice for lossless or near-lossless archival. Standard JPEG via our jpegli encoder remains the best option when universal compatibility is non-negotiable.
| Prompt hint | Likely format output |
|---|---|
| "Best for the web" / "web-ready" | WebP |
| "Tiny," "mobile," "smallest possible" | AVIF |
| "Lossless," "archive quality" | JPEG XL |
| "Compatible everywhere," "email" | JPEG (jpegli) |
| "Instagram," "social" | WebP or AVIF |
The AI cross-references your described use case against format strengths. You don't need to know what AVIF is — just describe what the image is for.
Smart resizing and cropping from plain English
Dimension hints in your description translate directly to resize and crop flags. "Instagram square" tells the model to target 1080×1080 with appropriate cropping. "Thumbnail banner" produces a landscape crop scaled to a web-friendly width. "Full-size but compressed" keeps the original dimensions and only adjusts quality and format.
This removes the most common source of friction for content teams: looking up the "right" dimensions for every platform every time. The model maintains an understanding of common publishing contexts — social formats, email headers, blog post images — and maps those to real pixel dimensions before calling /squish.
EXIF stripping on request
Any prompt mentioning privacy, clean uploads, or sharing publicly will trigger EXIF metadata removal. Camera model, GPS coordinates, shooting time — all of it gets stripped before the file is returned. (For more on why this matters, see our EXIF risks guide.)
Importantly, EXIF stripping only activates when relevant to your prompt. If you're compressing personal archive images and don't mention privacy, the metadata is left intact. The AI respects your intent rather than applying blanket rules.
How to use it right now
- Go to mochify.xyz
- Upload your image (or drop it in)
- In the natural language field, describe your goal — a single sentence is enough
- Hit squish — the AI handles format, size, and metadata settings automatically
- Download your optimized image
No account required. Zero retention — your image is processed in memory and never stored.
Cheat sheet: prompt examples and what they produce
| Your description | Format | Resize | EXIF strip |
|---|---|---|---|
| "Instagram square, web-ready" | WebP | 1080×1080 | No |
| "Tiny for mobile, strip location" | AVIF | ~800px wide | Yes |
| "Blog post hero, fast load" | WebP | ~1200px wide | No |
| "Email thumbnail, clean" | JPEG (jpegli) | ~600px wide | Yes |
| "Archive quality, no compression" | JPEG XL | Original | No |
| "Social banner" | WebP | ~1200×630 | No |
FAQ
What AI model does Mochify use for natural language compression?
We use Mistral Small 3.2, called via the Mistral API. It parses your prompt and determines the appropriate compression flags before passing instructions to the squish engine.
Do I need to write precise technical descriptions?
No. Casual descriptions like "make it small for my blog" work well. The model is built to handle vague intent and translate it into specific settings.
What formats can the AI choose from?
WebP, AVIF, JPEG XL, and JPEG (encoded with jpegli for better efficiency). The choice depends on what you describe.
Does the AI always strip EXIF data?
Only when your prompt indicates a privacy or clean-upload intent. If you don't mention it, metadata is left untouched.
Is my image stored after processing?
No. Mochify processes images in memory only — zero retention. This applies whether you use the standard tool or the natural language feature.
Can I use this for batch compression?
The natural language feature on the homepage applies the same settings to your upload. For high-volume batch workflows, see our CLI and MCP options.
What if the AI picks the wrong format?
You can specify the format explicitly in your description ("give me a WebP") and the model will follow that instruction directly.
Try it yourself — just describe what you need
No account. No dropdowns. Type your goal, upload your image, and let the AI handle the rest.
Start Optimizing Free