Magnific is the old Freepik under a new name: a stock library of hundreds of millions of assets, the Magnific upscaler and a hub that resells other labs' image, video and audio models for credits, with a commercial license on what you make. None of that fits in a weekend project, since the library and license are content you don't own and the generators run on other companies' hardware. What an agent can build is the small slice many people actually use: a local upscaler on open ESRGAN-family models that enlarges faithfully but does not invent detail the way Magnific's creative mode does. If upscaling is all you need, Upscayl already does it for free.
Before you buildMagnific's Free plan gives up to 20 AI images a day, and Upscayl is a free desktop app for upscaling. Try both before building anything.
Build me a local image upscaler and enhancer to replace Magnific's upscaling for my own photos. Requirements: - Python 3.11 with FastAPI and a single-page front end in vanilla JS (no build step). `python app.py` serves http://127.0.0.1:7890 and must not bind to any other interface. - Models: load ESRGAN-family upscalers (.pth or .safetensors) through `spandrel` and `torch` from ./models/. `python fetch_models.py` downloads the models listed in models.json (one 4x general model and one 2x photo model) and checks each SHA-256 sum. Use CUDA, then Apple MPS, then CPU, automatically. - Processing: split large images into overlapping tiles (default 512 px with 32 px overlap, blended) so a 24-megapixel photo doesn't run out of memory; cap output at 16384 px on the long edge; keep EXIF and the ICC profile; write PNG or JPEG. - Upscale page: drop one or many images, choose 2x or 4x and a model, optional sharpen and denoise sliders (Pillow), a before/after slider with synced zoom and a queue with progress per tile. - Storage: sqlite3 at ./data/upscales.db (source name, model, scale, seconds, output path); outputs in ./data/out/YYYY-MM/; a gallery with file-name search and "reuse settings". - Batch: `python upscale.py <folder> --scale 4` writes next to each file with an _x4 suffix and skips finished ones. - Single user. No accounts, no telemetry, local-first; the only outbound calls are the model downloads. - Out of scope: creative or prompt-guided upscaling that invents detail, face restoration, video, image generation, stock images, sharing. - README: GPU and memory needs, the measured seconds for a 12-megapixel image at 4x on this machine, which model licenses allow commercial use, and an honest note that results are faithful enlargements and will not match Magnific's creative mode.
Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.
The first prompt builds the core. Run these one at a time in the same folder to grow it into your own Magnific AI.
Extend the upscaler in this folder with model comparison. Requirements: - On the Upscale page, tick up to three models and run each on a 512 px crop the user draws on the image. - Show the crops side by side with the original and the seconds each took. - A "use this model on the full image" button queues the full job with the chosen model. - Cache crop results by image hash, crop box and model so re-runs are instant. - Update the README.
Add a folder watcher to the upscaler in this folder. Requirements: - `python watch.py ~/Pictures/to-upscale --scale 2 --model <name>` upscales each new image once it has finished copying (stable size for 2 seconds) and writes it to ~/Pictures/upscaled/. - Skip files that already have an output, and log each job to the same SQLite database as the web page. - Handle HEIC and WebP input by converting to PNG first. - Add a launchd example to the README. - Update the README.
Add a print-size helper to the upscaler in this folder. Requirements: - On the Upscale page, take a target print size in cm or inches and a DPI (default 300), and show the pixel size needed, the scale factor from the current image, and whether 2x or 4x gets there. - Warn when the result would be below 150 DPI at the chosen size. - Add presets for A4, A3, 8x10 in and 16x20 in. - Add unit tests for the arithmetic. - Update the README.
One bill covers a stock library, a stack of competing generators and the upscaler people came for, with a license that lets agencies use the results for clients. Keeping up with every model's own API is the work the subscription takes off your hands.
Estimate · Free on a GPU or Mac you already own, apart from electricity. Renting a GPU by the hour is the only recurring cost and adds up with heavy use; light use favours the $20 Premium plan or a free desktop app. The build is token usage on your agent plan.
| Plan | Monthly | Annual, per month | What you get |
|---|---|---|---|
| Free | $0 | $0 | Up to 20 AI images a day with the in-house model; video, audio and advanced editing aren't available or are restricted. |
| Premium | $20 | $14.50 | 20,000 credits a month (240,000 a year), the stock library, Spaces, a commercial AI license, the Magnific upscalers, API access and ComfyUI plugin. |
| Premium+ | $45 | $33.75 | 45,000 credits a month (600,000 a year), unlimited generations on most image models, extra credit packs, music rights in the license. |
| Pro Starter | $110 | $82.50 | 112,500 credits a month (1.5 million a year), everything in Premium+. |
| Business | custom | — | 45,000 credits per seat a month, pooled by an admin, with unlimited generation on most image models. |
Free gives limited daily AI use, up to 20 images a day with the in-house model, and no credit balance; stock downloads stay at free-tier limits.
Not really. One bill covers a stock library, a stack of competing generators and the upscaler people came for, with a license that lets agencies use the results for clients. Keeping up with every model's own API is the work the subscription takes off your hands. The prompt on this page builds the closest honest substitute.
About $20 a month on the Premium plan, $240 a year (checked 2026-10-19).
The stock library of photos, vectors, PSDs and videos, and the commercial license that comes with it; Many image, video and audio generators from other labs under one credit balance; Magnific's creative upscaler, which adds detail guided by a prompt, and the video upscaler; Spaces for team canvases, Brand Kit, the API, an MCP server and a ComfyUI plugin; Unlimited image generation on most models, on Premium+ and above. If any of those matter to you, keep paying.
Yes: Upscayl (open source), ComfyUI (open source), chaiNNer (open source).