toolspool
toolspool
Open weights vs local · 15 tools

Open Source AI Image Generation Tools: Three Different Things

Published July 20, 2026

The phrase hides three different shopping lists

Type open source ai image generation tools into a search box and you will get back products that have almost nothing in common with each other. That is because the phrase is doing three jobs at once, and the person searching usually wants exactly one of them.

The three jobs, plainly:

  • Open weights. The model is published under a licence anyone can use — Stable Diffusion, FLUX.1, Z-Image Turbo, GLM-Image. The website you type your prompt into is still a normal commercial product with a subscription. Open model, closed business.
  • Software you can run yourself. A downloadable app that generates on your own machine, with your prompts and pictures never leaving it. This is the rarest of the three by a wide margin.
  • Free, no account, no card. Plenty of people say "open source" when what they mean is "I do not want to pay or sign up." That is a pricing question, not a licensing one, and it has completely different answers.

Conflating them costs you an afternoon. You sign up for something advertised as built on an open model, expect to download it, and discover it is a web app with a $19.90 tier. Sorting the image generation category by which of the three you actually care about is the whole trick.

Built on open models, sold as a web service

Eleven tools in the category describe themselves as running on open or openly licensed models. Nearly all of them are hosted front-ends: the weights are open, the product is not. That is a legitimate and often sensible arrangement — someone else pays for the GPUs — but it is not software freedom, and none of these give you the app's source.

ToolModel it builds onPricing
DreamStudioStable Diffusion — this is Stability AI's own team interface and APIFreemium
Flux 1 AIBlack Forest Labs' FLUX.1 (Schnell, Dev, Pro). A third-party site, not the official BFL serviceFreemium
FluxPictureFLUX.1 pro/dev/schnell plus Krea and Wan 2.2; images up to 2 megapixels, commercial use allowedFreemium — Hobby $11.90/mo, Pro $22.90/mo, Max $26.90/mo
Z-ImageZ-Image Turbo; text-to-image and image-to-image, usable without login for basic generationFreemium — Basic $9.90/mo, Professional $19.90/mo, Ultimate $39.90/mo
Z Image Turbo AIThe open 6-billion-parameter Z-Image Turbo model; English and Chinese promptsFreemium — Basic $19.90/mo, Pro $31.92/mo, Max $47.92/mo
GLM-imageThe Apache-2.0 licensed GLM-Image model from Z.AI, deployed on the operator's own infrastructurePaid — Starter $5 for 500 credits, Professional $29.90 for 4,200
FloyoRuns ComfyUI workflows in the browser on managed GPUs, mixing open models, custom LoRAs and proprietary APIsFreemium — free trial, then $12–$56 per user/mo
StickerBakerAn open-source web tool itself, running Replicate's sticker-maker and face-to-sticker modelsFree
TextWallpaperOpen-source wallpaper generator — typed text and emoji into a sized, coloured background imageFree
Kolors Virtual Try On AIThe open-source Kwai-Kolors virtual try-on project; upload a photo plus a garment imageFree
QRBTFAI and parametric QR code art, with an open-source parametric library alongsideFreemium — 10 AI codes free, Pro $20/mo

Two of those eleven are open source as products rather than merely open-model consumers: StickerBaker and TextWallpaper are each described as open-source web tools, application code included. QRBTF sits just short of that line — it publishes an open-source parametric library alongside its AI generator, which is an adjacent component rather than the product itself. All three are narrow in scope too — stickers, wallpapers, QR art. Breadth and openness rarely arrive together at this end of the market.

If you specifically want a browser front-end pointed at Stable Diffusion or FLUX, the Model Frontends (SD/Flux) subcategory collects 37 of them and is a better starting point than a general search.

Four that actually run on your own hardware

Here is the honest number: four tools in this category are described as running locally or offline. Not forty. Four. Anyone promising you a rich landscape of downloadable generators is describing software that exists in the wider world, not products listed here.

  • DiffusionBee — runs Stable Diffusion locally on a Mac, bundling text-to-image, image-to-image, generative fill, upscaling, variants and custom model training. It runs 100% offline, so prompts and images never leave the device. Priced Free.
  • Makeayo — a Windows desktop app that runs image models locally rather than in the cloud, with hundreds of community models, image-to-image, 4K upscaling and ControlNet pose matching. The base app is free; subscriptions from $12/mo add credits and priority rendering.
  • DrawingPics — a native macOS app turning sketches into images in real time as you draw, entirely offline on Apple Silicon via Core ML. Sold as a one-time licence: $9.90 for one Mac, $19.90 for up to three.
  • OpenNFT — a downloadable desktop app for composing layered NFT art collections without code, running offline in 16 languages, with a free tier and a PRO version.

Note what that list is not: general-purpose. Two are Mac-only, one is Windows-only, one makes layered collections rather than prompt-driven art. Self-hostable general image generation software certainly exists outside this directory — the well-known workflow builders and web UIs are community projects, not listed products, and are not covered here. Within the listed set, the local option is a short shelf.

What each route actually costs you

The trade is boring and consistent. A hosted front-end costs money per month and gives you zero setup: you open a tab and type. A local app costs nothing per month, keeps your prompts on your machine, and costs you a download, a configuration session, and whatever your computer can manage. Neither is smarter than the other — they price the same thing differently, in dollars versus your own time and hardware.

Two things worth holding onto when comparing open-source AI image generation tools on cost:

  • Freemium is not free. Seven of the eleven model-based front-ends above are Freemium, meaning a free entry point exists and paid tiers sit behind it. Only StickerBaker, TextWallpaper and Kolors Virtual Try On are priced Free outright. GLM-image is Paid with no free tier at all.
  • Open weights say nothing about your usage rights on a given site. Where commercial use is spelled out — FluxPicture and Z Image Turbo AI both state it — you can rely on it. Elsewhere, read the terms of the site you are using rather than the licence of the model underneath it.

Hardware is the one thing to reason about rather than look up. Running a model on your machine means your machine does the work; how fast depends on it. Nobody can tell you a number for your laptop from a directory listing, and any article that quotes you one is guessing.

If what you meant was simply "free"

This is the most common case. Someone wants pictures without a bill and reaches for the word "open source" as shorthand. If that is you, licensing is a detour — go straight to the pricing question, where the category has 138 tools priced Free and many more with usable free tiers. The rundown of genuinely free image generators sorts those by what the free tier actually permits.

And if openness was never the point but output quality was, two neighbouring pieces answer that directly: one on which generators hold up for photorealistic work, and one on generators that take reference images and custom styles — 140 tools in this category document accepting an uploaded photo or reference image, which matters far more for control than any licence does.

The short version: decide which of the three questions you are asking before you open a single tab. Open source ai image generation tools is a phrase that answers all three and none of them.

FAQ

Are open source AI image generation tools actually free?

Not automatically. Open weights describe the model's licence, not the website's price. Of the eleven tools here built on open or openly licensed models, only StickerBaker, TextWallpaper and Kolors Virtual Try On AI are priced Free; seven are Freemium and GLM-image is Paid, starting at $5 for 500 credits.

Which image generators run entirely on my own computer?

Four. DiffusionBee runs Stable Diffusion 100% offline on a Mac and is free. Makeayo runs image models locally on Windows, with a free base app and subscriptions from $12/mo. DrawingPics runs offline on Apple Silicon for a one-time $9.90. OpenNFT is a downloadable offline app for layered NFT collections.

What is the difference between an open model and open-source software?

An open model means the weights are published for anyone to use — Stable Diffusion, FLUX.1, Z-Image Turbo, the Apache-2.0 GLM-Image model. Open-source software means the application's own code is published. Most products built on open models are ordinary commercial web services; StickerBaker and TextWallpaper are the ones that publish their own code, while QRBTF publishes an open-source parametric library alongside its AI generator.

Can I use images from these tools commercially?

Check the site's terms rather than the model's licence. FluxPicture states that commercial use is allowed, and Z Image Turbo AI describes its base model as free and open source for personal or commercial use. Where a site does not spell it out, the model's open licence does not settle the question for you.

Do I need a powerful computer to run one of these locally?

Running a model on your own machine means your machine does the work, so speed depends on your hardware rather than on a subscription tier. No fixed requirement applies across DiffusionBee, Makeayo and DrawingPics — DrawingPics targets Apple Silicon specifically, and the practical answer is to try the free options before buying anything.

Related articles