I'm Andy Zhang. In 2025 I started ImgEdify as a solo founder. It's my product, so I'll say that before I go any further.
I didn't set out to train a new image model. I was just sick of a dumb workflow. I'd write a prompt in one tab, throw the same photo at a background remover in another, then go looking for a resizer because the export was 4000 pixels wide and some form wouldn't take it.
So I built a browser workspace for that job. You can start from a written idea or from a photo, pick a model, see the credit cost before you create, and download the result.
Start with a prompt, or start with a photo
If you need something that doesn't exist yet, you describe the subject, lighting, and style, pick a model, and generate from text. If you already have a photo, you upload it and describe the change. Replace a background, pull out a distraction, recolor a jacket, turn a portrait into an illustration.
The AI image editor is that second path. You can add up to six JPG or PNG references, 5 MB or less each. Then you write what should change and what should stay put, pick a model and an aspect ratio, and check the cost on the Create button. It generates a new file. It doesn't overwrite your original, which sounds obvious, but I've used tools that get this wrong.
If the first pass is close, you can load the result from history and describe the next change. I use that loop a lot. Sometimes the second pass makes things worse, so you go back. I know 5 MB is tight for some phone photos, which is part of why the local compress and HEIC tools are there.
Show the price before anyone hits Create
I don't like generate buttons that hide the price until after the request is already gone.
ImgEdify uses one credit balance for image generation and editing. New accounts can get up to 60 free credits when you sign up, plus a daily +5 reward in the editor. The button shows the selected model's cost before you submit. You can set up a prompt first, but you need an account before the AI request goes out. If a task fails before it produces a result, it shouldn't be charged.
Getting every model's cost onto that same button was more bookkeeping than I'd expected. Each model prices differently, and the quote has to match what you actually get billed. I'd rather wrestle with that than surprise someone after they click. You can start for free and upgrade later if you need more credits, longer videos, or watermark-free downloads.
Not everything needs a model
AI is the right tool for "change the jacket but keep the couple and the street scene." It is a ridiculous tool for "make this 1920 pixels wide" or "turn an iPhone HEIC into a JPG so a form will accept it." Those jobs shouldn't cost credits, and they shouldn't require an upload to my servers.
So the file-prep tools run in the browser. No AI account. The file stays on your device. You can resize (and lock the aspect ratio), compress, convert HEIC or HEIF to JPG, or export WebP.
Doing HEIC conversion locally, instead of just accepting the upload and converting on a server, was the more annoying path. I still think it's the right one for this kind of utility. There's also an image-to-pixel-art tool that works the same way, all in the browser with Canvas: a pixel grid, a numbered worksheet, and a color key. I wanted a sprite or craft reference, not another filter that still sends your photo somewhere.
The other tools
Around the main editor there's a background remover that gives you a transparent PNG, plus white-background and transparent-background versions. You can remove a background for free. Higher-resolution downloads may use credits. There's face swap, including a multiple-face version for group photos, and text-to-video or image-to-video if you want a short clip.
Animate a Drawing is narrower on purpose. You upload a sketch or a photo of a drawing, pick a dance, a wave, a hop, or a float, and download a short crayon-style MP4. Settings are free to poke at; generating the video needs a sign-in and credits. I like the constraint. A blank "make a video" box is more flexible, and also a lot easier to get lost in.
The sticker generator can make a PNG sheet from a photo or a written character, or animated GIF reactions. You can try the demo and export an existing sheet without an account. AI sheets use credits, and free accounts get a watermark. Native GIF packs need a paid plan. Some people won't love the watermark. I get it. It's the line I drew for now.
What's actually going on under the hood
I'm not going to pretend I trained a secret foundation model in my garage. ImgEdify sits in front of third-party AI models, and the settings and costs depend on which one you pick.
Text-to-image models make a new picture from a prompt. Instruction-following models take one or more reference images plus a written change, then generate a new image from both. That's why faces, hands, and text can still drift if you run it again. Most of these systems are built on large diffusion or multimodal image models, which is the general approach, not a private recipe for any model on the site. Results vary. Look at the output before you publish it.
Background removal is more like segmentation: find the subject, cut it out. Hair and fur are still the annoying cases. The local tools use the browser's Canvas, plus client-side decoding for HEIC, so a resize or format conversion doesn't have to leave the device.
Commercial use depends on the model you picked, what you uploaded, and applicable law. Read the model's terms, and don't upload reference images, brands, or recognizable people you don't have rights to.
If you already live in a coding assistant
There's a remote MCP integration you can add to Cursor or Claude Code with a personal API key. The assistant can list models, check your balance, import a reference, quote a job, and only start generation after you approve it. Quotes and status checks don't use generation credits.
I also limited an account to one active generation at a time. If you wanted a queue, this will feel strict. I did it anyway, because accidentally starting the same paid job twice is worse. Direct ChatGPT connections aren't supported yet, or any client that needs OAuth. I'd rather just say that than promise a connector I haven't shipped.
If you want to try it
Open it in the browser, start with a prompt or a photo, and look at the cost before you create. New accounts can get free credits for the AI tools. The local file-prep stuff is free and doesn't need an account.
If a tool felt half-finished, or a model you wanted isn't there yet, tell me. That's the feedback I actually ship from.
Top comments (0)