Introducing ChatPic
I'm building ChatPic because most photo editing software makes you learn its menus before it lets you edit a photo.
The problem
I spent years around photographers while building Snapgle, and the pattern was always the same. The people with the best eye were rarely the ones who knew the software best. They knew exactly what they wanted, "less orange in the skin, keep the sunset warm," and then lost twenty minutes finding which of forty sliders did it, or gave up and slapped a filter on.
Lightroom and Photoshop are extraordinary tools built for people who've already paid the learning cost. Phone editors went the other way and hid everything behind presets. Neither one lets you say what you want and get it.
What it is
You describe the edit you want, "warm this up and brighten the shadows," and ChatPic does it. Then you keep talking: "a bit less," "only the sky," "undo the last one." The conversation is the history of the photo.
Why this is possible now
Two things changed in the last few months.
Language models got good enough at turning a loose sentence into a precise intent. ChatGPT made that obvious to everyone. Ask for "make it moodier" and you can get back something concrete: lower exposure, crush the blacks a little, cool the highlights.
Image models also learned to follow instructions. InstructPix2Pix, out of Berkeley in November, edits a photo from a written instruction, and ControlNet, released this month, gives far more control over what a diffusion model keeps and what it changes. A year ago "edit this photo the way I describe" was a research demo. Now it's a weekend of prototyping.
How it works
It's a mix of computer vision and a model that maps your request to editing operations, using the image and your past edits as context.
Most requests don't need generative AI at all. "Brighter," "more contrast," "crop to square" map onto the same operations every editor has had for twenty years, and those are exact, fast and reversible. ChatPic translates the sentence into those operations first. It only reaches for a generative model when the request needs one, like "remove the guy in the background."
I care about that split because it keeps the photo yours. I don't want a tool that invents a better picture than the one you took. Every edit ChatPic makes shows up as a step you can read, adjust with a normal slider, or throw away, so if you already know what you're doing, the conversation is a shortcut you can ignore whenever you like.
Where this goes
Today the bet is narrow: one photo, one conversation. If it works, the same idea stretches further. Edit fifty photos from a shoot with one sentence. Save "the way I like my portraits" as something you can say instead of a preset you have to rebuild. Ask why a photo looks flat and get an answer, not just a fix.
Further out, I think most creative software ends up like this. The menus don't disappear, but you stop needing them to start.
What I don't know yet
Plenty. Language is vague, and "make it pop" means something different to everyone, so ChatPic will need to ask follow-up questions without becoming annoying. Generative edits can still change things you didn't ask them to touch. And I don't know yet whether people who already know Lightroom will want this, or only the ones who never learned it.
Where it stands
Early. The waitlist is open at chatpic.online. I'm building this around what people actually ask for, not what I assume they want, so if you sign up, expect to hear from me, and expect your feedback to change what ships.