The editor
Every generated thumbnail opens in an editor at /edit/<id>. It is a stage on the left showing the image, and a tabbed panel on the right holding every change you can make. There is no timeline, no layer stack to learn, and no export settings — the whole thing is built around "the image is nearly right, fix this one thing".
The seven tabs
| Tab | What it does | Plan |
|---|---|---|
| Mood | Ten colour grades — Cinematic, Horror, Anime, Retro 80s and more | Starter+ |
| Scene | Ten background treatments layered into the existing scene | Starter+ |
| Expression | Eight facial expressions, identity preserved | Starter+ |
| Lighting | Six relights — golden hour, neon, spotlight, rim | Starter+ |
| Avatar | Face swap, pop-out, position and size, remove person | Pro+ |
| Text | AI rewrite, find & replace, and free text layers | Starter+ |
| Custom text chat | Any change, described in plain English | All plans |
Locked tabs are visible rather than hidden, with a lock icon, so you can always see what a tier would give you. The custom chat tab is never locked — on any plan, including Free, you can ask for any change in words.
Two kinds of edit
Understanding this split explains most of the editor's behaviour:
- Generative edits — moods, scenes, expressions, lighting, face swap, text rewrites, custom instructions. These send the image back to the model and return a new image. Each one costs one request, takes a few seconds, and is undoable.
- Layer edits — text layers and popped-out image layers. These live on top of the image, are edited instantly in the browser, cost nothing, and are flattened into the file at download.
The practical consequence: do your generative edits first, then your layer work. Layers survive generative edits, but it is much less confusing to settle the artwork before you position type on it.
Click-to-edit is on by default
Click anywhere on the image and the editor targets that spot for your next instruction. It is the fastest way to change one specific part of a frame. See click to edit.
A workflow that wastes fewer credits
- Fix identity first. If the face is wrong, swap it before anything else. Every later edit inherits whatever face is in the frame.
- Then composition. Move or resize the subject, or remove them. Big structural changes before small ones.
- Then mood and light. A grade applied to the final composition looks coherent; one applied before you move things does not survive the move.
- Then expression. Small, and easy to lose if you keep regenerating over it.
- Then text. Baked text via find & replace, or free layers on top.
- Download last. Original resolution, no crop. See downloads and exports.
Undo, autosave and losing nothing
Every generative edit pushes the previous image onto an undo stack, and the whole draft — image, text layers, cutouts, history — autosaves to your account about a second after you stop working. Close the tab, come back tomorrow, and the editor offers your draft back. Detail in autosave and drafts.
Notes on behaviour
- Edits compound. Six stacked generative edits drift further from the original than one. If you find yourself six deep, it is usually cheaper to undo to a good state and make one well-worded custom edit.
- One edit at a time. The panel disables while an edit runs. This is deliberate — concurrent edits on the same image would race each other.
- The stage is a preview. Text layer sizes are percentages of the frame, so what you see is proportionally what you get at full resolution.