Edit part of a picture with Mask AI

Image and photo · Intermediate · 14 min read · Updated

Select an area with a brush, wand, lasso or shape, describe what should change and let the AI rework only that spot, then compare, undo or carry the result on to the next module.

Mask AI with a teapot photo loaded and an ellipse mask drawn over it: selection tools on the left, mode tabs, model picker and the IDA Q buttons on the right

What Mask AI is and when to use it

Mask AI is masked editing. You load one picture, mark the part you want changed, say in a sentence what should be there, and the AI reworks only the marked area while the rest of the photo stays as it was. It covers the jobs that a global prompt gets wrong: swap one object, remove another, change a background, extend the frame, restyle a region, or change a colour without touching the rest.

Compared with Image Studio (where you upload a ready-made mask file), Mask AI is where the mask is made: you paint and shape it directly on the picture, stack several selections as layers, and run the edit in the same screen. Compared with the photo editor, it is generative: it invents new pixels instead of adjusting existing ones.

You want toMode
Put something new into a spotInpaint
Take an object out and rebuild what was behind itRemove object
Keep the subject, replace what is behind itChange background
Paint beyond the edges of the pictureExtend
Change the style of one regionStyle
Say in plain words what to changeInstruction edit

Loading a picture and the workspace

The start screen asks you to Choose an image (PNG, JPG or WebP, up to 20 MB) and offers a From library tab with your own pictures. A back link, Back to the generator, leads out. When the picture is loaded, the title bar shows its size in pixels and the workspace opens.

Mask AI with a teapot photo loaded and an ellipse mask drawn over it: selection tools on the left, mode tabs, model picker and the IDA Q buttons on the right
The Mask AI workspace with an ellipse mask drawn over the teapot. Tools on the left, the picture in the middle, mode, model, description and the IDA Q buttons on the right. Marked on the screenshot: Selection tools, add and subtract, undo, zoom; The mask: a blue overlay on the selected area; Mode tabs; Model picker with its price; Edit description; IDA Q Edit with its price.
  • Toolbar. Seven selection tools, then add and subtract mode, undo and redo (the history keeps 60 steps), clear, a mask preview toggle and zoom controls. The wheel zooms at the cursor, the space bar pans.
  • Mode. Six tabs that decide what the edit does.
  • Model. Which engine runs the edit. A badge says whether it respects the mask (MASK) or processes the whole picture (WHOLE IMAGE), and the price or Included in plan sits next to it.
  • Edit description. One or two sentences about what should appear in the selected area.
  • IDA Q. Two ready buttons that run FOTOhub's own queue models, with the price and a notice that they process outside the EU.
  • Mask layers, Edge softness and FOTOhub tools. Lower in the right panel: the list of your selections, one slider and a set of single-purpose tools.

How masks are made

A mask is the part of the picture the AI may change. In the editor it appears as a blue overlay; white on the exported mask means edit, black means keep. Every selection you make becomes a layer, so you can build the mask from several pieces and hide, refine or delete each one.

Screen recording in Mask AI: a brush stroke, a lasso loop and an ellipse are drawn on the teapot photo and the last shape is undone
Building a mask: a brush stroke, a lasso loop and an ellipse, then undo the last one. Each shape is its own layer.
ToolKeyUse it for
BrushDPaint the area to change. Size with the slider or Alt plus the wheel; the size is measured on the photo, so zooming does not change it
Auto-selectClick an object and the AI finds its edges: left click adds, right click subtracts
Magic wandSelect similar colours: a flat wall, a sky. Tolerance decides how different a shade can be; start at 32. Turn off contiguous to catch every similar patch in the photo
LassoLTrace an irregular outline freehand; the shape closes itself
RectangleA box: a banner, a crop, a whole section
EllipseRound things: faces, plates, a bowl
GradientGA soft transition: drag a line, the start is no change, the end is full change. For skies and vignettes
  • Add and subtract. The plus button (or the E key) switches between adding to the selection and cutting away from it. Use subtract when a selection spilled over an edge.
  • Edge softness. One slider, 15 percent by default. Zero is a hard cut; raise it for hair and fur, lower it for objects.
  • Stroke strength. Below 100 percent the brush builds a selection up gradually in layers.
  • Mask expansion. Grows the mask by a few pixels before it reaches the model. Helps when a rim of the old image remains at the edge of an edit; 0.01 to 0.05 is usually enough.
  • Per-layer description. Each layer can carry its own text. Generating all layers sends a separate request per layer, so the cost grows with the number of layers.

Modes, models and what they cost

ModeNeedsWhat happens
InpaintA mask and a descriptionFills the selection with what you describe; the rest is kept
Remove objectA maskRemoves the object and rebuilds the background; no description needed
Change backgroundSometimes a maskReplaces the background and keeps the foreground; some engines find the background themselves
ExtendDragging the blue handles at the edgesPaints beyond the frame, up to 2000 pixels per side
StyleA mask and a style descriptionRestyles the selected area; describe the style, not the content
Instruction editAn instruction; a mask is optionalFOTOhub AI edits by instruction; with a mask only that area changes

The model list changes with the mode. Two things are worth knowing before you choose one. A MASK model changes only the selection. A WHOLE IMAGE model re-renders the whole picture and treats your selection as a hint, so it can alter areas you did not mark; pick MASK models when the rest of the photo must stay exactly as it is. And every model shows its price, or Included in plan, next to its name, and the Generate button repeats it, so you know before you press.

  • FOTOhub Studio and FOTOhub Remover run on FOTOhub's own GPUs. Studio is listed as included in paid plans, Remover removes an object without a prompt at a low price. When the GPUs are busy or unavailable the screen reports an error instead of a picture; see the troubleshooting section.
  • IDA Q Edit changes only the selected area, following your description. It runs in a queue on servers outside the EU (the screen says so and asks you to accept it) and the price is shown on its button.
  • IDA Q Backdrop replaces the background from a description; it cuts the subject out itself, so a selection is not used.
  • FOTOhub AI powers Instruction edit. It offers a fast mode (about 45 seconds) and a best-quality mode (about 6 minutes), with the cost shown for each.
  • The picker also lists partner models. They are labelled MASK or WHOLE IMAGE like the rest.

Your first edit, step by step

  1. Load the picture Choose an image or take one from your library. The size in pixels appears next to the title.
  2. Pick the mode Click Inpaint to put something new into a spot, Remove object to take something out.
  3. Mark the area Choose a selection tool and mark the spot. Make the mark a little bigger than the object.
    Mask AI with a teapot photo loaded and an ellipse mask drawn over it: selection tools on the left, mode tabs, model picker and the IDA Q buttons on the right
  4. Describe the change Write what should appear in the selected area. Describe the object and the light, not the whole photo. For removal, leave the field empty.
  5. Run it Press the Generate button of the model, or Edit selected area under IDA Q. The button carries the price. IDA Q Edit runs in a queue and can take several minutes.
  6. Check the result and keep going The picture on the canvas is now the result and the title shows Iteration 2. Press Before to see the original, Undo edit to go back, or edit the result again: every new edit starts from the latest picture.
    The Mask AI canvas after an edit with the Before and Undo edit buttons under the picture

Replace the teapot with a cactus (and see a ghost)

Goal. Swap the teapot for a plant with IDA Q Edit.

Input you can copy.

Prompt: a small potted cactus in a terracotta pot
SettingValue
ModeInpaint
MaskOne ellipse over the teapot body
EngineIDA Q Edit
The teapot area after IDA Q Edit with the prompt a small potted cactus: two cacti in terracotta pots on the linen, with a faint rest of the spout on the left

Why it works. Two cacti stand on the linen in the same side light. The ellipse did not cover the tip of the spout, so a ghost of it remains on the left. The mask was the problem, not the prompt: widen the selection until it covers the whole object with a margin, or add a second shape over the spot, then run again.

A bigger mask, a different surprise

Goal. Cover the whole teapot with a generous rectangle and ask for something quite different.

Input you can copy.

Prompt: a stack of three blue ceramic bowls
SettingValue
ModeInpaint
MaskA large rectangle around the teapot
EngineIDA Q Edit
IDA Q Edit with a large rectangle mask and the prompt a stack of three blue ceramic bowls: two blue bowls stacked with the teapot body still sitting on top

Why it works. The spot is filled with blue bowls, but the AI built two, not three, and kept the teapot body sitting on top of them, as if it were part of the scene to respect. When you want an object fully replaced, run Remove object first, then Inpaint into the empty spot. Two short steps are more predictable than one big change.

Instruction edit: say what to change

Instruction edit trades the mask-and-describe routine for a plain instruction, in the manner of telling a retoucher what you want. The mask is optional: with one, only the marked area changes, without it the whole picture follows the instruction. The panel shows two modes, Fast (about 45 seconds) and Best quality (about 6 minutes), and the cost of each.

Mask AI in Instruction edit mode with a rectangle mask over the tablecloth, the FOTOhub AI model, the instruction in the description field and the Generate button
Instruction edit with a rectangle mask over the tablecloth and the instruction typed in the description field.

Recolour the tablecloth

Goal. Change one surface to another colour while leaving the teapot alone.

Input you can copy.

change the linen tablecloth to deep navy blue, keep everything else the same
SettingValue
ModeInstruction edit
MaskA rectangle over the lower part of the picture
EngineFOTOhub AI
Instruction edit with a rectangle over the lower part of the picture: the tablecloth in front is now navy blue while the teapot is untouched, with a hard edge where the mask ended

Why it works. The front of the cloth is now navy and the teapot did not change, so the mask did its job. The result also shows the rule of masks: where the rectangle ended there is a hard edge, and the cloth behind the teapot stays pale. Paint the whole surface you want to change, with the lasso or the wand on the cloth, and the colour will be even.

Results, comparison and next steps

  • Before shows the original, Undo edit brings the previous version back.
  • Iteration N in the title counts how many edits stack up on this picture.
  • Results at the bottom of the right panel lists every result of the session as small thumbnails.
  • Continue with… opens the shared menu of next steps for the result, such as sending it to the photo editor, Design Studio or image to video.
  • New at the top loads a different picture.

Limits, price and privacy

TopicHow it works
Start imagePNG, JPG or WebP, up to 20 MB
PriceShown next to each model and again in the button that runs it; Included in plan appears for the FOTOhub model that is part of paid plans
Several layersGenerating every layer sends one request per layer, so the price grows with the layers
Queue modelsIDA Q Edit, IDA Q Backdrop and Instruction edit run as queued jobs. IDA Q Edit took about four and a half minutes in our test; Instruction edit Fast about 45 seconds
Where it runsIDA Q buttons say Processed outside the EU and ask you to accept it. Instruction edit says it runs on servers in the United States. Other models follow their own notice in the picker
ExtendUp to 2000 pixels per side

Troubleshooting and FAQ

Auto-select shows Auto-select failed.
Press Retry. If it keeps failing, select with the Magic wand, the Lasso or a shape instead; they ran fine when we tried them.
An FOTOhub GPU tool answers Failed to fetch or shows a technical error.
The GPU service did not answer. Nothing was produced. Try again later, or use IDA Q Edit or Instruction edit, which run on a different path.
A ghost of the old object is still visible.
The mask missed part of it. Widen the selection with a margin, raise Mask expansion a little, or run Remove object first and then Inpaint.
The model changed things outside my selection.
You probably picked a WHOLE IMAGE model, which treats the selection as a hint. Choose a MASK model, or Instruction edit with a mask.
Why is there a hard line where my mask ended?
With Edge softness at zero or a rectangle that stops mid-surface, the change stops abruptly. Select the whole surface and keep Edge softness above zero.
Do I need a description for Remove object?
No. Mask the object with a little margin and run. The description is optional there.
Can I keep working after a result?
Yes. The result becomes the current picture and the next edit starts from it. Use Before and Undo edit to compare and step back.

Open Mask AI

Open Edit part of a picture with Mask AI in FOTOhub