Single image edit

The Single image edit mode changes specific parts of a photo while keeping everything you do not explicitly point to untouched.

It is the most controllable workflow in the app: you can describe the change with a prompt and, when you want maximum precision, guide the AI with the Markup Editor (masks and annotations drawn directly on the image).

The idea is simple:

1. upload a photo,
2. say what to change,
3. indicate where with markup (optional),
4. generate the result,
5. iterate until you reach the final version.

Use this mode when you want control and precision, for example:

• local retouching (skin, outfit, product, details);
• removing distractions (cables, stains, objects in the background);
• precise additions (small objects, details, micro corrections);
• local color and light fixes;
• product/e-commerce cleanup (dust, light scratches, imperfections).

• 1 image (PNG, JPG, GIF up to 10MB).
• A prompt (recommended, almost always).
• Markup/mask (optional, but recommended if the change is local).

When you start processing, the app sends to the model:

• the image (possibly resized/converted to be processed correctly);
• your prompt (if present);
• and, if you created markup, an overlay with your instructions (in one of the formats described below).

The model returns a new image. The goal is an output where:

• the requested change is applied in the right areas;
• the rest of the image is not “rewritten” unnecessarily;
• light, shadows, perspective, and texture remain believable.

In this mode you typically see:

• Mode: workflow selector (make sure it is on Single image edit).
• Image upload: area to drag or select the file.
• Prompt: text field to describe the change.
• Add markup / Edit markup: opens the Markup Editor.
• Markup mode: how to send the overlay to the model (Separate overlay / Embedded overlay / Dual images).
• Force markup instructions: makes the prompt optional and gives priority to markup (useful if you want to work with markup only).
• Edit: generate the result.
• Generate variants (2-4): create alternatives for the same request (great for choosing the best take).
• Use as input: use the output as the new starting image for a second pass.

To avoid unwanted changes and get repeatable results:

1. Write a prompt with goal + constraints.
2. Add markup only where needed (small, but precise).
3. Run a first generation.
4. If needed, generate 2-4 variants and pick the best.
5. Use Use as input to refine with a second pass.

Open the “Mode” selector and choose Single image edit.

You can:

• drag the file into the upload area, or
• click and select the file from your computer.

Tip: if the image has very small text or important tiny details, use a high-quality source (avoid compressed screenshots).

A complete prompt always says two things:

• what to change (goal);
• what not to change (constraints).

Useful constraint examples:

• “Keep identity, face, and expression unchanged.”
• “Do not change pose, framing, or overall lighting.”
• “Do not change global colors or white balance.”
• “Do not add text or watermark.”

If the change is local or delicate, use markup:

• click Add markup (or Edit markup);
• highlight the target areas;
• add short text labels near the areas.

Then save/apply markup to return to the editor.

You have three options:

• Overlay (separate image): send clean image + transparent overlay.
• Overlay embedded in the image: send one image with markup drawn on top.
• Dual images (clean + overlay): send the clean image plus a reference with embedded overlay.

If you enable Force markup instructions:

• the prompt becomes optional;
• the model still receives a sample/instruction that pushes it to follow markup.

Use it when:

• you want to work mostly “by brush” with few text instructions;
• you notice the model often ignores the overlay.

Click Edit and wait for the output.

You have three paths:

• Perfect -> download/save and you are done.
• Almost right -> click Use as input and do a second pass (usually the most effective).
• Want alternatives -> use Generate variants (2-4), then pick the best and use Use as input.

Markup communicates to the AI:

• where to intervene (highlighted area);
• what operation to perform (text labels);
• any priority (e.g. “keep” on a zone that must remain identical).

• Brush: paints the target area (where you want intervention).
• Eraser: fixes markup.
• Shapes: quickly define zones (circles/rectangles).
• Text: often the most powerful part: write short, operational instructions.

• Mark only what you want to change.
• If the area is large, split it into zones with different labels (e.g. “remove” vs “repair”).
• Avoid covering details you want to preserve (especially faces and logos).
• If the goal is to remove an object, circle it and write “remove”.

• Remove an object: circle the object + text “remove”.
• Fix a defect: brush the defect + text “repair / cleanup”.
• Change a local color: brush the area + text “color: dark blue, keep texture”.
• Add a detail: highlight the spot + text “add ___ here, realistic”.

The model receives:

• the clean photo (to edit);
• a transparent overlay with markup (guide only).

Why it works well:

• the “real” image stays clean;
• the overlay is clearly a guide;
• it often reduces the risk of markup traces in the output.

The model receives a single image with markup drawn on top.

Pros:

• the AI cannot ignore it easily.

Cons:

• if you ask for complex changes, there is more risk of artifacts or “copying” overlay parts.

When to use it:

• if the separate overlay gets ignored;
• if you need a very obvious visual instruction (edges/areas).

The model receives:

• the clean photo (to edit);
• a second image with embedded overlay (reference).

Pros:

• combines the clean photo with a very clear overlay reference;
• useful when you have many zones or multiple instructions.

1. Goal: one clear sentence.
2. Constraints: short but explicit list.
3. Quality: realism, coherence, no artifacts.

Base template “Goal: [WHAT TO CHANGE]. Constraints: keep [WHAT NOT TO CHANGE] unchanged. Realistic output, no artifacts, consistent light and texture.”

Template with markup “Follow the markup as a guide. Inside marked areas: [ACTION]. Outside marked areas: do not change anything. Keep identity and unmarked details.”

Product/e-commerce template “Clean the product in marked areas: remove dust/scratches while preserving original shape and texture. Accurate colors, no invented elements, professional result.”

If you use different colors in markup, declare it: “Markup legend: red=remove, blue=add, green=keep.”

Use it when:

• you want alternatives (e.g. 3 interpretations of the same change);
• you want the version with fewer artifacts;
• you want to explore multiple solutions without rewriting the prompt.

This is the most important function for a studio-grade workflow. Use it when:

• the result is almost perfect but needs refinement;
• you need to fix small artifacts;
• you want a second step (remove first, refine after).

1. Pass 1: do the main change (prompt + markup).
2. Pass 2: “Use as input” and refine edges/texture/color.
3. Pass 3: if a defect remains, markup only the defect + minimal prompt.

• Markup is available only in this mode.
• If the prompt is too generic, the AI may change more than needed: always add constraints.
• If the selected model supports advanced controls (ratio/resolution), you will find them in the editor when available.

• Add constraints: “Outside marked areas do not change anything”.
• Reduce markup (smaller and more precise).
• Try Separate overlay or Dual images if the overlay seems ignored.

• Ask for “realistic, natural textures, no oversharpen, no plastic skin”.
• Do a second cleanup pass with Use as input.

• Highlight artifacts with markup and write “clean up edges, rebuild texture”.
• If the area is large: remove first, then refine (two passes).

This can be due to policy blocks (sensitive/copyrighted content) or ambiguous requests.

• Simplify the prompt.
• Avoid requests involving sensitive or protected content.
• Try a different photo or a less drastic change.

In Single image edit you can also use Control Map when prompt + markup are not enough.

• you want stronger pose/structure consistency;
• you need tighter composition control;
• you want fewer structural changes across variations.

1. enable Control Map;
2. choose Canny (edges) or Depth (depth);
3. generate/use the map;
4. run Edit.

Note: you can use “control-map-only” mode when you want map guidance to dominate.

In Single image edit, Batch mode is not available.

Batch is available only in:

• Enhance photo quality
• Restore photo

Recommended high-volume workflow:

1. run batch in Enhance/Restore;
2. refine individual outputs in Edit.