The short answer: upload one main image, add up to 9 reference images, and describe the change in one sentence: swap the background, restyle the whole frame, add or remove elements, or box a region to edit only that area. AI generates one new image based on the main one; each run handles a single image, and failures are free.
What one-sentence editing can do
This is the general-purpose editor with no preset buttons: you tell the AI what to change in one required paragraph:
| Scene | One-sentence example |
|---|---|
| Change background | "Put the subject on a rainy Tokyo street after the rain, keep the subject's lighting" |
| Change style | "Turn the whole photo into cyberpunk style, keeping pose and composition" |
| Add / remove elements | "Add a giant cat in the background" / "Remove the passer-by on the right and rebuild the ground" |
| Product staging | "Keep the product and its logo, move the white-background shot into a lifestyle scene with natural light" |
The tool page's four official examples cover exactly these: a giant cat in a street scene, a photo turned into an illustration, a passer-by removed, and a white-background product staged into real life.
Main image vs reference images
The multi-upload rule is simple: the first image is the main image; the next up to 9 are references.
- The main image is what gets edited; AI generates the new image from it, and only the main image can be boxed for local edits;
- References guide: they specify style, people, products or backgrounds; AI references their light, elements or composition but never pastes them directly into the result;
- Each reference can be cropped individually, uploading only the boxed part (say, just the sky from one photo, or one product);
- Your description can name them directly: "use reference 1's light and reference 2's background; change only the boxed area".
In one line: the main image decides what gets edited, the references decide which way it goes, and your description decides what it becomes.
How to write edit descriptions
The description is required and the only source of effect (there are no preset buttons). The formula: say what to change and what it should become, then list what must stay the same:
- ✅ "Replace the taxi on the right with a red sports car, keeping the other cars, road and lighting unchanged";
- ✅ "Keep the pose, expression and composition, and restyle the whole photo as a comic";
- ❌ "Fix this image"; too abstract, results unpredictable.
The more specific the description, the closer the result.
How to edit step by step
- Upload the main image + references: the first is the main one, the rest are references; up to 10 total; JPG / PNG / WebP, each up to 10 MB;
- Optional: box a region: box the area on the main image; AI changes only that part and leaves the rest untouched; without a box, the whole image is editable;
- Write the edit description: required; follow the formula above;
- Pick the output ratio: 12 presets (1:1, 4:3, 3:4, 16:9, 9:16 and more) + custom; defaults to the main image's ratio, output short side 512–2048;
- Download: compare and save. Billed by actual token usage, failures are free.
Four scenes in practice
- Add creative elements: a giant cat appears behind a street photographer, and perspective and lighting stay consistent:
The cat's perspective matches the street light; the photographer and crowd stay untouched
- Remove and replace: the taxi on the right becomes a red sports car, with other cars, road and lighting unchanged:
Only that one car changed; the street's mood and every other element stayed
- Change style: a real scene rebuilt as a cyberpunk museum, and the composition survives the restyle:
Composition and subject relationships kept; color, light and mood fully reworked
- Product staging: a white-background Polaroid moves into a lifestyle scene with plants, frames and natural light:
The product and its logo stay; background, lighting and shadows adjust together
Local edits and the limits
- Box to edit locally: after boxing a region on the main image, AI changes only that part; fixing one car or one patch of sky carries zero risk to the rest;
- References aren't collage: they guide style and elements; for true multi-asset compositing, describe the layout in your description;
- One image per run: each request produces one result with no batch queue, so submit a set image by image;
- Output ratios: 12 presets + custom; the main image is center-cropped to the target ratio then resized (multiples of 16, 1:3–3:1, long side ≤2560).
vs Photoshop AI and Canva AI
| Photoshop AI | AI image editor | |
|---|---|---|
| Getting started | Layers, masks, curves and Generative Fill | Browser upload + one sentence, up to 9 references |
| Core capability | Manual selections and local repainting, skill-dependent | A main image plus references, edited by description |
| Consistency | Manual work is hard to keep consistent | Descriptions and references keep similar edits stable |
| Environment | Desktop install, hardware-dependent | Runs in the browser, no install |
| Best for | Professional retouchers and design teams | Portrait retouching, product scenes, content visuals, creative composites |
| Canva AI | AI image editor | |
|---|---|---|
| Core capability | Templates and Magic Edit; complex work limited | A main image plus references, edited by description |
| Steps | Choose a template, edit, export | Upload, describe or box a region, download |
| Original preserved | Depends on templates and edit scope | Keeps the main image's composition, subject and proportions |
| Best for | Social posters, template design | Portrait retouching, product scenes, content visuals, composites |
In one line: Photoshop wins on fine control, Canva on template speed, and the general editor wins on "one sentence + references, for everything without a dedicated tool". For clear-cut jobs the dedicated tools are faster: background removal, background changing, blur, object erasing; each has its own entry.
Frequently asked questions
How does AI Image Editor work?
Upload the main image first, optionally add up to 9 reference images, box a region to change (optional), then describe the edit. AI generates a new image based on the main image.
What are reference images for?
They specify style, people, products or backgrounds. The first image is always the main image being edited; references are guidance only and are not pasted into the result.
Can I edit only part of the image?
Yes. Box a region on the main image and AI changes only that area, leaving the rest untouched. Without a box, the whole image is editable.
How is the output size decided?
The server computes a size close to the main image's aspect ratio: both sides divisible by 16, ratio between 1:3 and 3:1, longest side at most 2560. The main image is center-cropped to the target ratio, then resized; you can also pick from 12 presets or custom dimensions.
Can I batch process?
No. Each request processes one main image and produces one result. For multiple images, submit them one by one; within a single run, up to 10 images participate (main + references).
Do I get charged if it fails?
No. Credits are deducted only after the result is generated and saved; upload failures and upstream errors are free.
How do I change the background or style?
Upload the main image and describe the new scene or style, stating what to keep: subject, composition and lighting. Add up to 9 references for visual guidance.
How do I write a better edit description?
Say what to change and what the result should look like, then list what must stay the same, e.g. "replace the taxi on the right with a red sports car, keeping the other cars, road and lighting unchanged". Specific descriptions give predictable results.
Features and details reflect the tool as of November 12, 2026 and may evolve; the before/after examples are from the tool's real demo assets, with the cover composited from the original and illustrated halves.
Ready to edit with one sentence? Eyesme's AI Image Editor takes a main image plus up to 9 references and changes only what you describe; for clear-cut jobs, the dedicated tools (background removal, portrait blur and more) get there faster.




