AI image editing uses a generative model to modify an existing image based on text instructions rather than manual pixel manipulation. Unlike traditional editing tools where you select regions with lasso tools and apply filters, text-based editing lets you describe the change: "replace the sky with a sunset", "remove the person on the left", "make it look like a watercolor painting". The model understands both the image content and the instruction, modifying only the relevant regions while keeping the rest intact. Multi-turn editing extends this — you can make sequential edits to the same image, refining the result in steps.
Describe edits in plain English: "change the wall color to sage green", "add a cat sitting on the windowsill", "remove the car from the background". No masking, no selection tools, no layer management. The model identifies what to change based on your description.
Make sequential edits in the same session. First pass: "change the background to a beach." Second pass: "make the sky more orange." Third pass: "add palm trees on the right." Each edit preserves the cumulative changes from previous rounds without starting over.
Add objects that weren't in the original image ("add a vase of flowers on the table") or remove unwanted elements ("remove the power lines"). Added objects cast appropriate shadows, match the scene lighting, and respect perspective. Removed objects leave clean, filled backgrounds.
Transform the visual style while keeping the content: "convert to oil painting style", "make it look like a Wes Anderson film still", "apply anime aesthetic". GPT Image 2 changes the rendering style but maintains the original composition, subject identity, and spatial layout.
Change the time of day ("make it nighttime with street lights on"), weather ("add rain and wet reflections on the ground"), or mood ("make the lighting warmer and cozier"). These changes cascade correctly — a nighttime edit adds shadows, turns on windows, and adjusts the color temperature.
Go to Sunra Image and select GPT Image 2 from the model dropdown.
Click the image upload area or drag and drop your source image. GPT Image 2 accepts JPEG, PNG, and WebP. The image is displayed in the preview panel alongside the prompt field.
Write what you want changed — not a full scene description, just the edit instruction. Good: "Replace the blue chair with a red sofa." Bad: "A room with wooden floors, white walls, and a red sofa." Be specific about what changes and what stays.
Click Generate. Use the before/after comparison to check that the edit was applied correctly and that unmentioned areas remain unchanged. Check edges where edits meet original content for blending artifacts.
If the edit looks good but you want further changes, keep the result and write your next instruction. Each round builds on the previous. If the result went in a wrong direction, regenerate from the original image with a revised prompt.
Whether you're a solo creator, an agency, or a brand — every model adapts to how you work.
Café window portrait
Change the view outside the window to a rainy Paris street with blurred car headlights and wet cobblestones. Keep the person and interior lighting exactly the same.
Record shop interior
Add warm string lights hanging across the ceiling and a neon "NOW PLAYING" sign glowing in pink on the back wall. Keep all the vinyl records and shelves unchanged.
Mediterranean alley still life
Replace the drinks on the table with a vase of fresh sunflowers and a stack of old books. Keep the Mediterranean alley background, table, and lighting the same.
Replace the background with a modern minimalist office. Keep the person and their clothing exactly as they are. Soft natural lighting from the left.
Model: GPT Image 2 · Mode: Edit · Source: portrait photo with busy background
Replace the coffee mug in her hand with a glass of iced tea. Keep the hand pose, lighting, and shadows consistent.
Model: GPT Image 2 · Mode: Edit · Source: lifestyle photo
Convert this photograph to a Studio Ghibli animation style. Keep the exact composition, poses, and expressions. Soft cel-shading, warm color palette, visible brush texture.
Model: GPT Image 2 · Mode: Edit · Source: family photo
Round 1: Change the sky to golden hour sunset. Round 2: Add warm light reflections on the building windows. Round 3: Add a few birds flying in the distance.
Model: GPT Image 2 · Mode: Edit · Sequential rounds on same image
Take a single product photo and generate dozens of variants: different background colors, seasonal themes ("add Christmas decorations around the product"), context shots ("place the bag on a café table"). One photo shoot produces an entire season of marketing assets.
Photograph a room and edit it to explore design options: "replace the hardwood floor with light gray tile", "change the curtains to floor-length linen", "paint the accent wall dark navy". Show clients multiple design directions from the same base photo without 3D rendering.
Fix common photo problems with text instructions: "remove the trash can from the background", "fix the overexposed sky", "remove the shadow across her face". Useful for wedding photographers, real estate agents, and anyone cleaning up good-but-imperfect shots.
Designers upload a work-in-progress composition and rapidly test variations: "try the logo in gold instead of silver", "make the background more textured", "increase the contrast between foreground and background". Reduces the Photoshop round-trip time from minutes to seconds per iteration.
| GPT Image 2 Editing | Other Tools | |
|---|---|---|
| Edit interface | Natural language instructions — describe the change in words | Photoshop AI: text + manual selection hybrid. Canva: template-based with limited AI fills. Nano Banana Pro: natural language, similar to GPT Image 2 |
| Multi-turn editing | Sequential edits in one session, each building on the last | Most tools treat each edit independently. Photoshop preserves layers but requires manual management |
| Content understanding | Understands scene context — added objects match perspective, lighting, shadows automatically | Photoshop generative fill: good context matching. Canva: limited scene awareness. Basic inpainting tools: no context |
| Precision for pixel-level edits | Best for semantic edits (change object, swap style). Less precise for exact pixel-level adjustments like 2px border changes | Photoshop: unmatched for pixel-precision. Figma: precise for UI elements |
| Cost and access | Free daily credits on Sunra. No software install. Pay per edit beyond free tier | Photoshop: $22.99/mo subscription + learning curve. Canva Pro: $12.99/mo. Nano Banana Pro: free credits on Sunra |
Write "change the shirt color to navy blue" not "make it look better". Ambiguous instructions produce unpredictable results. If you want to preserve specific elements, say so: "keep the face, hair, and background exactly the same."
Multi-turn editing works best when each round focuses on a single change. "Change the background AND swap the shirt AND add a hat" in one prompt is less reliable than three sequential single-edit rounds. Each focused edit gives you a checkpoint to evaluate before continuing.
GPT Image 2 excels at semantic edits — things described in words ("replace", "remove", "change style"). For precise geometric operations (rotating an object exactly 15 degrees, aligning elements to a grid, pixel-perfect masking), traditional tools like Photoshop are more appropriate.
Editing works better on high-resolution inputs (1024px+). Low-resolution images may produce soft or artifact-heavy results in edited regions, especially when adding detailed objects. If your source is small, consider upscaling it first.
Join thousands of creators, agencies, and brands who use Sunra every day.
