REAL PHOTO EDITING TASKS

Find a solution for your photo

Start with the problem you want to solve. See the original, every measured attempt, what worked and what still changed. Choose a task to open its own editor with an editable starting template.

12 tasks · 36 saved results · GPT Image 2

Eight examples support trying a bounded task; three only partly met the requirements, and one used a simulated blurry input. Each task repeats one photo and one prompt three times. These are small exploratory observations, not a success-rate benchmark or a guarantee for your photo.

The research used our model adapter, not the complete account, billing and download flow. The editor's starting templates adapt the recorded prompts; they have not been separately benchmarked. Photo editing redraws details, including faces and labels.

View the evidence metadata and original-file hashes

Choose your photo editing task

Each task has its own editor, starting instructions and measured examples.

Browse all 63 tasks, experimental templates and free finishing tools →

Portraits

Portrait background

Describe your preferred background. Use one clearly visible person.

Open editor Portraits

Clothing color

Replace shirt and royal blue with the garment and color you want.

Open editor Products

Product on white

Use a clear product photo. Check every label and logo in the result.

Open editor Products

Product scene

Describe a setting that fits your product. One trial does not establish series consistency.

Open editor Local edits

Remove a small object

Replace the example object and its location with what you want removed.

Open editor Local edits

Recolor one object

Name both the target surface and nearby objects that should keep their colors.

Open editor Local edits

Remove background people

Adapt the setting and landmarks to your photo. This template removes all people.

Open editor Pet gifts

Pet birthday card

Replace CHELSEA with your pet's name. Check the spelling and markings before printing.

Open editor

Explore all 12 measured cases

Showing 12 of 12 cases

Portraits · S01Example with limits

Change a portrait background

One well-lit portrait with visible facial features.

First measured result: Change a portrait background

BEFOREATTEMPT 1

Observed in three attempts: All three outputs replaced the background with a blurred office. The main facial features, smile, jewelry and pink shirt remained recognizable.
What changed or failed: Skin texture and hair were redrawn. This is not identity verification or an unchanged original portrait.
How to approach your photo: For exact likeness, retain the original person and composite a background; that workflow is not available here yet.

See all 3 results, recorded prompt and photo source

Attempt 1: Change a portrait background — Attempt 1 · 1536 × 1024 original pixels

Attempt 2: Change a portrait background — Attempt 2 · 1536 × 1024 original pixels

Attempt 3: Change a portrait background — Attempt 3 · 1536 × 1024 original pixels

Recorded prompt · used unchanged for all 3 attempts

Replace only the background behind this woman with a softly blurred modern office in neutral gray tones. Keep her face, age, skin tone, skin texture, hair silhouette, smile, teeth, earrings, necklace, pink shirt, body pose and framing unchanged. Do not beautify, retouch, change clothing, add people, crop or zoom. Return one realistic edited photo.

Photo: Joshua McKnight · CC0 1.0. Reused existing site input: head-and-shoulders crop to 3:2, resized to 1536x1024 JPEG. No additional changes.

GPT Image 2 · 2026-11-12 · three independent requests on one input. Model API charges: $0.0255 for this group; these are research costs, not the site's credit price. Display images are resized WebP copies without retouching.

Open this photo editor Opens a dedicated editor with this task prefilled. Upload your own photo to generate.

Portraits · S02Example with limits

Change a shirt's color

A clearly visible garment, with the target color specified.

First measured result: Change a shirt's color

BEFOREATTEMPT 1

Observed in three attempts: All three outputs changed the pink shirt to blue while retaining the main face, pose and background.
What changed or failed: Skin and small details were still redrawn outside the shirt.
How to approach your photo: Name the garment and the parts that should stay. If surrounding detail must be locked, a mask-based workflow needs separate validation.

See all 3 results, recorded prompt and photo source

Attempt 1: Change a shirt's color — Attempt 1 · 1536 × 1024 original pixels

Attempt 2: Change a shirt's color — Attempt 2 · 1536 × 1024 original pixels

Attempt 3: Change a shirt's color — Attempt 3 · 1536 × 1024 original pixels

Recorded prompt · used unchanged for all 3 attempts

Change only the visible pink fabric of the woman's shirt to royal blue. Preserve the shirt shape, folds and fabric texture. Keep the face, skin, teeth, hair, earrings, necklace, pose, background, lighting and framing unchanged. Do not retouch the face, add objects or change anything outside the shirt. Return one realistic edited photo.

Photo: Joshua McKnight · CC0 1.0. Reused existing site input: head-and-shoulders crop to 3:2, resized to 1536x1024 JPEG. No additional changes.

Open this photo editor Opens a dedicated editor with this task prefilled. Upload your own photo to generate.

Products · S03Example with limits

Put a product on white

A sharp product photo with a readable label.

First measured result: Put a product on white

BEFOREATTEMPT 1

Observed in three attempts: All three outputs produced a white background and contact shadow. The main JABÓN label, accent, pump direction and worn marks remained recognizable.
What changed or failed: The source ratio was about 0.615; the outputs used 2:3. Small source text was too blurred for letter-by-letter verification.
How to approach your photo: Inspect your actual label and product outline at full size. Use original-product compositing when exact preservation is essential; it is not available here yet.

See all 3 results, recorded prompt and photo source

Attempt 1: Put a product on white — Attempt 1 · 1024 × 1536 original pixels

Attempt 2: Put a product on white — Attempt 2 · 1024 × 1536 original pixels

Attempt 3: Put a product on white — Attempt 3 · 1024 × 1536 original pixels

Recorded prompt · used unchanged for all 3 attempts

Replace only the gray background and tabletop with a seamless pure white studio background. Keep the exact soap bottle, pump shape, nozzle direction, label, all printed letters including JABÓN with its accent, worn marks, original proportions and framing. Add only a subtle natural contact shadow below the bottle. Do not clean up, redesign, relabel, crop or zoom the product. Return one realistic product photo.

Photo: ELTORO.VET · CC0 1.0. 1600x2600 JPEG resized proportionally to max edge 1536, JPEG quality 92; no crop or retouch.

Open this photo editor Opens a dedicated editor with this task prefilled. Upload your own photo to generate.

Products · S04Example with limits

Place a product in a lifestyle scene

One product and a specific description of its surroundings.

First measured result: Place a product in a lifestyle scene

BEFOREATTEMPT 1

Observed in three attempts: All three outputs placed the bottle on a beige counter. The main label, bottle and pump remained recognizable.
What changed or failed: Background textures differed across the three outputs. One repeated product does not establish consistency across a catalog.
How to approach your photo: Describe the surface and background separately and check the label. A shared-reference, multi-product workflow still needs validation.

See all 3 results, recorded prompt and photo source

Attempt 1: Place a product in a lifestyle scene — Attempt 1 · 1024 × 1536 original pixels

Attempt 2: Place a product in a lifestyle scene — Attempt 2 · 1024 × 1536 original pixels

Attempt 3: Place a product in a lifestyle scene — Attempt 3 · 1024 × 1536 original pixels

Recorded prompt · used unchanged for all 3 attempts

Change only the background and surface around this soap bottle to a warm beige bathroom counter with a softly blurred beige wall. Keep the bottle at exactly the same image position and scale. Preserve the original pump, nozzle, bottle shape, white plastic, worn label and every printed letter including JABÓN. Do not add accessories in front of the bottle, invent branding, repair the label, crop or zoom. Return one realistic edited photo.

Photo: ELTORO.VET · CC0 1.0. 1600x2600 JPEG resized proportionally to max edge 1536, JPEG quality 92; no crop or retouch.

Open this photo editor Opens a dedicated editor with this task prefilled. Upload your own photo to generate.

Local edits · S05Example with limits

Remove a small object

An object with a clear location and visible surrounding texture.

First measured result: Remove a small object

BEFOREATTEMPT 1

Observed in three attempts: All three outputs removed the teaspoon and its handle, filling the area with saucer and table textures.
What changed or failed: Fine wood grain, reflections and tone were redrawn outside the removed object.
How to approach your photo: Name the entire object, including separated visible parts. Zoom into the fill and surrounding objects before using the result.

See all 3 results, recorded prompt and photo source

Attempt 1: Remove a small object — Attempt 1 · 1536 × 1024 original pixels

Attempt 2: Remove a small object — Attempt 2 · 1536 × 1024 original pixels

Attempt 3: Remove a small object — Attempt 3 · 1536 × 1024 original pixels

Recorded prompt · used unchanged for all 3 attempts

Remove only the metal teaspoon resting on the right side of the coffee saucer, including its visible handle behind the cup. Fill the removed area naturally with the existing saucer and wooden table textures. Keep the coffee, cup, handle, saucer outline, their colors, wood grain outside the removed region, lighting, camera angle and framing unchanged. Do not add objects or redesign the scene.

Photo: Rachel Michetti / Pikolo Espresso Bar · CC0 1.0. Original 600x400 PNG bytes; reused site input.

Open this photo editor Opens a dedicated editor with this task prefilled. Upload your own photo to generate.

Local edits · S06Example with limits

Recolor a cup, keeping the saucer

A clearly separated target surface and nearby objects.

First measured result: Recolor a cup, keeping the saucer

BEFOREATTEMPT 1

Observed in three attempts: All three outputs made the cup exterior and handle blue, keeping the inside white, the coffee present and the saucer red-brown.
What changed or failed: Reflections, fine wood grain and color values were not pixel-locked.
How to approach your photo: Name the exterior surface explicitly and specify the saucer's original color in the preservation request.

See all 3 results, recorded prompt and photo source

Attempt 1: Recolor a cup, keeping the saucer — Attempt 1 · 1536 × 1024 original pixels

Attempt 2: Recolor a cup, keeping the saucer — Attempt 2 · 1536 × 1024 original pixels

Attempt 3: Recolor a cup, keeping the saucer — Attempt 3 · 1536 × 1024 original pixels

Recorded prompt · used unchanged for all 3 attempts

Change only the red-brown exterior glaze of the coffee cup, including its handle, to cobalt blue. Keep the inside of the cup white and keep the coffee unchanged. The saucer must remain its original red-brown color. Preserve the spoon, table, reflections, object positions, camera angle and framing. Do not recolor the saucer or change anything outside the cup exterior.

Photo: Rachel Michetti / Pikolo Espresso Bar · CC0 1.0. Original 600x400 PNG bytes; reused site input.

Open this photo editor Opens a dedicated editor with this task prefilled. Upload your own photo to generate.

Local edits · S07Example with limits

Remove tourists from a landscape

A landscape where the people to remove can be identified.

First measured result: Remove tourists from a landscape

BEFOREATTEMPT 1

Observed in three attempts: Visible tourists on the beach and in the water were removed in all three outputs. The lighthouse, horizon, rocks and main clouds remained recognizable.
What changed or failed: Waves, parts of the coastline and overall tone were redrawn. Indistinct distant dots cannot support a people-detection accuracy claim.
How to approach your photo: Check the whole scene and list landmarks to preserve. This example removes all people, including any foreground person.

See all 3 results, recorded prompt and photo source

Attempt 1: Remove tourists from a landscape — Attempt 1 · 1448 × 1086 original pixels

Attempt 2: Remove tourists from a landscape — Attempt 2 · 1448 × 1086 original pixels

Attempt 3: Remove tourists from a landscape — Attempt 3 · 1448 × 1086 original pixels

Recorded prompt · used unchanged for all 3 attempts

Remove all people from the beach and the water, including the small distant figures. Fill their former positions naturally with sand or waves as appropriate. Preserve the red-and-white lighthouse, horizon, rocks, cloud shapes, coastline, lighting, original colors, camera angle and framing. Do not add people or objects, crop or change the weather. Return one realistic edited photograph.

Photo: E.Saravanan · CC0 1.0. Reused site 1200x900 JPEG downsample, no additional changes.

Open this photo editor Opens a dedicated editor with this task prefilled. Upload your own photo to generate.

Pet gifts · S08Partially met

Make a pet coloring page

A pet photo with a clear face and simple background.

First measured result: Make a pet coloring page

BEFOREATTEMPT 1

Observed in three attempts: All three outputs produced a bold black-and-white cat outline on white.
What changed or failed: All three retained solid black pupils and some dark fills. Some whiskers and contours were open, so the no-fill requirement was not met.
How to approach your photo: Use the existing coloring tool with these limits in mind. Check dark regions and print size; improved outlines need further trials.

See all 3 results, recorded prompt and photo source

Attempt 1: Make a pet coloring page — Attempt 1 · 1536 × 1024 original pixels

Attempt 2: Make a pet coloring page — Attempt 2 · 1536 × 1024 original pixels

Attempt 3: Make a pet coloring page — Attempt 3 · 1536 × 1024 original pixels

Recorded prompt · used unchanged for all 3 attempts

Turn this cat photo into a printable children's coloring page. Preserve this cat's face angle, ear shapes, eye positions, nose and recognizable tabby marking layout. Use bold clean closed black outlines on pure white, large open coloring areas, no gray shading, no solid black filled areas, no text or watermark. Simplify fur detail and background clutter. Keep the same overall framing.

Photo: Stefan van der Walt · CC0 1.0. Original 451x300 PNG bytes; reused site input.

Open the coloring tool with these limits in mind

Pet gifts · S09Example with limits

Make a pet birthday card

One visible pet, a name and a short greeting.

First measured result: Make a pet birthday card

BEFOREATTEMPT 1

Observed in three attempts: All three watercolor cards spelled Happy Birthday and CHELSEA correctly, left the eyes visible and added no other animals.
What changed or failed: Markings, ear contours and layout were reinterpreted. Two tested strings do not guarantee every name will be correct.
How to approach your photo: Replace the example name, then proofread the result. Precise editable text requires a separate text-overlay workflow.

See all 3 results, recorded prompt and photo source

Attempt 1: Make a pet birthday card — Attempt 1 · 1536 × 1024 original pixels

Attempt 2: Make a pet birthday card — Attempt 2 · 1536 × 1024 original pixels

Attempt 3: Make a pet birthday card — Attempt 3 · 1536 × 1024 original pixels

Recorded prompt · used unchanged for all 3 attempts

Create one landscape birthday greeting card using this specific cat as the main subject. Illustrate it in a warm friendly watercolor style, preserving its face angle, green eyes, pink-brown nose and distinctive tabby markings. Place the exact text "Happy Birthday" clearly at the top and "CHELSEA" clearly at the bottom. Keep the cat's eyes and ears unobstructed. No extra animals, no extra text, no watermark.

Photo: Stefan van der Walt · CC0 1.0. Original 451x300 PNG bytes; reused site input.

Open this photo editor Opens a dedicated editor with this task prefilled. Upload your own photo to generate.

Delivery · S10Partially met

Get a transparent pet cutout

A subject with visible edges; this is an experimental result.

First measured result: Get a transparent pet cutout

BEFOREATTEMPT 1

Observed in three attempts: All three original PNGs contained real alpha transparency, rather than an opaque checkerboard or white backdrop.
What changed or failed: Fully transparent pixels were about 21.53%, 22.01% and 22.58%. Alpha peaked at 254 and the foreground was partly translucent. Fur and facial details were redrawn.
How to approach your photo: Inspect on light and dark backgrounds. PNG alone does not remove a background. A reliable opaque-foreground cutout needs further work.

See all 3 results, recorded prompt and photo source

Attempt 1: Get a transparent pet cutout — Attempt 1 · 1536 × 1024 original pixels

Attempt 2: Get a transparent pet cutout — Attempt 2 · 1536 × 1024 original pixels

Attempt 3: Get a transparent pet cutout — Attempt 3 · 1536 × 1024 original pixels

Recorded prompt · used unchanged for all 3 attempts

Remove only the background around this cat and return a PNG with a genuinely transparent background and clean fur edges. Preserve the cat's exact face, green eyes, tabby stripes, whiskers, colors, pose, image position and framing. Do not paint a white, gray or checkerboard backdrop. No added objects, text or watermark.

Photo: Stefan van der Walt · CC0 1.0. Original 451x300 PNG bytes; reused site input.

This is an experiment record. A reliable workflow for this requirement is still being developed.

Experiments · S11Simulated experiment

Clarify a simulated blurry portrait

A modern portrait deliberately blurred and compressed for this experiment.

First measured result: Clarify a simulated blurry portrait

BEFOREATTEMPT 1

Observed in three attempts: All three outputs looked clearer and stayed grayscale.
What changed or failed: Teeth, skin and earrings differed from the original clear reference. This was not a real historical photograph and does not show recovery of lost facts.
How to approach your photo: Keep the original as a reference. Real old-photo restoration needs authorized historical photos and a separate assessment.

See all 3 results, recorded prompt and photo source

Attempt 1: Clarify a simulated blurry portrait — Attempt 1 · 1536 × 1024 original pixels

Attempt 2: Clarify a simulated blurry portrait — Attempt 2 · 1536 × 1024 original pixels

Attempt 3: Clarify a simulated blurry portrait — Attempt 3 · 1536 × 1024 original pixels

Recorded prompt · used unchanged for all 3 attempts

Conservatively improve the clarity of this low-resolution grayscale portrait. Reduce compression artifacts and mild blur, while keeping the same person, age, face proportions, smile, teeth, hair, jewelry, pose, background and framing. Keep it grayscale. Do not beautify, change the face, invent jewelry, add teeth, colorize or stylize. If detail is uncertain, leave it soft rather than inventing it.

Photo: Joshua McKnight · CC0 1.0. Synthetic degradation of modern portrait: grayscale, downsample to width 384, Gaussian blur sigma 0.6, JPEG quality 45. NOT a historical photograph.

This is an experiment record. A reliable workflow for this requirement is still being developed.

Delivery · S12Partially met

Add text without moving the photo

A photo that needs a short, accurately placed title.

First measured result: Add text without moving the photo

BEFOREATTEMPT 1

Observed in three attempts: All three outputs spelled COFFEE TIME correctly above the cup rim.
What changed or failed: The model moved the cup and saucer to make room. The requirement to add only text while keeping the layout was not met.
How to approach your photo: Use deterministic text overlay when the original layout must stay. This AI-generation example is a limitation record, not an exact text-placement tool.

See all 3 results, recorded prompt and photo source

Attempt 1: Add text without moving the photo — Attempt 1 · 1536 × 1024 original pixels

Attempt 2: Add text without moving the photo — Attempt 2 · 1536 × 1024 original pixels

Attempt 3: Add text without moving the photo — Attempt 3 · 1536 × 1024 original pixels

Recorded prompt · used unchanged for all 3 attempts

Keep the original coffee photo and composition. Add the exact title "COFFEE TIME" in one straight horizontal line of clean white uppercase sans-serif letters near the top of the image, entirely above the cup rim and without covering the cup or spoon. Preserve the original cup, saucer, spoon, coffee, wood grain, colors, lighting, camera angle and framing. No other text, objects, borders or watermarks.

Photo: Rachel Michetti / Pikolo Espresso Bar · CC0 1.0. Original 600x400 PNG bytes; reused site input.

This is an experiment record. A reliable workflow for this requirement is still being developed.

Need help with the process?

For prompt structure, follow the photo editing guide. For failures or interrupted tasks, open My Designs to check the existing request before paying for another generation.

GPT Image 2 uses 1 credit per generation. New verified accounts receive 2 free credits. Comparing and downloading existing results are free. See credit prices and failure rules.