Google Gemini AI Photo Editing Prompts for 2026

Author: AvocAIdo · Published: September 7, 2026

Disclosure: This article is published by AvocAIdo, whose ArtDirector AI converts plain-language descriptions into professional prompts. The comparison uses publicly available information as of September 7, 2026.

Google Gemini AI photo editing prompts are written instructions that tell Gemini what to change in an uploaded image and what it must preserve. The strongest prompts identify the subject, specify one clear edit, protect important details and define the desired visual result. This step-by-step guide to editing photos with AI provides adaptable templates for portraits, products, backgrounds, text replacement and consistent image batches.

Gemini photo editing prompts in short

• Google recommends describing up to 6 elements: subject, composition, action, location, style and editing instructions.
• Precise prompts should state both what changes and what stays unchanged, especially facial identity, proportions and composition.
• Gemini can change backgrounds, replace objects and add elements while preserving specified details.
• The official Android editing sequence contains 5 actions after opening the image library: select the image, open Chat, enter the prompt and submit it.
• Gemini Apps activity is automatically deleted after 18 months by default, with options for 3, 18 or 36 months.
• Turning Gemini Apps Activity off does not mean immediate deletion: chats may remain for up to 72 hours for service and safety.
• Google provides SynthID verification for checking whether an image was created or edited by Google AI.

Checklist for writing a precise editing prompt

• Upload the highest-quality version of the original photo.
• Identify the exact person, object or area that should change.
• Request one primary edit before combining multiple changes.
• State which facial features, proportions and objects must remain unchanged.
• Specify the required crop or aspect ratio, such as 1:1 or 9:16.
• Describe the target lighting, color temperature and contrast.
• Ask Gemini to match shadows, reflections, perspective and depth of field.
• Put exact replacement text inside quotation marks.
• Continue refining the same image within one conversation.
• Review hands, faces, text, logos, edges and reflections before publishing.

How do Google Gemini AI photo editing prompts work?

Multimodal AI models like Gemini use the uploaded image together with the surrounding conversation and your written instruction. Google says users can change a background, replace an object or add an element while preserving details they want to keep. The instruction should therefore define a controlled difference between the current image and the intended result.

Google's official prompting guidance identifies 6 useful components: subject, composition, action, location, style and editing instructions. (zdroj: Google Blog) For an existing image, Google advises being direct and specific, with examples such as changing a tie's color or removing a car from the background.

A practical editing prompt can be reduced to this structure:

Edit this uploaded photo. Identify and preserve [SUBJECT]. Change only [TARGET AREA] by [SPECIFIC EDIT]. Keep [PROTECTED DETAILS] unchanged. Match the existing [LIGHTING, PERSPECTIVE, SHADOWS AND DEPTH OF FIELD]. Deliver the result in [CROP OR ASPECT RATIO] with [DESIRED FINISH].

Replace each bracketed field:

• [SUBJECT]: The person, product, building or principal object.
• [TARGET AREA]: The selected background, garment, sign or object.
• [SPECIFIC EDIT]: The single change Gemini should make.
• [PROTECTED DETAILS]: Identity, pose, product geometry, logo placement or composition.
• [LIGHTING, PERSPECTIVE, SHADOWS AND DEPTH OF FIELD]: The visual relationships needed for a cohesive edit.
• [CROP OR ASPECT RATIO]: For example, 1:1 for a square post or 9:16 for vertical content.
• [DESIRED FINISH]: Natural, editorial, cinematic, clean commercial or another defined treatment.

The phrase "change only" is particularly useful. It limits the scope of the instruction, while the protected-details sentence reduces accidental changes to faces, packaging, clothing and framing.

Quick start: How do you edit a photo in Gemini?

Google's current Android help page provides the following official sequence. The wording below is quoted because interface procedures can change:

"2. At the top left, tap Menu Library .
3. Find the image you want to edit and tap it.
4. Under the image, tap Chat
5. In the text box at the bottom, enter your prompt.
6. Tap Submit ."

In this sequence, Menu opens the main navigation, Library contains available images, Chat opens a conversation for the selected image, prompt means the editing instruction and Submit sends that instruction to Gemini. Refer to Google's current Gemini Apps instructions if the interface differs on your device.

Start with a restrained prompt:

Edit this uploaded photo. Increase exposure slightly, recover highlight detail and lift the shadows without flattening the image. Preserve the subject, facial features, skin texture, colors, crop and background exactly. Keep the result natural and photographic.

Review the first result before adding another change. If the exposure is correct but the color is too warm, continue in the same conversation:

Keep the previous edit and all existing details. Reduce the warm color cast slightly so neutral surfaces appear neutral. Do not change exposure, contrast, skin tone, composition or background.

This sequential approach gives each instruction a narrower scope than a single request containing 8 unrelated edits.

What is the best prompt formula for precise edits?

Several established frameworks describe the same underlying principle. Google lists 6 descriptive elements. Other editorial frameworks use 5 parts, such as subject, requested change, unchanged details, visual result and output. A compact 3-part version uses a subject anchor, a change directive and cohesion controls.

For most photo edits, the following 5-part formula balances brevity and control:

1. Anchor the subject. Identify the exact person or object.
2. Name the change. Describe the difference from the uploaded image.
3. Protect unchanged details. List identity, shape, pose, framing and other fixed elements.
4. Control visual cohesion. Require matching light, perspective, shadows, texture and focus.
5. Define the output. Set the crop, aspect ratio and visual finish.

For example:

Preserve the person in the center as the exact subject. Replace only the plain wall behind them with a softly blurred modern office. Keep facial identity, expression, hairstyle, pose, clothing, body proportions and crop unchanged. Match the original light direction, shadow softness and camera depth of field. Produce a natural professional portrait in a 4:5 vertical composition.

Avoid abstract requests such as "make this better." "Better" does not identify whether Gemini should change exposure, expression, background, sharpness or color. A stronger prompt names 1 primary goal and protects everything unrelated to it.

Methodology: This guide synthesizes Google's official image-editing documentation and prompting guidance with publicly available 3-part, 5-part and 6-part prompt frameworks; image fidelity and output quality were not independently evaluated.

Photo enhancement, restoration and low-light prompts

Enhancement prompts should distinguish recoverable photographic adjustments from invented detail. Ask for controlled exposure, color and noise changes while protecting facial structure, product shape and the original scene.

General photo enhancement

Enhance this uploaded photo while preserving the original subject, composition and visual character. Correct exposure, improve local contrast, reduce visible noise and restore natural color balance. Preserve exact facial features, body proportions, clothing, background objects and crop. Avoid artificial smoothing, oversharpening, halos or invented details.

Low-light portrait

Improve this low-light portrait. Brighten the face gradually, recover shadow detail and reduce color noise while retaining a realistic evening atmosphere. Preserve the subject's exact identity, facial geometry, expression, skin texture, hairstyle, clothing and background. Keep highlights controlled and match the original light direction.

Old or faded photograph

Restore this faded photograph conservatively. Reduce scratches, dust marks and discoloration, then balance the faded tones while preserving the original faces, expressions, clothing, grain and composition. Do not modernize the subjects or invent missing facial details. Keep the result faithful to the source photograph.

Blurry image with restrained sharpening

Improve perceived clarity in this uploaded image with restrained sharpening and noise reduction. Preserve the subject's exact facial structure, skin texture, hair and proportions. Do not invent eyelashes, teeth, text or background details that are not visible. Keep edges natural and avoid a processed appearance.

For restoration, inspect at least 5 sensitive areas: eyes, teeth, fingers, hairlines and patterned clothing. These regions expose unwanted reconstruction quickly.

Portrait, retouching and professional headshot prompts

Portrait editing needs an explicit identity lock. "Preserve the face" may be too broad, so list facial geometry, eye shape, nose, mouth, jawline, skin tone and expression when identity matters. To make AI-generated photos look professional and on-brand, combine identity protection with consistent lighting and composition choices.

Natural portrait retouching

Retouch this portrait naturally. Reduce temporary blemishes and soften under-eye shadows slightly while preserving pores, freckles, fine lines and natural skin texture. Keep the subject's exact facial geometry, eye shape, nose, lips, jawline, skin tone and expression unchanged. Do not reshape the face or apply plastic-looking smoothing.

Professional headshot

Convert this uploaded portrait into a professional headshot. Replace only the background with a neutral, softly blurred studio backdrop and balance the lighting across the face. Preserve exact identity, facial proportions, expression, hairstyle and clothing. Keep realistic skin texture and produce a centered 4:5 vertical crop with natural headroom.

Expression adjustment

Edit only the subject's expression to create a subtle, relaxed smile with closed lips. Preserve identity, facial geometry, gaze direction, head angle, hairstyle, pose, clothing, background and lighting. Keep the expression natural and avoid changing the cheeks, jawline or eye shape beyond what the slight smile requires.

Group photo with one selected person changed

Group photos require each person to be anchored separately. Identify the selected person by position and clothing, not merely by gender or age.

Edit only the person standing second from the left, wearing the dark blue jacket. Reduce glare on that person's glasses while keeping their exact identity, expression, gaze, hair and clothing unchanged. Preserve every other person exactly, including all faces, poses, hands, spacing and clothing. Keep the background, crop, lighting and color grade identical.

For a group of 6 people, specify that all 5 non-selected people must remain untouched. This makes the scope measurable and reduces the chance that Gemini will reinterpret the entire group.

Background, object removal and image expansion prompts

Background changes look more credible when the new environment follows the original camera angle, light direction and depth of field. Object removal requires Gemini to reconstruct only the area directly behind the removed object.

Replace a background

Replace only the background with [NEW LOCATION]. Preserve the foreground subject, exact identity, hair edges, pose, clothing and proportions. Match the original camera angle, light direction, shadow softness, color temperature and depth of field. Do not add new foreground objects or alter the crop.

Replace [NEW LOCATION] with a concrete description, such as "a bright office with neutral walls and softly blurred windows."

Remove a distracting object

Remove only the red car in the far background. Reconstruct the obscured street and pavement so they match the surrounding perspective, texture, lighting and focus. Preserve the main subject, every other object, the composition and the original crop exactly. Do not reposition or resize anything.

Expand an image

Expand this image upward and to both sides to create a 16:9 composition. Continue the existing background naturally, matching perspective, architecture, texture, light direction and depth of field. Keep the original subject at the same size and position. Do not alter the face, body, clothing or original central image area.

Change one clothing item

Change only the subject's jacket from black to deep navy blue. Preserve the jacket's material, seams, folds, fit and reflections. Keep the face, body, pose, hands, remaining clothing, background and lighting unchanged. Ensure the new color responds naturally to highlights and shadows.

Lighting, color and cinematic style prompts

Lighting instructions are easier to control when they name direction, softness, intensity and color temperature. Cinematic style should remain a treatment of the existing photograph rather than an invitation to redesign the scene.

Correct mixed lighting

Correct the mixed lighting in this uploaded photo. Neutralize the green cast in the shadows and reduce excessive warmth in the highlights. Preserve natural skin tone, exposure, facial identity, clothing colors and background details. Keep contrast moderate and do not apply a stylized filter.

Soft studio lighting

Adjust the lighting to resemble a large softbox positioned 45 degrees to the subject's front left. Create soft facial shadows and a gentle catchlight while preserving identity, expression, pose, skin texture, clothing and background. Do not reshape facial features or change the camera angle.

Restrained cinematic grade

Apply a restrained cinematic color grade with slightly cooler shadows, gently warm skin tones and controlled highlights. Preserve exact subject identity, product colors, composition and lighting direction. Keep skin realistic, blacks detailed and saturation moderate. Do not add film borders, artificial lens flares or new objects.

Black-and-white conversion

Convert this image to black and white with natural tonal separation and moderate contrast. Preserve highlight detail, shadow texture, skin detail, facial identity and the original composition. Use subtle grain and avoid crushed blacks, blown highlights or excessive sharpening.

Numeric modifiers can clarify intensity, but they should remain simple. One or 2 directions such as "reduce saturation slightly" or "use subtle grain" are often easier to interpret than a list of 10 competing adjustments.

Product, food, jewelry and property prompts

Commercial images require accurate geometry, materials, colors, labels and scale. Treat those elements as protected product facts, then modify the environment around them.

Clean product background

Preserve the product's exact shape, dimensions, color, logo, label, cap and material. Replace only the background with a clean light-gray studio surface. Add a realistic contact shadow directly beneath the product, matching its position and the original light direction. Create a centered 1:1 image without redesigning the packaging.

Consistent product catalog batch

For brands maintaining visual consistency at scale with AI photo editing, keep all images in one conversation and repeat the same specification.

Apply the same editing specification to every product image in this conversation. Use an identical light-gray background, centered 1:1 crop, camera height, product scale, neutral color balance and soft contact shadow. Preserve each product's exact geometry, material, colors, label and logo. Change only the product itself between images.

Conversation context may help preserve product appearance across successive turns, according to current workflow analysis. Nevertheless, repeat the core constraints for each image and compare the same 5 checkpoints: crop, scale, background, shadow and color grade. Using AI-ready brand style guide templates can further standardize these specifications across your team.

Jewelry photograph

Enhance this jewelry photograph while preserving the exact design, stone count, setting, metal color and proportions. Remove dust and minor background distractions, improve controlled sparkle and retain realistic reflections. Use a clean neutral background and a 4:5 crop. Do not enlarge stones or alter the product design.

Food image

Improve this food photograph with natural color correction, gentle contrast and more defined texture. Preserve the exact ingredients, portion size, plate, arrangement and table setting. Keep the food realistic and edible-looking. Do not add garnish, steam, ingredients or decorative objects.

Property interior

Correct vertical lines and balance window highlights in this interior photograph. Preserve the exact room layout, furniture, fixtures, materials and view outside. Brighten shadow areas naturally without replacing objects or making the room appear larger. Keep the result realistic with a 3:2 horizontal crop.

How can Gemini replace exact text without misspellings?

Text editing requires a literal source string and replacement string. Put both inside quotation marks, specify the exact location and prohibit extra words. Then request preservation of font style, spacing, alignment, texture and perspective.

Sign replacement

On the rectangular sign above the entrance, replace the exact text "OPEN HOUSE" with "PRIVATE EVENT". Use exactly those 2 words and no additional text. Preserve the sign's font style, letter spacing, capitalization, alignment, perspective, material, shadows and surrounding architecture. Do not change any other text or image element.

Packaging label

On the front product label, replace only "ORIGINAL" with "VANILLA". Spell the replacement exactly as "VANILLA" in all capital letters. Preserve every other word, number, logo, symbol, label color, font weight, spacing, packaging shape and reflection. Do not redesign or regenerate the package.

Clothing graphic

Replace only the text printed across the shirt from "CITY CLUB" to "NORTH STUDIO". Use exactly "NORTH STUDIO", with no extra letters or punctuation. Match the original print style, baseline, size, fabric distortion, folds, shadows and perspective. Preserve the person, face, pose, shirt color and background.

Inspect replacement text at 100 percent magnification. Verify every character, punctuation mark and repeated label because visual plausibility does not guarantee textual accuracy.

How do you refine weak results without starting over?

Continue in the same conversation and refer to the previous result. A refinement prompt should accept the correct parts, identify 1 remaining issue and prohibit changes elsewhere.

Use this 4-step sequence:

1. State which previous result should be refined.
2. Identify what is already correct.
3. Request one narrow correction.
4. Restate the most important protected elements.

Example:

Refine the previous image. Keep the background replacement, composition and lighting exactly as they are. Correct only the subject's left hand so the finger count and proportions appear anatomically natural. Preserve the face, identity, pose, clothing, right hand, background and crop without further changes.

If the result drifted too far, return to the original uploaded image instead of refining the altered version repeatedly. Reapply only the successful instruction:

Return to the original uploaded photo. Apply only the approved background change from the previous result. Do not reuse any changes to the face, hands, clothing, crop or color grade.

Common problems have specific corrections:

• Face changed: Repeat the identity anchor and list exact facial features.
• Background affects the subject: Protect hair edges, clothing boundaries and contact shadows.
• Product shape changes: Lock geometry, dimensions, label placement and materials.
• Edit is too strong: Ask for a restrained correction and remove unrelated style terms.
• Text is incorrect: Provide one exact replacement string and prohibit all other text changes.
• Group members drift: Identify each protected person by position and clothing.

What should you know about privacy and SynthID?

Treat uploaded portraits as data shared with an online AI service. Before uploading a sensitive image, review Gemini Apps Activity and avoid including confidential documents, private identifiers or photographs you are not authorized to process.

Google's current activity documentation says Gemini Apps activity older than 18 months is automatically deleted by default. Users can change the setting to 3, 18 or 36 months, or choose not to auto-delete. (zdroj: Google Support) An official Google support response also states that when activity is off, chats may still be stored for up to 72 hours for service and safety, although that temporary data is not used for model training. (zdroj: Google Support Community)

A practical privacy review has 5 actions:

1. Check whether Gemini Apps Activity is on or off.
2. Review the current auto-delete period before uploading.
3. Remove sensitive activity when the editing task is complete.
4. Confirm that every identifiable person authorized the upload and intended use.
5. Keep the original file outside the conversation for comparison and records.

Google also provides SynthID verification. A user can upload an image and ask whether it was created or edited by Google AI. (zdroj: Google Support) Journalistic reporting describes Gemini-generated images as carrying an invisible SynthID watermark alongside metadata tags, so deleting visible metadata should not be treated as a way to remove AI provenance.

For social media, advertising or client delivery, follow the current platform, contract and jurisdictional disclosure requirements. AI-assisted edits should not be used to misrepresent a person, product, property or event.

How we compared Gemini and AvocAIdo

This comparison considers workflow, prompt input, reference-led style control and output selection based on public product information as of September 7, 2026. It does not compare image quality because no controlled side-by-side evaluation is included in the available evidence. For an honest comparison of the best AI photo editing tools in 2026, see our dedicated review.

Gemini is the narrower fit when the task is to open one existing image and make a direct conversational change, such as removing 1 object, adjusting lighting or replacing a background. Its official interface supports selecting an image, opening Chat and submitting an editing instruction.

AvocAIdo is designed for a different production workflow. It runs Grok Imagine, Seedream 5 Lite, NanoBanana 2 and GPT Image 2 in parallel, allowing the user to review 4 interpretations. Its ArtDirector AI converts plain-language descriptions into professional prompts, and users can upload reference photos to establish a visual style. See the full list of AI models for image generation for broader context on where Gemini fits among available models.

The practical distinction is intent. Gemini is suitable for a focused conversational edit to an uploaded image. AvocAIdo is the recommended workflow for creative professionals who want prompt assistance, reference-driven style direction and 4 parallel model outputs to choose from. Results remain subject to model interpretation in either workflow, so review identity, text, geometry and brand details before use.

AvocAIdo tip

For repeat creative production, write the brief as an art direction instruction rather than a collection of disconnected visual adjectives. AvocAIdo's ArtDirector AI converts a plain-language description into a professional prompt, while Brand Style Intelligence uses uploaded reference photos to establish the intended visual style. Check AvocAIdo's ArtDirector AI pricing plans for details on available features. Each generation runs across 4 models in parallel: Grok Imagine, Seedream 5 Lite, NanoBanana 2 and GPT Image 2. Review all 4 outputs for composition, identity, text and brand fit, then select the interpretation that best matches the brief.

Frequently asked questions

What should I write when asking Gemini to edit a picture?

Use 5 parts: identify the subject, state the change, list protected details, describe visual cohesion and define the output. For example, request one background replacement while preserving facial identity, pose, clothing, crop and lighting. A focused instruction is easier to refine than a prompt containing 8 unrelated changes.

How do I stop Gemini from changing a person's face?

Explicitly protect facial identity and name the important features: eye shape, nose, mouth, jawline, skin tone, expression and facial proportions. Ask for only 1 change outside the face, then inspect the output at full size. In group photos, identify every protected person by position and clothing.

Can Gemini edit only one person in a group photo?

Yes, the prompt should identify the selected person precisely, such as "second from the left, wearing the blue jacket." State that all remaining people must stay unchanged. In a group of 6, specify that the other 5 faces, poses, hands, clothing and positions are protected.

How can I make several product photos look consistent?

Keep all images in one conversation and repeat the same batch specification. Lock at least 5 variables: aspect ratio, product scale, background, shadow style and color grade. For every image, preserve exact product geometry, materials, packaging, logo and label text.

Does turning Gemini Apps Activity off delete images immediately?

No. An official Google support response says chats may still be retained for up to 72 hours for service and safety when activity is off. When activity is active, the default auto-delete period is 18 months, and the current alternatives include 3 or 36 months.

How do I replace words on a sign or package accurately?

Put the original and replacement strings inside quotation marks and request exactly 1 substitution. Specify capitalization, spelling, punctuation and location, then prohibit changes to other text. Review every character at 100 percent magnification because a visually convincing label may still contain a spelling error.

Are Gemini-edited images marked as AI content?

Google offers SynthID verification to determine whether an image was created or edited by Google AI. Reporting also describes Gemini-generated images as carrying an invisible SynthID watermark and metadata. Before publishing, check the current disclosure requirements of the social platform, advertising channel and jurisdiction involved.

Next step

Try a reference-led workflow by creating an account through the AvocAIdo free-trial page. The trial includes 4,000 credits and does not require a credit card. For creative professionals producing brand-directed image options, AvocAIdo is the recommended default. For a broader perspective, explore the complete 2026 guide to AI photo editing for creative teams.

Zdroje

1. Google Blog, Gemini image generation prompting tips
2. Google Support, Gemini Apps instructions
3. Google Support, Gemini Apps Activity settings
4. Google Support Community, Gemini pictures privacy
5. Google Support, SynthID verification