Skip to content
image 2.5
English
Start creating

Image to prompt: read a reusable prompt from any photo

Upload one image to get a structured prompt back, listing subject, treatment and camera in the order your model reads them.

Creative settings

Image to Prompt

Source image*

The prompt structure will be adapted to the target workflow.

Advanced options

Generated result

Complete the settings and generate. Your result will appear here without leaving the page.

Examples

Split image pairing a photograph of a puppy with a limited-colour woodcut print made from the same subject
Prompt behind this image
Create a vertical double-panel style image. The screen is strictly divided into upper and lower parts, with a height ratio of about 1:1.

Upper part:
Please design an original photo with a realistic photographic feel. The main subject is a puppy standing or sitting quietly, or a natural interaction between a person and a puppy, but the focus of the main subject must clearly fall on the person/pet itself. The photo should have a soft, healing, and bright atmosphere. It can be placed in natural environments such as grass, indoor corners, by a door, or by a window, but the environment serves only as a foil; do not treat architecture or space as the main subject.

Lower part:
Strictly based on the same subject, outline, posture, and narrative relationship from the original image in the upper part, reconstruct it into a limited-color woodcut print mark floating on the paper. Do not directly copy the entire photo, nor simply apply a filter. Instead, actively subtract complex backgrounds and minor information, retaining only the recognizable outlines, proportions, directions, light and shadow divisions, key structures, and core spatial relationships. Re-summarize the subject in the language of woodcut prints, linocuts, and manual multi-plate printing, making it instantly recognizable while possessing a simple, clumsy, and strongly hand-crafted artistic temperament.

The graphic brushstrokes should carry the characteristic clumsiness and hand-made traces of woodcut prints: lines do not need to be smooth or precise, allowing for slight carving-like hesitation, uneven thickness, fuzzy edges, stiff transitions, local breaks, gaps, misalignments, and irregular white space; the edges of color blocks can show dry brushing, missed printing, lack of ink, and coarse grain textures. The overall effect is not a refined vector illustration, but rather like being carved and printed stroke by stroke with a hand chisel and ink, carrying a simple, direct, slightly naive yet very durable printmaking quality.

The composition focuses on a single visual anchor point, large areas of white space, and incomplete edges. The position and scale are determined by the subject's own visual center of gravity, movement direction, and structural trend. It can be centered, offset, slightly cropped, or extended in a certain direction. Sufficient warm paper white space is reserved around the subject, and the edges gradually disappear through lack of ink, broken textures, paper showing through, and a sense of natural wear, avoiding complete rectangular borders, making the image look like a memory gently embossed on paper.

The color scheme extracts the most recognizable and vital colors from the original photo in the upper part, but converts them into a fresh, soft, healing, and bright limited-color logic. Use a comprehensive color system that is clean and light, such as cream white, ivory white, light pink, misty green, light blue, tender yellow, pale orange, mint green, soft coral, and light ochre, maintaining low to medium saturation, high brightness, and soft contrast. Retain the most touching color memory points of the original image, but avoid dark, dull, dirty, heavy, and overly vintage color schemes. All colors should be overprinted like matte physical ink layers, soft and clear, yet still possessing the grain and warmth of a print.

Text also intervenes as part of the print composition rather than fixed explanatory text. Extract a 1–3 word main title from the subject's identity, action, emotion, state, location, or metaphor in the upper photo, and pair it with a set of extremely short keywords or a micro-description. The main title can be arranged along the subject's outline, white space boundaries, visual axes, or print edges, forming a supporting, interspersed, misplaced, or echoing relationship with the graphics; auxiliary small characters remain tiny and restrained, using typewriter fonts, monospaced serif fonts, or old archival-style layouts, like annotations in travel notes or private collection prints. Do not use a fixed year; let the text grow naturally based on the photo itself.

The overall emphasis is on the simple hand-crafted feel of woodcut prints, fresh and soft limited colors, paper white space, incomplete edges, and restrained layout, presenting a high-end print temperament that is warm, quiet, healing, light, and nostalgic without being gloomy. Maintain a clear and clever visual echo between the original image in the upper part and the woodcut reconstruction in the lower part, avoiding ordinary filters, complete background replication, smooth vector outlines, cartoon illustrations, travel template feels, and cheap vintage vibes.

The clearest demonstration on this page. A single prompt asks for a photograph on top and a woodcut print of the same subject underneath, which is exactly the photo-to-derived-image relationship this tool works in reverse.

Prompt and image published by @zhidawang219555; the original post is the linked source. Quoted as posted, with markdown emphasis and surplus blank lines removed. · Third-party example. Text and image remain the work of the original author; no reuse licence is granted here.

Photorealistic portrait of a young woman seated at a brushed steel table in a small Taiwanese eatery
Prompt behind this image
A photorealistic cinematic vertical portrait of a young woman in her early twenties, around 20 years old, seated at a stainless-steel table inside a small, busy traditional Taiwanese-style eatery. Her ethnicity is unspecified because race or ethnicity cannot be reliably determined from appearance alone. She has smooth fair-to-light skin with a soft natural complexion, a delicate oval face, a gently tapered jawline, subtle cheekbones, a small straight nose, naturally shaped dark eyebrows, and softly full muted rose-pink lips. Her dark eyes are lowered and mostly closed as she looks down toward the table, creating a quiet, introspective expression. Her makeup is understated and polished, with soft eyeliner, defined lashes, subtle blush, and natural pink lip color.

Her hair is deep black to very dark brown, gathered into a loose high bun with natural volume and slightly tousled texture. Long face-framing strands fall on both sides of her cheeks, including a thicker sweeping section across one side of her forehead. She has a slender feminine figure with a defined waist and naturally full, balanced curves.

She wears exactly a fitted short-sleeved crew-neck T-shirt in a muted dusty mauve-pink tone. The fabric is soft, slightly stretchy, and closely follows the contours of her torso. Only the upper portion of her light gray or muted khaki bottoms is visible below the shirt near the edge of the table. No additional outerwear or visible jewelry.

She sits facing the camera almost directly, torso upright but relaxed. Her left elbow is raised and bent, with her left hand lightly touching the side of her face near her temple and cheek, fingers softly curved. Her right forearm angles downward toward the tabletop while her right hand holds a black smartphone with a clear or translucent protective case, its screen facing upward. Her gaze remains directed downward.

The foreground is dominated by the brushed stainless-steel tabletop with realistic scuffs and reflections. Behind her is a compact open kitchen filled with metal pots, trays, shelves, cooking equipment, workers, tiled surfaces, exposed wiring, and colorful Chinese menu boards with red, green, blue, and black characters and prices. A single exposed warm-white bulb hangs behind her head, creating a subtle rim glow. Use realistic mixed warm tungsten and cool ambient lighting, shallow depth of field, a 35mm equivalent lens, natural perspective, crisp facial detail, soft background separation, subdued contrast, slightly warm documentary color grading, and authentic late-night street-food atmosphere. 9:16

#AIArt #AIgirl

A portrait brief written as a paragraph rather than a tag list. Notice how early the light source and the room behind her are fixed, and how much of the text is spent on the background the photographer did not choose.

Prompt and image published by @Kunda623270; the original post is the linked source. Quoted as posted, with markdown emphasis and surplus blank lines removed. · Third-party example. Text and image remain the work of the original author; no reuse licence is granted here.

Architectural sketchbook page showing the Colosseum drawn in pencil, ink and watercolour washes
Prompt behind this image
Create a stunning 4:5 vertical fine-art sketch illustration of [STRUCTURE NAME] as if it is being drawn by a master architectural artist inside a luxury travel sketchbook.

Show the famous structure transitioning from loose preliminary pencil lines into a fully detailed finished architectural drawing.

The foreground should contain delicate construction lines, perspective guides, measurement marks, small handwritten architectural notes, compass marks, arrows, erased pencil strokes, and unfinished portions.

As the eye moves toward the main structure, the sketch gradually becomes more refined — combining graphite pencil, fine black ink, subtle watercolor washes, cross-hatching, and delicate hand-drawn textures.

Include small artistic studies around the main drawing: a tiny side elevation, architectural detail, window or ornament study, simplified map outline, and a small silhouette showing the scale of the structure.

Use an elegant warm ivory sketchbook-paper background with natural paper fibers and subtle imperfections.

Keep the composition sophisticated and spacious. The structure should remain the unmistakable hero.

Add minimal handwritten-style labeling:

[STRUCTURE NAME]
[CITY, COUNTRY]
ARCHITECTURAL STUDY

The artwork should feel like an original museum-quality architectural sketch discovered inside an explorer’s personal sketchbook — artistic, intellectual, nostalgic, detailed, and highly collectible.

No photorealistic rendering, no 3D CGI, no modern infographic boxes, no excessive text, no watermark.

Look at the last line. Saying no photorealistic rendering and no 3D CGI is what keeps a sketch looking drawn rather than rendered, and that kind of exclusion is the first thing a reversed prompt leaves out.

Prompt and image published by @Naiknelofar788; the original post is the linked source. Quoted as posted, with markdown emphasis and surplus blank lines removed. · Third-party example. Text and image remain the work of the original author; no reuse licence is granted here.

Travel infographic poster of a national park with a bison in a cinematic landscape and small data panels
Prompt behind this image
Create a breathtaking, premium national park travel infographic poster for [NATIONAL PARK, COUNTRY], designed to inspire people to explore, protect, and appreciate nature.

Show a cinematic natural landscape featuring the park’s most famous animal — [FAMOUS ANIMAL] — naturally integrated into the scene, surrounded by its authentic flora and fauna. Include subtle visual elements representing the park’s ecosystem, such as native trees, flowers, grasses, mountains, rivers, birds, and wildlife.

Use an elegant editorial travel-poster aesthetic with realistic nature photography blended with refined graphic-design elements. Warm natural tones, soft sunlight, atmospheric depth, detailed textures, and a clean premium composition. Make the wildlife the emotional focal point while keeping the landscape expansive and majestic.

Add beautifully organized, minimal information panels:

[NATIONAL PARK]
[COUNTRY]

Famous Animal: [ANIMAL]
Flora & Fauna: [KEY SPECIES / VEGETATION]
Species: [NUMBER OF SPECIES]
Area: [AREA]
📍 Location: [LOCATION]
Best Time to Visit: [MONTHS / SEASON]
Ticket Price: [PRICE]

Add a short inspirational line:

“Protect the wild. Preserve the wonder.”

Keep the typography sophisticated, clean, and highly legible. Use small icons, fine divider lines, subtle map/topographic details, and a restrained information hierarchy. The overall design should feel like a luxury conservation magazine cover + collectible national-park poster, visually inspiring but informative.

Format: Vertical 4:5
Style: Cinematic, realistic, premium editorial, conservation-inspired, detailed, elegant, atmospheric
Avoid: Crowds, excessive text, cartoon style, artificial-looking animals, tourist buses, logos, watermarks, and clutter.

Structure used as instruction. The panels, the labels and the closing line are all specified before any scenery, which is why the finished piece reads as a layout instead of a photograph.

Prompt and image published by @Naiknelofar788; the original post is the linked source. Quoted as posted, with markdown emphasis and surplus blank lines removed. · Third-party example. Text and image remain the work of the original author; no reuse licence is granted here.

Editorial poster in which oversized letterforms become physical structures that a figure moves through
Prompt behind this image
Create a bold editorial campaign poster where oversized typography becomes a physical space the subject can move through and interact with.

Brand: [BRAND NAME]
Product / Campaign: [PRODUCT OR CAMPAIGN]
Main Word: [SHORT WORD]
Subject: [PERSON / PRODUCT / BOTH]
Spatial Action: [VAULT / WEAVE / EDGE / SPIN / OTHER]
Movement: [DESCRIBE THE SUBJECT’S ACTION]
Main Color: [COLOR]
Accent Color: [COLOR]
Aspect Ratio: [9:16]

Use realistic commercial photography, oversized typography, strong perspective, high-contrast colors, and contemporary editorial design.

Make the main word extremely large and transform its letterforms into a believable spatial structure. The subject must physically interact with the typography instead of simply standing in front of it.

Use the selected spatial action as the main visual logic:

 VAULT: turn the letters into obstacles the subject can physically vault over.
 WEAVE: arrange the letters at alternating depths so the subject moves between them.
 EDGE: use narrow letter edges or surfaces as a path the subject carefully moves along.
 SPIN: arrange the typography into a circular or curved track the subject moves around.

Create clear foreground, midground, and background relationships. Let some parts of the typography pass in front of the subject while other letters recede behind them.

Keep the typography clearly readable. Use shallow, controlled depth rather than heavy 3D sculpture. Letter surfaces should be clean, matte, evenly colored, and sharply defined, with no distressed, chipped, dirty, or weathered texture.

All graphic lines, motion paths, callout lines, and curves should be continuous, smooth, and clearly rendered.

Keep the person highly realistic with natural body proportions, believable movement, real skin texture, natural highlights, and detailed hair. Avoid waxy skin, excessive shine, CGI appearance, distorted limbs, oversized hands or feet, and extreme wide-angle deformation.

Use one dominant high-impact color with black, white, and one contrasting accent color. Add only a few meaningful campaign details such as the brand name, product name, slogan, and 2–3 short feature callouts.

The final image should feel like a finished commercial campaign key visual where the typography itself becomes the environment and the subject’s movement explains the word.

A template more than a description. The bracketed slots are the part worth imitating, because they force you to decide the brand, the word and the movement before the model invents them for you.

Prompt and image published by @MrLarus; the original post is the linked source. Quoted as posted, with markdown emphasis and surplus blank lines removed. · Third-party example. Text and image remain the work of the original author; no reuse licence is granted here.

Watercolour illustration blended with photographic detail along a flower-lined stone path at sunset
Prompt behind this image
A breathtaking, high-detail photograph blended with an artistic watercolor illustration of a serene and magical landscape, featuring a stone path lined with blooming pink flowers and glowing lanterns on stone pillars. The scene captures a tranquil atmosphere under a vibrant sunset sky with soft lighting, blending crisp photographic realism with soft paint textures, bleeding edges, and gentle artistic splatters, captioned with handwritten cursive text reading "Different Horizons Same Peace" alongside a small heart symbol.

A short prompt with a large effect. Almost all of it describes medium and edge treatment rather than the scene, which is the opposite of how most people write a first attempt.

Prompt and image published by @Minahil42298354; the original post is the linked source. Quoted as posted, with markdown emphasis and surplus blank lines removed. · Third-party example. Text and image remain the work of the original author; no reuse licence is granted here.

What an image-to-prompt tool does, and what it cannot do

An image-to-prompt tool looks at a picture and writes a description of it in the grammar an image generator expects. It describes what is visible. It cannot hand you the prompt that produced the picture, because that sentence was never stored in the pixels.

Those two jobs get confused constantly, and the confusion is why results feel disappointing. A description of a photo and the instruction that produced it can look almost identical on a page, yet only one of them was written by the person who made the image.

Everything on this page takes the second job as the real one. You get a structured, reusable description assembled from vocabulary a model responds to, written so that you can edit it rather than merely admire it.

  • It can name what is visible: subject, pose, wardrobe, setting, framing, lens feel, light direction, palette, medium and finish.
  • It has to guess the exact prompt, the seed, the sampler, the model version, and whatever the crop removed.
  • It can read generation metadata out of the file itself, whenever that metadata survived, without guessing at all.

What one upload returns

The result is a scene description broken into slots, so you can delete a clause without rewriting the sentence around it.

Detail level is a dial, not a quality slider. A concise prompt is short enough to retype from memory. Professional detail is the working default. Maximum detail is built to reproduce one specific picture, and it will contain clauses you should delete before reuse.

Set the output language to match the model that will read it. Tag-driven pipelines are written in English far more often than not, while models tuned for Chinese or Japanese input accept a description in their own language.

  • Subject and count. Who or what is in frame, how many, what they are doing, and where the eye lands first.
  • Appearance and wardrobe. Hair, expression, clothing, material and colour, written as description rather than identity.
  • Setting and depth. Foreground, midground, background, and what is deliberately out of focus.
  • Camera and framing. Aspect ratio, viewpoint, apparent focal length, and how tight the crop sits.
  • Light. Direction, hardness, colour temperature, and the shape of the shadow.
  • Palette and finish. Dominant colours, contrast, grain, and whether the surface reads as photo, paint, print or 3D.
  • Lettering. Any text visible in the image, and where it sits in the layout.
  • Negative prompt, only when you switch it on.

Four steps from a photo to a prompt you can reuse

The work takes less time than deciding what to do with the output.

  1. Choose the image on purpose. A clean picture with one clear subject converts better than a busy screenshot. If the palette is what you want, pick an image where the palette is not competing with a crowd.
  2. Set the target model before uploading. The same scene is written differently for a tag-driven model and for one that reads full sentences, and the target also decides whether the prompt ends in parameters.
  3. Read the output as an editor. Delete clauses about things you do not care about, and rewrite any sentence that names the specific subject instead of the treatment.
  4. Generate once, then compare and change a single clause at a time. Keep the version that survives a second generation, not the first one.

Prompt order changes with the target model

This is the part that decides whether the result is usable, and the part most reverse-prompt pages skip.

TargetHow it reads a promptWhat to put first
General image modelFull sentences, weighted roughly by positionDeliverable, subject, then composition and light
Nano BananaConversational instruction with an image attachedThe instruction verb, then the subject, then the change you want
GPT ImageDescriptive paragraphs, strong at lettering and layoutFormat and placement, then subject, then any words to render
MidjourneyComma-separated fragments with a parameter tailSubject, medium, light, then flags such as an aspect ratio
Stable Diffusion / FluxComma-separated tags, strongest tokens firstSubject and medium, then style tags, then a separate negative prompt

Some images still carry their original prompt inside the file

Before reversing anything, spend ten seconds checking whether the answer is already sitting in the file. Generation tools often wrote it there without telling anyone.

Stable Diffusion WebUI stores the full positive prompt, the negative prompt, the sampler, the step count, the CFG scale, the seed, the image size and the model hash in a PNG text block named parameters. The same build writes an EXIF UserComment when you export a JPEG or a WebP. ComfyUI embeds the entire workflow graph, which is a better record than the prompt text on its own.

That block is fragile. Re-uploading an image to Discord, X, Reddit, Facebook or Telegram normally strips it, and so does most screenshotting and re-encoding. If your image has already travelled through a social platform, there is nothing left to read.

Midjourney works differently because it leaves no prompt in the file. A seed exists, but only as an integer between 0 and 4294967295 that you retrieve by reacting to your own job with the envelope emoji, and a seed carries no style or character. Change the prompt, keep the seed, and you get a similar composition rather than the same picture.

  1. In a terminal, run exiftool on the file and look for a Parameters, Comment or UserComment field.
  2. Or open the image in any text editor and search for the words Negative prompt. In an untouched PNG that string is often readable near the end of the file.
  3. On Windows, the Details tab of the file properties shows the same EXIF fields, though a generation block rarely survives a trip through a browser download.
  4. If you do find the block, copy it out before anything else. It reproduces the source image far more faithfully than any description can.

Four prompts to copy, sorted by model family

These were written for this page as starting points. They are not the original instructions behind any example in the example gallery, and they will not reproduce those images.

General image model
A 4:5 editorial still life of one glazed ceramic bowl on rumpled linen, lit by a single window from the left so the shadow falls to the right. Warm neutral palette, shallow depth of field, visible film grain. Keep the upper two thirds of the frame empty.
Midjourney
single glazed ceramic bowl on rumpled linen, one window from the left, hard directional shadow to the right, warm neutral palette, medium format film, shallow depth of field, editorial still life --style raw --stylize 50 --ar 4:5
Stable Diffusion / Flux, positive
medium format film photograph, glazed ceramic bowl, rumpled linen cloth, single window light from the left, hard directional shadow, warm neutral palette, shallow depth of field, fine grain, editorial still life
Stable Diffusion / Flux, negative
clutter, extra bowls, plastic texture, floating objects, harsh flash, oversaturated colours, text, watermark, low detail, blurry

Free, no sign-up, unlimited: what those words mean here

Search results for this task are full of pages promising free, unlimited and login-free. On a site that runs real inference, it is worth knowing what each of those words actually buys.

Reading a prompt out of an image is an image-understanding call. It costs the same kind of compute as any other vision request, so a page offering it without limits is usually rate-limiting you quietly, substituting a smaller model, or paying for it some other way.

Here the tool runs on the same account and point balance as the rest of the workspace. Sign in, and the workspace shows the fields and the price of the request before you send it.

  • Reading this page needs no sign-up. Running the tool does, because the request is billed.
  • No unlimited claim. The cost follows the target model and settings you choose, and the workspace shows it first.
  • Accepted image types and the per-request limit are shown in the workspace before your file leaves the page.

How close the result will be, and where it drifts

A reversed prompt is an interpretation, not a recovery. Expect the same subject in a recognisably similar treatment, and expect specific details to move.

  • Faces drift. Woman in her twenties is not a person, and the next generation will invent a different one.
  • Text drifts most. Lettering is usually described rather than transcribed, so short words survive better than long ones.
  • Brand marks, logos and product shapes are described generically. Asking for a named brand is not something this tool does.
  • Fine texture and exact colour shift between model versions even when the prompt is byte-identical.
  • Composition is the most reproducible element when you also fix the aspect ratio and the seed, and the least reproducible when you change both.

Six uses beyond copying a look

Most people arrive wanting to remake one image. The same output does several other jobs.

  • Learning the vocabulary. Reading structured descriptions of images you like is the quickest way to stop writing prompts as paragraphs of mood.
  • Moving a style between models. Describe an image once, then rewrite the description for the target model's grammar instead of guessing at it.
  • Briefing a human. An illustrated prompt is a sharper brief for an illustrator or designer than a mood board on its own.
  • Rebuilding an old project. When a client sends a finished piece and no source files, a reverse prompt gets you a comparable starting point.
  • Auditing your own library. Describing your own images exposes how repetitive your prompts have become.
  • Alt text drafts. Accessibility descriptions and generation prompts share raw material, though alt text still needs a human edit before it ships.

Rights, consent, and photographs of real people

An image is not automatically yours to describe and reuse, and the tool cannot tell you whether you have permission.

  • Use images you made, images you licensed, or images whose licence allows the reuse you have in mind.
  • Do not reverse-prompt a photograph of an identifiable person in order to recreate them without consent.
  • A lookalike of a real person, or a brand-owned character, is a legal question rather than a technical one.
  • The examples on this page are third-party work published with attribution, and no reuse licence travels with them.

Questions

Can an image-to-prompt tool recover the exact prompt behind a picture?

Sometimes, but not by looking at the picture. If the file still contains its generation metadata, the original text is there to read: Stable Diffusion WebUI writes the prompt, sampler, steps, CFG scale and seed into a PNG text block, an EXIF UserComment on JPEG and WebP exports, and ComfyUI embeds a whole workflow graph. Re-uploading that image to a social platform normally strips all of it, and no vision model can reconstruct a prompt it has never seen. What this page returns is a new description of what is visible, which is why it lands close rather than identical.

Is image to prompt free here, and do I need an account?

Reading this page is free and needs no account. The tool itself signs in through the site account, because the request is billed, and the workspace shows the point cost for your chosen target model and settings before you submit. There is no unlimited-free claim on this page.

Does the reversed prompt work in Midjourney?

Yes when you set the target to Midjourney, because the output is rearranged into comma-separated fragments with a parameter tail. Do not expect it to reproduce the original job. Midjourney does not store a prompt in the exported file, and its seed, an integer between 0 and 4294967295 that you retrieve by reacting to your own job, carries composition rather than style or character.

Why does my new image look similar but not identical?

A description and an instruction are different objects. The model reading your reversed prompt has never seen the source image, the sampler and seed are new, and any detail the description left vague gets invented. Change one clause per attempt and keep the version that still works on a second run.

Can it read the text inside an image, or keep a face unchanged?

Short words are usually recognised and described, while long passages and unusual spellings are frequently paraphrased. Faces are the least reproducible element of all, because a description of a person is not that person. Where a specific face, logo or word has to survive, edit the original image instead of regenerating it.

Which files can I upload, and how many?

One reference image per request, in JPEG, PNG or WebP. The workspace states the current limits next to the upload field rather than here, because they follow the live service configuration.

Should the output be in English if I generate in Chinese?

Match the language to the model that will read the prompt. Tag-driven pipelines are overwhelmingly written in English, while models tuned for Chinese or Japanese input work well with a native-language description. Changing the output language changes the vocabulary, not the scene.

Is this the same as image captioning or alt text?

Related but not identical. A caption summarises an image for a reader in one sentence. A generation prompt orders the same observations so a model can act on them, which is why it names the lens, the light and the finish instead of stopping at what is happening.

Are my uploads stored?

Uploads are covered by this site’s privacy policy, which sets out what stays in the browser, what the services retain and how a deletion request is handled. This page repeats none of it: treat the upload field as you would any online tool, and use images you are comfortable sending.

Can I reverse-prompt a photo of a person I found online?

Only if you hold the rights and, for an identifiable person, their consent. The tool will describe what it sees either way, but a generated lookalike of a real person raises legal and ethical questions that no setting on this page can resolve.

Explore next