15 copy-paste prompts that turn any photo into something people stop scrolling for β
written to hold the face, the pose and the background instead of quietly redrawing them.
Read them here, or take the PDF.
Take the whole thing as a PDF
All 15 prompts, the troubleshooting table and the free-access guide β 20 pages, 141 KB.
Yours to keep and to share.
Read this page first β it is the difference between these working and not
Every prompt in this pack is an edit prompt, not a create prompt. That means you upload a photo first, then paste the prompt. If you paste one of these into an empty chat with no image attached, the model will invent something from scratch and you will think the prompt is broken. It isn't. It just had nothing to work on.
Each prompt is written as a single block. Copy the whole thing β including the boring-looking parts about lens length, shadow direction and what not to change. Those clauses are doing most of the work. Image models drift: they slim faces, whiten teeth, restyle clothing and quietly redraw the background unless you explicitly forbid it. Roughly half of every prompt here exists to stop that drift.
Where you see text in [SQUARE BRACKETS], that part is yours to fill in. Replace the brackets and everything inside them with your own words β don't leave the brackets in.
If the first result is close but not right, do not argue with the model in follow-up messages. Edit one clause in the prompt and run it again from a fresh upload. Conversational patching is where good results go to die.
Which model?
Each prompt lists the model it was written for. Nano Banana Pro (Google) is the strongest at rebuilding a scene while keeping a face intact, and it is the only one you can trust with readable text in an image. ChatGPT / GPT Image is better at painterly and stylised transformations, and is more forgiving of vague instructions. Where a prompt says either, use whichever you already have open. The Where to run these, free section at the back shows how to reach Nano Banana Pro without paying for anything.
Fix and restore
Four prompts for making a photo you already have look like it was taken on better equipment, on a better day.
1
True-Detail Upscale
Nano Banana Pro or ChatGPT
Recovers real detail from a small, compressed or screenshotted photo without changing anything about it.
Best for Old phone photos, saved-from-WhatsApp images, screenshots, anything that looks soft or blocky when you zoom in.
Re-render the supplied reference image at maximum resolution as a faithful photographic upscale. This is a detail-recovery task, not a re-imagining: identity, facial geometry, apparent age, expression, gaze direction, pose, hand position, framing, crop and camera perspective must all be reproduced exactly. Reconstruct genuine micro-detail β individual hair strands, eyelashes, skin pores, fabric weave and stitching, printed text, wood grain, brushed metal β as it would have been recorded by a full-frame sensor at the same focal length. Eliminate JPEG blocking, colour banding, chroma noise and upscaling halos. Keep the original white balance, colour response, contrast curve and shadow depth; the only thing that changes is clarity and dynamic range. Do not relight, do not restyle, do not beautify, do not slim or reshape any feature, do not add makeup, jewellery, teeth or catchlights that are not already there. Do not invent detail in areas that are genuinely out of focus β leave the natural depth of field intact. Skin must stay skin: no wax, no gloss, no airbrushing. Avoid: warped hands, extra fingers, drifting facial likeness, duplicated background objects, added text or watermarks, painterly rendering, HDR crunch, over-sharpened edge halos.
Pro tip. Likeness drifts hardest when the face in your source is under about 200 pixels wide. If it comes back as a stranger, crop tighter on the face first and upscale that, then use the result as the reference for a second pass.
2
Invisible Retouch
Nano Banana Pro or ChatGPT
Professional skin retouching that keeps pores, freckles and lines β the opposite of a beauty filter.
Best for Portraits for a profile, a website team page, or any photo where you want to look rested rather than airbrushed.
Retouch the supplied portrait to a high-end editorial standard while keeping the edit invisible. Remove only temporary features: active blemishes, transient redness, oily shine, dust and sensor spots, stray lint on clothing. Keep everything permanent β freckles, moles, scars, laugh lines, under-eye structure, skin texture and pores must all survive at full strength. Even out blotchy tone by reducing colour variance, never by flattening luminosity; the natural highlight-to-shadow transition across the cheek, nose and jaw must stay intact. Preserve micro-contrast so the skin still reads as a surface with grain. Do not reshape the nose, jaw, eyes, lips or body. Do not whiten teeth or eyes beyond a subtle clean-up. Do not add makeup, glow, bloom or a beauty filter. Do not alter hair colour, hairline, clothing or background. Keep the original lighting direction, colour temperature and depth of field exactly. The result should look like the same photograph taken on a better day β someone comparing the two should not be able to say what was changed.
Pro tip. If it still overdoes it, add "apply the retouch at roughly 50 percent strength" as the last line. Stating an intensity works surprisingly well on every current model.
3
Old Photo Rescue
Nano Banana Pro
Repairs tears, creases, stains and fading in a scanned photo β and adds believable period colour to black-and-white.
Best for Family photos, inherited albums, anything scanned off paper. The single most emotionally valuable thing on this list.
Restore the supplied scanned photograph as a careful archival conservation, not a re-creation. Repair the physical damage: fill tears, creases, emulsion cracks, missing corners, water stains, mould spotting and scanner dust by reconstructing what the surrounding image logically continues into. Where damage crosses a face, rebuild only what anatomy makes unambiguous, and keep that reconstruction conservative. Correct fading and colour cast, recover shadow and highlight detail, and re-establish neutral greys. If the source is black and white, add historically plausible colour: period-accurate clothing dyes, muted natural fabrics, believable skin undertones matched to the visible lighting, era-appropriate paint and vehicle colours, natural foliage and sky. All colour must be restrained and slightly desaturated, like a well-preserved period print β never modern, saturated or digital. Preserve the original film grain structure, lens softness and depth of field of the era. Do not modernise faces, do not change expressions, apparent ages, hairstyles, clothing or the number of people present, do not sharpen into artificial crispness, and do not add objects, text or borders that were not in the frame.
Pro tip. Crop off the white scan border before you upload. Left in, models frequently try to "restore" the paper edge into the picture and you get a phantom wall running down one side.
4
Clean Cutout, New Background
Nano Banana Pro
Moves a subject onto a new background with a real cutout β hair strands intact, shadow correct, no halo.
Best for Product listings, profile shots, anything where the background is doing you no favours.
Isolate the main subject of the reference image and place it on [DESCRIBE THE NEW BACKGROUND IN ONE SENTENCE]. Cut the subject with true edge accuracy: individual hair strands, fur, fabric fuzz and every semi-transparent area β glass, lenses, veils, liquid, smoke β must carry across with their transparency preserved rather than filled in. No white or dark halo, no rubbery outline, no missing contact shadow. Re-light the subject so it belongs in the new scene: match the new background's light direction, colour temperature, intensity and softness onto the subject, cast a physically correct contact shadow and ambient occlusion where it meets the surface, and add a matching reflection if that surface is reflective. Keep the subject's own shape, proportions, colour identity, texture and detail completely unchanged β only the light falling on it changes. Render the background at a depth of field consistent with the subject's apparent focal length. Avoid: floating subjects, hard pasted-on edges, shadow direction that disagrees with the light, duplicated limbs, cropped fingers, added text or logos.
Pro tip. Describe the new background by its light, not just its content: "a sunlit oak table by a north-facing window" gives the model everything it needs to relight the subject. "A nice table" gives it nothing.
Look professional
Four prompts that turn a phone snap into something you would put on a shop listing, a press page or a profile.
5
Agency Headshot
Nano Banana Pro
Converts a casual photo into a studio headshot, keeping your actual face, hair and clothes.
Best for LinkedIn, casting profiles, company about-pages, author bios.
Convert the supplied photo into a professional headshot as a commercial portrait studio would shoot it. Keep the person's face, hairstyle, glasses, facial hair, clothing and accessories exactly as they appear in the reference β this is a change of lighting and framing, not a makeover. Reframe to a head-and-shoulders crop: subject centred, chin level, shoulders turned slightly off-axis, eyes to camera, expression relaxed and confident with a trace of a smile. Render as if shot on an 85mm lens at f/2.8 from eye height, with sharp focus on the near eye and ears and shoulders falling gently soft. Light with a large softbox key just above eye level and slightly to one side, a white fill card opposite to keep the shadow side open, and a subtle hair light for separation from the background. Two clean catchlights in the eyes, a soft shadow under the jaw, no hotspot on the forehead or nose. Background: seamless mid-grey studio paper with a soft gradient falloff β no texture, no props, no visible edges or corners. Grade the colour neutrally and accurately: true skin tones, controlled contrast, no teal-and-orange styling. Retouch to magazine standard but keep pores, fine lines and freckles visible. The output should pass as a genuine studio capture.
Pro tip. If it keeps changing your clothes, name them: "the subject is wearing a navy crew-neck sweater β reproduce it exactly". Models redesign anything you don't describe.
6
White-Room Backlight
ChatGPT
Blown-out white background with a hard rim light wrapping the subject β the fitness and fashion editorial look.
Best for Gym and transformation photos, dance and sport shots, high-contrast portraits.
Re-light the supplied image as a high-key studio portrait against a blown-out white background. Place a powerful, lightly diffused source directly behind the subject so the background reads as pure clipped white and the light wraps around the silhouette as a bright rim along the shoulders, arms and hair. Let a controlled amount of that backlight bloom into the frame as soft atmospheric haze, heaviest close to the subject's edges. Keep the front of the subject noticeably darker than the background for depth, lit only by a soft low-intensity fill so that form, muscle definition, fabric folds and texture all stay readable rather than falling into silhouette. Increase overall contrast and deepen the shadow side. Preserve the subject's identity, pose, clothing and proportions exactly. Skin stays matte and textured β no gloss, no oil sheen, no plastic. Remove every element of the original background. Photorealistic, high resolution, editorial lighting.
Pro tip. This one is dramatic by design. If the subject goes too dark, add "raise the fill light so the face is clearly exposed" β the model will hold the rim and lift the front.
7
Product Hero Shot
Nano Banana Pro
Re-photographs your product as a commercial hero image with correct material behaviour and space for a headline.
Best for Etsy, Shopify, Amazon listings, ad creative, anything you shot on a kitchen table.
Re-photograph the product in the reference image as a commercial hero shot for an online store. The product's shape, proportions, colour, material, labelling and every word of printed text must be reproduced exactly as they appear in the reference β do not redesign, re-letter or "improve" the packaging. Stage it on [SURFACE β e.g. a warm oak worktop / a matte concrete slab / a seamless off-white sweep], shot slightly above eye level on a 100mm macro lens at f/8 so the entire product is sharp front to back. Light with a large overhead softbox plus a narrow strip light down one side to define the edge, and a subtle bounce card to open the shadow at the front. Render a true contact shadow directly beneath the product plus a soft ambient shadow around it, correct specular behaviour for the actual material β matte card absorbs, glass refracts and carries a bright edge highlight, metal shows a gradient reflection β and physically plausible reflections of the surroundings. Colour-accurate, neutral white balance, clean commercial grade. Compose with generous negative space above the product for a headline. Avoid: warped or invented text, altered logo shapes, a floating product, doubled or contradictory shadows, background clutter, props that were not requested, wide-angle distortion.
Pro tip. Text on packaging is where these break. Use Nano Banana Pro, and if the label matters, retype it in the prompt: "the label reads exactly: SEA SALT & THYME".
8
Brand Mockup
Nano Banana Pro
Puts your artwork onto a physical product β shirt, mug, framed print β with real fabric drape and print texture.
Best for Print-on-demand and digital-download shops that need listing images without ordering samples.
Apply the supplied artwork onto [PRODUCT β e.g. a heavyweight cream cotton t-shirt / an 11oz white ceramic mug / a framed A2 poster on a plastered wall] as a realistic product mockup. The artwork must be reproduced exactly: same colours, same proportions, same crop, same typography β nothing redrawn, re-lettered or re-coloured. Wrap it onto the object with correct geometry, following the curvature of the surface, the drape and folds of the fabric, and the perspective of the frame, and let the print take on the material underneath so that weave texture, ceramic sheen or paper tooth shows faintly through the ink. Print edges should be very slightly soft, the way real screen or direct-to-garment printing is, never a hard digital paste. Light the scene naturally with soft window light from one side, cast a correct shadow, and keep the object in a clean, neutral, uncluttered setting shot at f/4 so the background falls gently out of focus. Photorealistic commercial product photography, colour-accurate. No added logos, no text on the product other than the supplied artwork, no warped or re-rendered lettering.
Pro tip. Upload the artwork on a transparent or plain white background with no mockup around it. If your reference is already a mockup, the model will cheerfully mockup the mockup.
Transform the world
Four scene-rebuild prompts. These are the ones that get shared β and the ones that most often come back as a cheap filter instead of a rebuild, which is exactly what the long "reconstruct, do not filter" clauses are for.
9
Sandbox City Render
Nano Banana Pro
Rebuilds your photo as a screenshot from an open-world driving-and-crime game β polygons, hard sun, punchy grade.
Best for Group photos, street scenes, car shots. Reliably the most-shared image in this pack.
Rebuild the entire reference scene as a real-time render from a modern open-world sandbox video game, captured as an in-game screenshot on a high-end PC at maximum settings. This is a full 3D reconstruction, not a filter: every person, vehicle, building, prop and surface must be rebuilt as game-engine geometry with visible polygonal structure, baked normal maps, tiling textures and optimised level-of-detail. Characters take slightly exaggerated, stylised proportions with rigid, sculpted features; skin is matte with no photographic pores; hair is built from clumped card geometry; clothing uses flat fabric shaders with visible seams and repeating texture. Architecture is modular with crisp edges, and roads, kerbs, street furniture and vegetation are dense but simplified and clearly asset-built. Lighting is real-time: one strong directional sun, hard-edged shadows, ambient occlusion in the corners, imperfect screen-space reflections, none of the softness of a ray-traced render. Grade the image with punchy contrast and lightly boosted saturation. The camera is a third-person gameplay or free-camera view with mild wide-angle distortion and stable framing β no depth of field, no motion blur, no lens flare, no photographic grain. Keep the original composition, subject placement and pose. The result must read unmistakably as a video game and never as a photograph. No HUD, no interface elements, no text, no watermarks.
Pro tip. Written without naming a game on purpose β trademarked titles trip content filters and are a liability if you post commercially. Describing the engine behaviour gets you the look without the brand name.
10
Voxel World
ChatGPT
Turns the entire background into a blocky cube world while leaving the person completely photorealistic.
Best for Holiday and outdoor photos with recognisable scenery. The contrast between real person and block world is the whole joke.
Keep the human subject from the reference image fully photorealistic and completely untouched β same face, pores, hair strands, clothing texture, expression, pose and body lighting. Apply no stylisation, no pixelation and no relighting to the person whatsoever. Rebuild everything else in the frame as a blocky voxel world constructed from uniform cubes, matching the original location's structure exactly rather than reinventing it. Reproduce the real layout, spacing, depth and geometry: a path keeps its exact curve and width in gravel-toned blocks; trees keep their trunk positions, heights and canopy spread in trunk and leaf blocks; buildings keep their silhouette, window placement and roof pitch in block geometry; benches, fences, rocks and signs become cube-built equivalents in the same positions and at the same scale relative to the subject. The horizon line, camera height, focal length, framing and subject scale must match the source photo precisely. Anything the subject touches, holds, sits on or stands on converts to voxel form at the correct scale for their hands and body, with correct contact shadows so they are grounded in the scene rather than pasted onto it. Voxel surfaces use flat, low-resolution square textures with a visible pixel grid and no smoothing or bevelling. Light the blocks from the same direction and at the same intensity as the original photo. The location must stay instantly recognisable from its composition alone.
Pro tip. Works best on wide shots with obvious landmarks. On a tight portrait there is almost no background to rebuild, and the effect vanishes.
11
Brick-Built World
ChatGPT or Nano Banana Pro
Same trick, built from moulded plastic bricks β studs, seams, glossy ABS and all.
Best for Landmarks, holiday photos, city streets, anything with recognisable architecture.
Keep the human subject from the reference image fully photorealistic and completely untouched β same face, pores, hair strands, clothing texture, expression, pose and body lighting, with no stylisation and no relighting applied to the person. Rebuild everything else in the frame as a structurally accurate replica assembled from moulded plastic construction bricks, matching the original location exactly rather than reinventing it. Reproduce the real layout, spacing, depth and geometry: a path keeps its exact curve and width in tile and plate construction; trees keep their trunk positions, height ratios and canopy spread using brick-built trunks and layered foliage elements; buildings keep their silhouette, window placement and roof angle in authentic brick geometry and proportions; rocks, benches, fences and signs become brick-built equivalents in the same positions and at the same relative scale. Perspective must match the original photograph exactly β horizon line, camera height, focal distance, framing and subject scale all identical. Anything physically touching the subject converts to brick logic: an animal becomes a moulded figure, a held object becomes a brick-built version correctly scaled to the hand. Render the plastic honestly: visible studs and tube undersides, mould seam lines, injection sprue marks, faint surface scuffs, and the specific semi-glossy specular response of ABS plastic. Light it as real daylight interacting with plastic, matching the direction and intensity of the original photo, with soft shadow falloff and subtle environmental reflections. Maintain recognisable background identity β the location should be identifiable from composition alone, simply rebuilt in brick. High-resolution, cinematic realism. No interface, no text, no overlays.
Pro tip. Never name the brick brand in a prompt for anything you plan to sell or run as an ad. Describing the plastic gets the same result with none of the trademark exposure.
12
Anime Frame
Nano Banana Pro
Redraws the photo as a frame from a hand-drawn anime film, keeping the composition and the likeness.
Best for Portraits, couples, pets, street scenes. Note the long negative list β that is what keeps it from going glossy AI-slop.
Redraw the reference image as a single frame from a high-budget hand-drawn anime feature film. This is an adaptation, not a new scene: composition, camera angle, focal length, crop, perspective, pose, gesture, gaze direction, subject placement, clothing layout and background structure must all match the reference exactly. Translate each person into an anime character who remains recognisably themselves β keep facial structure, apparent age, skin tone, hairstyle, hair colour, build and expression. Eyes are stylised but proportioned to the real face and pointed in the same direction; do not use oversized generic anime eyes. Simplify the nose and mouth in the anime idiom while preserving the likeness. Draw with clean, confident 2D linework of varying weight, accurate anatomy and a clear silhouette. Shade skin and fabric with soft cel shading in two to three tones, gentle gradients and highlights that follow the original light direction. Hair becomes defined strand groups that keep the original volume and flow. Clothing, props and accessories keep their real shapes, colours, logos and material weight. The background becomes a painted anime environment with the same depth, perspective, time of day and atmosphere as the reference. Final quality: theatrical anime production, crisp, richly coloured, film-still framing. Avoid: 3D or CGI rendering, western comic or cartoon style, oil or watercolour painting, semi-realistic digital painting, glossy AI sheen, smoothed plastic faces, changed identity, changed pose or crop, distorted hands, extra fingers, empty or simplified backgrounds, messy linework.
Pro tip. If faces come back generic, add one identifying detail per person: "the woman on the left has a gap between her front teeth and a small mole on her left cheek".
Make something people share
Three prompts with no practical purpose whatsoever, which is precisely why they perform.
13
Clay Miniature
Nano Banana Pro
Recreates the scene as a frame from a hand-made stop-motion film, complete with fingerprints in the clay.
Best for Pets, kids, couples, any scene with character. Warmer and less overused than the anime conversion.
Recreate the reference scene as a frame from a hand-made stop-motion film shot on a physical miniature set. Everything must read as a real object built by hand and photographed with a macro lens. Characters are sculpted from modelling clay over wire armatures, with visible fingerprint texture, thumb-smoothed surfaces, faint tool marks, slight asymmetry and a soft matte sheen. Hair and fur are sculpted or felted, never rendered. Clothing is real fabric at miniature scale, so the weave looks oversized and the folds sit stiffly. Sets are built from card, foam, wood and painted plaster with visible seams, brush strokes and traces of glue. Keep the original composition, pose, expression and character likeness clearly recognisable through the clay. Light the set with small practical lamps: a warm key from one side, a cool fill opposite, hard shadow edges, and a shallow depth of field with the pronounced macro falloff that makes the scale of the miniature obvious. A little dust hangs in the air. This is a photograph of a physical model β not a 3D render, not a cartoon, not a digital painting. No perfect symmetry, no glossy CGI plastic, no text.
Pro tip. The phrase carrying most of the weight is "visible fingerprint texture". Drop it and you get generic 3D-cartoon output instead of something that looks handmade.
14
Liquid Pixel-Stretch
ChatGPT
Keeps the subject razor-sharp while the rest of the frame dissolves into flowing ribbons of its own colour.
Best for Poster art, album covers, profile banners, anything you want to look expensive and abstract.
Turn the reference image into a gallery-grade abstract artwork built around an untouched subject. The main subject stays completely sharp, correctly lit and instantly recognisable β same identity, silhouette, colours and position in frame. From the subject's edges outward, dissolve the rest of the image into long ribbons of stretched pixels that pull the photograph's own colours and textures into smooth, sweeping curves, as though the image were being drawn out into flowing streams of light and paint. Transitions from photograph to stretch must be seamless, with no visible seam, no glitch blocks, no banding, no broken pixels and no hard edges. The flow should follow a deliberate composition β one dominant sweep with secondary counter-curves β rather than a random smear, and should leave clean negative space so the subject reads clearly. Preserve realistic depth, shadow and reflection where the stretch meets the subject. Take the entire colour palette strictly from the source image; introduce no new hues. Style: liquid motion blur, light-trail streaks, silky flowing energy, minimalist premium poster art, ultra-high resolution.
Pro tip. Works best on a subject with a clean silhouette against a simple background. Busy backgrounds turn into mud when stretched.
15
Candid Phone Selfie
Nano Banana Pro
Builds a convincingly unglamorous selfie of two people from two separate reference photos.
Best for Putting two people in one frame who were never photographed together.
Generate an unposed phone selfie of the people from the supplied reference images standing together, as though one of them is holding the phone. Faces must match the references closely β same bone structure, skin tone, hair and apparent age. Hold the camera at arm's length, tilted slightly up and off-axis, so the framing is a little crooked and one person sits closer to the lens than the other, with the mild wide-angle distortion that comes from that proximity. Both lean in shoulder to shoulder, relaxed, looking at the lens with easy half-smiles. Light it like a real indoor snapshot: mixed room light plus a direct phone flash that slightly blows out the nearest forehead and cheek, with a hard shadow thrown onto the wall behind. Keep every imperfection β visible pores, uneven skin tone, stray hairs, creased clothing, a shiny nose, faint sensor noise in the shadows, mildly incorrect white balance, a hint of motion blur toward the edges. The background is an ordinary everyday interior, cluttered in a normal way and slightly out of focus. Nothing about the image should look composed, retouched or professionally lit. A straight-out-of-camera phone photo: no filter, no beauty smoothing, no portrait-mode bokeh.
Pro tip. Upload both references in the same message and say which is which β "image 1 is the man on the left". Sent one at a time, the model usually forgets the first.
Why these prompts look the way they do
Every prompt in this pack is built from the same five blocks, in the same order. Once you can see the pattern you can write your own β which is the actual skill.
1. The task verb
Opens with rebuild / re-light / retouch / redraw / re-photograph β never "make a picture of". An edit verb tells the model the reference image is the subject, not just inspiration.
2. The lock list
Everything that must not change: identity, pose, crop, camera angle, clothing, colour balance. This is the single highest-value part of any edit prompt and the part everyone skips.
3. The build list
What to add or change, described in physical terms β materials, geometry, construction, surface behaviour. "Make it look like plastic bricks" is a wish. "Injection-moulded ABS with visible studs, mould seams and glossy specular highlights" is an instruction.
4. Camera and light
Focal length, aperture, camera height, light source, direction, hardness. Photographers' language moves these models more than adjectives do. "85mm at f/2.8, soft key just above eye level" beats "professional photo" every time.
5. The negative list
The specific failure modes for that effect, named out loud: plastic skin, extra fingers, warped text, invented background, HDR crunch. Naming a failure is what suppresses it.
When it goes wrong
Nine symptoms that account for almost every bad result, and the clause that fixes each one. Add the fix to the end of the prompt and regenerate from a fresh upload β don't try to correct it in conversation.
Symptom
The fix
The face is no longer them
Your source face is too small. Crop tight on the face and upscale that first, then use the result as your reference. Add: "the output must be recognisable as the same individual".
Text turned into gibberish
Switch to Nano Banana Pro and type the words out: "the sign reads exactly: OPEN DAILY". No other model handles lettering reliably.
It looks like a filter, not a rebuild
Add: "this is a complete reconstruction of the scene, not a filter, overlay or colour grade applied to the photograph".
Skin looks like wax or plastic
Delete any word like smooth, flawless or perfect. Add: "retain visible pores, skin texture and fine lines; matte finish, no gloss".
It changed the clothes
You didn't describe them, so it invented them. Name the garment, colour and neckline explicitly in the prompt.
Hands are wrong
Either crop them out of the reference, or say what they are doing: "both hands rest flat on the table, fingers visible and separated".
The effect is far too strong
State an intensity: "apply at roughly 40 percent strength". Almost every model respects a stated percentage.
The background got reinvented
Add: "match the layout, spacing and geometry of the reference exactly; do not reinterpret or reposition anything".
The result is soft or low-resolution
Run prompt 1 on the output. Generating and upscaling are separate jobs β asking for both in one pass gets you neither.
Where to run these, free
You do not need a paid subscription to use this pack β but where you run these matters far more than most guides admit. The same model, reached three different ways, gives you three wildly different free allowances. Start at the top of this list.
Google AI Studio β the best free option by a distance
Go to aistudio.google.com and sign in with a Google account. No credit card, no trial. Pick a Gemini image model, attach your reference photo, paste the prompt. You get a far larger daily allowance than the consumer app, higher output resolution, and no visible watermark β which is the whole reason to start here. This is the one to use if you are working through several prompts in a sitting.
The Gemini app β convenient, but watch the cap
The same family of models sits in the Gemini app and at gemini.google.com. Handy on a phone, and fine for a quick try. The catch: the free tier gives you only a couple of Nano Banana Pro generations per day, and those come out with a visible watermark stamped on them. The standard image model is more generous but less capable at holding a face. If your output arrives watermarked, you have hit this limit β switch to AI Studio.
ChatGPT / GPT Image
The free tier includes a small daily allowance of image edits. Attach the photo to the message, paste the prompt, send. If it refuses on a photo of a real person, that is a policy filter on likenesses rather than a problem with the prompt β run it on Google instead.
Don't bother with a free API key
Worth saying because it wastes an afternoon: Google's image models are not offered on the free API tier. Every one of them is paid-only for developers. The free allowance exists in the web interfaces above, not through code.
Two things worth knowing.Aspect ratio: set it before you generate, not after β every model here handles a stated ratio far better than being asked to re-crop something it has already made. Invisible watermarking: even where there is no visible mark, Google embeds SynthID and content-credential metadata in the file. It does not affect how the image looks or your right to use it, but it does mean these images are detectable as AI-generated, which is worth knowing before you use one somewhere that matters.
Why is this free?
Paper Plane Workshop is a one-person studio β apps, games, tools and guides. This pack is my way of saying hi.
If it was useful, share the page link rather than the file, so people always get the current version.
Then have a look at the apps or the
shop. π
Fine print: the images you make with these prompts are yours, commercial use included.
Share this page freely; don't resell the PDF. AI models change without notice, so results will vary over time.