PromptLens

FreeBuild sheet 05 of 08 · Concept site · fictional company

HALIDEHow the site was made

Every photograph melts into the next.

How we made a concept site for a fictional photographer and filmmaker: every prompt exactly as it was sent, the steps in order and how long it took.

THE MELT · FOUR OF THE FIVE PHOTOSFOUR PICTURES
A woman with a short auburn crop in a charcoal coat looks back over her shoulder against a rust-red wall, a smear of motion beside her face.A single drop of clear serum falls from a glass pipette into an open bottle, against cold blue.Black and white: two hands hold a phone on a small gimbal, and the screen shows one wide eye.Two hands hold a vintage twin-lens camera open from above; its ground glass shows an empty studio stool under a lamp.Generated with AI
Four of the five photos the site melts between, faded into each other here. Photo 3, The Long Table, is a real photograph by Dan Michael Sinadjan on Unsplash, so it is not in this strip. Pictures generated with AI.
01 · The idea

One idea, one moment, one job.

The idea is what the page is about. The moment is the one thing built for this business alone. The job is what it does for the owner.

The idea

One long exposure.

Scroll through the work and each photograph smears sideways along its own light, its three colour layers slip out of register, and the next frame develops out of the smear shadows first, like a print in the tray. It is drawn live in the page, already moving before anyone scrolls, and the page's background takes the colour of each photo.

The one moment

The pointer is a slow shutter

Move the pointer across a frame and the picture streaks in the direction it moves. Stop, and it sharpens again.

What it does for the owner

Request the date

  • Three packages: a portrait session, a brand content day, a UGC pack
  • A calendar that shows the price and the deposit for the date you pick
  • A proofing sheet where the client picks frames
  • Questions and a contact form
02 · The steps

In the order we did them.

Each step says what we did and why. Open Details under a step for the tools and the numbers behind it.

  1. Pick the mood before the prompts

    Pick one reference picture on Pinterest for each photo you need, and write the prompts from what it shows: the light, the colour and the mood.

    Details
    • Pinterest
    • image analyzer

    One reference pin was chosen for each project. Our image analyzer read each pin, the five prompts were written from that reading, and they were approved word for word before any call. The pins are references only: none is on the site.

  2. Make five frames at full size

    Ask an image model for five tall photographs at its largest size, from the words alone.

    Details
    • nanobanana-pro
    • 4K
    • 4:5

    Calls 01 to 05, no reference. A short script sent them to the same image endpoint our tool uses, reading each prompt from the approved prompt file, so every word is as approved. Each came back at 3,712 × 4,608 px.

  3. Open every picture at full size

    Open every picture at full size before you use it: two had faults that only show up close (a ghost face, and a face that should have been hidden), so they were made again.

    Details
    • review

    Frame 1 carried a second, ghost face in its smear, and frame 4 showed the face behind the phone. Both were made again, calls 06 and 07; frame 4 failed the same way.

  4. Rewrite three prompts

    When a round of pictures loses the mood you wanted, rewrite the prompts instead of rolling again: three were rewritten, and two of them came back right.

    Details
    • nanobanana-pro
    • 4K

    The first round had lost its references' mood, so three prompts were rewritten, calls 08 to 11. Frames 1 and 4 passed. The restaurant frame failed twice: the plate filled most of the width.

  5. Try an edit, then use a real photograph

    When no prompt gets a picture right, use a real photograph instead: the restaurant frame is a licensed photo from Unsplash, credited on the site.

    Details
    • nanobanana-pro
    • edit

    Call 12 asked the model to zoom out from the restaurant frame, and it did not. A real photograph went in instead: Fish dish by Dan Michael Sinadjan, used under the Unsplash License and credited on the site.

  6. Finish each frame

    Crop every photo to the same shape with its subject at the same height, and add the film look as a separate step.

    Details
    • Python
    • 4:5 crop
    • film finish

    Each frame was re-cropped to 4:5 with its subject at mid-height, so the melt reads as one object changing, and given a film finish as a separate pass.

  7. Build the melt

    Write the melt as a small graphics program in the page, so each photo smears into the next as you scroll and the pointer streaks the picture.

    Details
    • WebGL

    The frames melt in a shader: highlight trails, the three colour layers slipping apart, the next frame developing shadows first. The pointer streaks the frame the way it moves.

  8. Audit twice, from scratch

    Check the page twice from scratch, on phones, a laptop and with motion off, and fix everything the first check finds.

    Details
    • site walk
    • axe
    • copy_check

    The first pass found twelve defects and fixed all of them, among them a hero melt too blurry to read and trails that stepped into six visible copies. The second, at 320, 375, 414 and 1,280 px and with motion off: 0 errors, 0 outside requests, no sideways scroll, axe 0. Lighthouse on a phone, measured later: 96 to 99 for speed, 100 for accessibility.

03 · The calls

Every prompt, exactly as sent.

A call is one request to an AI model: the words, plus any pictures sent with them. Each prompt is copied from our records, never retyped, beside what it made. The exact settings sit under each one. Eight calls made pictures the site does not use: they are listed, and their prompts are left out.

Call 01 · not usedFrame 1, first take: a second, ghost face showed in the smear. Not used.

Call 02: Frame 2 · Salt & Skin

A picture, from words alone

A single drop of clear serum falls from a glass pipette into an open bottle, against cold blue.Generated with AI
Made: picture 02The second frame of the melt, re-cropped so the drop sits at mid-height.

Nano Banana Pro, an image model, made this picture from the words below alone.

The prompt, exactly as sent

A macro product photograph of a single drop of clear serum falling from the tip of a glass pipette into the open, threaded neck of a frosted pale-blue glass serum bottle with no label, the drop suspended in mid-air between the pipette tip and the bottle mouth, perfectly round with a small tail, catching a silver-white specular highlight on its top and a soft glow through its interior. The pipette comes down from the top of the frame on the vertical centre line, its matte black rubber bulb just cropped at the top edge; the bottle's threaded glass collar sits below, each helical ridge catching a fine line of light, the frosted body softening into gentle out-of-focus blur at the bottom. A cold monochrome blue world: a smooth gradient backdrop from deep navy at the top to pale ice blue behind the bottle, a soft box key light from above and slightly behind, a thin cool rim light along the glass edges, clinical-luxury skincare photography, extremely shallow depth of field with focus split between the drop and the top of the bottle neck, telephoto compression. Avoid: a label, printed text or logo on the glass, two drops or duplicate drops, a deformed or flattened drop, a warm or yellow cast, a green tint, plastic-looking glass, a CGI render look, hands, harsh hard shadows. Vertical 4:5 frame, the subject's visual centre at exactly mid-height and horizontally centred, fine film grain, photorealistic. No text, no logo, no watermark.

The exact settings
tool
generate_image
sent by
script, POST /api/generate-image
provider
nanobanana-pro
aspectRatio
4:5
resolution
4k
referenceImageUrls
none
sent
25 Sep 2026, 02:58:17 UTC
prompt length
1,447 characters

Call 03 · not usedThe restaurant frame, first take. It passed review, then the first round was judged to have lost its references' mood; on the site this frame is a real photograph. Not used.

Call 04 · not usedFrame 4, first take: the face behind the phone was readable. Not used.

Call 05: Frame 5 · The contact

A picture, from words alone

Two hands hold a vintage twin-lens camera open from above; its ground glass shows an empty studio stool under a lamp.Generated with AI
Made: picture 05The last frame of the melt, where the booking sits.

Nano Banana Pro, an image model, made this picture from the words below alone.

The prompt, exactly as sent

Looking down into the waist-level viewfinder of a vintage twin-lens reflex medium-format film camera held in two hands, seen from slightly above so the square ground-glass screen faces the viewer at the centre of the frame, its folding metal hood open around it. On the ground glass glows a small, warm, softly focused image of an empty photographic studio: a single wooden stool under one soft light in front of a plain backdrop, waiting for the next sitter, with the faint fine grid lines of the focusing screen visible over it. A matte black body with textured leatherette, chrome knobs and a film-advance crank catching a gentle specular highlight, the camera unbranded with a blank nameplate; warm hands at the sides, a soft blurred taupe-brown wall behind. Low warm ambient indoor light, an amber and sepia palette, the ground-glass image glowing brighter than its surroundings, a tactile analog feel, a frame within a frame, shallow depth of field with the viewfinder and chrome sharp and the background falling away, 1970s documentary film photography. Avoid: a brand name, maker's logo or any lettering on the camera, a modern digital camera, a toy-like plastic body, people or faces inside the viewfinder, distorted or extra fingers, a ground-glass image that looks pasted on, a cold blue cast, harsh flash, garish saturation. Vertical 4:5 frame, the subject's visual centre at exactly mid-height and horizontally centred, fine film grain, photorealistic. No text, no logo, no watermark.

The exact settings
tool
generate_image
sent by
script, POST /api/generate-image
provider
nanobanana-pro
aspectRatio
4:5
resolution
4k
referenceImageUrls
none
sent
25 Sep 2026, 03:00:18 UTC
prompt length
1,497 characters

Call 06 · not usedFrame 1, second take: it passed review, and the rewritten prompt of call 08 replaced it. Not used.

Call 07 · not usedFrame 4, second take: it failed the same way as call 04. Not used.

Call 08: Frame 1 · Weight of a Glance, rewritten

A picture, from words alone

A woman with a short auburn crop in a charcoal coat looks back over her shoulder against a rust-red wall, a smear of motion beside her face.Generated with AI
Made: picture 08The hero, and the first frame of the melt.

Nano Banana Pro, an image model, made this picture from the words below alone.

The prompt, exactly as sent

An editorial fashion photograph for an independent magazine: a woman in her late twenties with short, tousled auburn hair cut close at the neck, wearing a long charcoal wool coat with the collar turned up. Her body is turned three-quarters away from the camera toward the left edge, and her head is turned back over her right shoulder toward us, chin slightly raised, eyelids lowered, her gaze resting just past the lens, lips closed and relaxed. It is a slow-shutter exposure with a flash at the end of the turn: her eyes, nose and mouth are frozen crisp, while a soft band of her hair and cheek streaks horizontally out to the right at eye level in pale warm streaks, the motion of the turn. She is small in the frame: her head sits at the exact centre, her shoulders and coat fill the lower third and are softly blurred by the movement, and the upper half of the frame is empty, evenly lit rust-red seamless paper. One soft frontal light, low contrast, open shadows that never go fully black, the coat reading as a deep greenish charcoal; a palette of rust red, muted oxblood and warm ivory skin. A medium telephoto at a wide aperture, focus falling off quickly so only her face is sharp and the rest of the picture is gently soft. Avoid: a second face or duplicated eyes inside the smear, a frontal passport pose, looking straight into the lens, a smile, glossy beauty retouching, hard shadows, a black border or frame edge, jewellery, leather, a fringe. Vertical 4:5 frame, the subject's visual centre at exactly mid-height and horizontally centred, a real unretouched photograph. No text, no logo, no watermark.

The exact settings
tool
generate_image
sent by
script, POST /api/generate-image
provider
nanobanana-pro
aspectRatio
4:5
resolution
4k
referenceImageUrls
none
sent
25 Sep 2026, 03:34:01 UTC
prompt length
1,617 characters

Call 09 · not usedThe restaurant frame, rewritten prompt: the plate filled about 60% of the width. Not used.

Call 10: Frame 4 · Second Take, rewritten

A picture, from words alone

Black and white: two hands hold a phone on a small gimbal, and the screen shows one wide eye.Generated with AI
Made: picture 10The fourth frame of the melt.

Nano Banana Pro, an image model, made this picture from the words below alone.

The prompt, exactly as sent

A low-contrast black-and-white photograph, extremely close: a young man's two hands hold a smartphone on a small matte black handheld gimbal out at chest height, the phone's screen turned toward us and filling most of the frame. The screen shows a live close-up of one of his eyes, wide and unblinking, the brow slightly furrowed, filling the screen edge to edge, with no icons, numbers or interface of any kind. The screen is dim, no brighter than the skin of his hands, the image on it slightly soft. One hand grips the gimbal handle below the phone, the other steadies the phone's edge with a thumb and two fingertips. Behind the phone there is only the dark, out-of-focus knit of his sweater; his face is above the top edge of the frame and not in the picture. Dim, flat room light, grey tones throughout, no deep blacks and no bright whites, the look of a quiet self-portrait made at night. A normal lens close up, shallow focus on the screen, the hands slightly softer. Avoid: a face, eyes or skin anywhere outside the phone screen, on-screen icons, text, numbers or recording symbols, a brand logo or lettering on the phone or the gimbal, extra or malformed fingers, glossy product-shot lighting, a bright glowing screen, any colour tint. Vertical 4:5 frame, the subject's visual centre at exactly mid-height and horizontally centred, a real unretouched photograph. No text, no logo, no watermark.

The exact settings
tool
generate_image
sent by
script, POST /api/generate-image
provider
nanobanana-pro
aspectRatio
4:5
resolution
4k
referenceImageUrls
none
sent
25 Sep 2026, 03:35:23 UTC
prompt length
1,404 characters

Call 11 · not usedThe restaurant frame, rewritten prompt, second take: the same, and a stray streak of light. Not used.

Call 12 · not usedAn edit of the picture from call 09, asking the model to zoom out. It did not zoom out. Not used: this frame is a real photograph on the site.

04 · The log and the time

Every call, and how long it took.

The log lists every call, used or not. The times come from the log and our saved versions, and say so where they are estimates.

One line per call, in the order they ran. Four made a picture or film the site uses; eight did not, and they are listed anyway.

Time

About 1 h

To make every picture, in three rounds and one edit, on 25 Sep 2026.

Details

Every call: 02:56 to 03:56 UTC, 25 Sep 2026, in three rounds and one edit.

1 h 49 min

From the first picture to the site's first working version.

Details

From the first call to the site's first commit at 04:45 UTC, wall clock.

About 1 h

The build itself, as we wrote it down at the time. Not timed.

Details

The build itself, as written in our register. Not timed. The real photograph went in at 05:09 UTC.

05 · The honest line

It took more than one prompt.

Twelve prompts to one model, eight of them not used, and one frame no prompt got right, so it is a real photograph. Then the work no prompt does: every picture opened at full size, four frames finished by hand, the melt written as a shader, and two audits.

Optional

The next build sheet, by email.

Every sheet is free and open like this one, whether you sign up or not.

Your address is kept for these emails only. Privacy

For business owners

A site like this, for your business.

The sites on these sheets are concepts. For a local business we build the real thing: a site that takes the order, the booking or the call.

See the websites