The short version: ChatGPT doesn’t edit your photo pixel by pixel like a filter, it rebuilds a new image based on your original plus your description, so the trick is writing prompts that name a lens, a light source, a colour grade, and a mood rather than just typing “make it cinematic.” Get specific about those four things and you’ll get results that look like a film still instead of an Instagram filter.
What “cinematic” means when you’re typing it into a chat box
Everyone says “make this look cinematic” and gets a slightly darker, slightly orange version of their original photo back. That’s because “cinematic” isn’t a style, it’s a shorthand for about six technical choices that films make and phone photos usually don’t. Films have controlled lighting, a shallow depth of field, a deliberate colour grade (usually teal shadows and orange skin tones), visible grain, letterboxing or a wide aspect ratio, and a lens with slight imperfections like flare or vignette.
When you ask ChatGPT for a cinematic edit, you’re really asking it to guess which of those six things you mean. If you don’t specify, it picks the most generic combination it’s seen a million times, which is why so many AI “cinematic” photos look identical: warm highlights, crushed blacks, a bit of haze. To get something that looks intentional rather than templated, you need to name your ingredients.
The prompt formula I use every time
After months of testing this for client work and my own LinkedIn content, I’ve settled on one structure that works about 80% of the time on the first or second try:
- Subject and action – what’s in the frame and what it’s doing
- Lighting source and direction – “single window light from the left,” “harsh overhead fluorescent,” “golden hour backlight”
- Lens behaviour – “shot on a 35mm lens with shallow depth of field,” “slight anamorphic flare,” “35mm film grain”
- Colour grade – “teal and orange grade,” “desaturated with warm highlights,” “muted Wes Anderson palette”
- Reference point – a director, cinematographer, or film era, used loosely as a style anchor rather than a demand for a copy
A prompt that works looks something like this: “Edit this photo of a woman at a desk, keep her pose and outfit, light her from a single window on the left with soft shadows, shoot it like it’s on 35mm film with visible grain, grade it with cool shadows and warm skin tones, feel similar to a Roger Deakins interior shot.” That’s specific enough that ChatGPT has real constraints to work within instead of guessing.
My own test: a boring headshot from a hotel room
I did this for the first time in a Manchester hotel room before a speaking gig. Bad lighting, a beige wall, a photo taken on my phone in about four seconds by a colleague. I uploaded it to ChatGPT and asked for “cinematic lighting” as my first attempt. What I got back was a warmer, slightly blurred version of the same flat photo. Fine, but not different enough to use.
Second attempt, I named the ingredients: “keep my face, hair, and outfit exactly the same, relight this as if it’s a single practical lamp glowing warm on the right side of my face with the left side falling into shadow, add a very light film grain, grade it teal in the shadows and warm on the skin, like a quiet dramatic film still, not a movie poster.” That version had actual shape to the light. It looked like it belonged on a speaker page rather than a hotel selfie.
I ran nine versions total before I had one I’d put on my own site. That’s the number worth remembering: expect somewhere between five and ten passes before a cinematic edit looks intentional rather than accidental. Anyone telling you one prompt nails it every time hasn’t done this enough times to hit the versions that come back with warped hands or a face that’s subtly not yours anymore.
The uncomfortable bit nobody selling “AI photo editing courses” tells you
Here’s the part that gets glossed over constantly: ChatGPT isn’t editing your photo. It’s generating a new image that’s inspired by yours. That distinction matters more than most guides admit, because it means your actual face, your actual background text, your actual product label can shift slightly with every pass. I’ve had a client’s shop signage turn into gibberish letters after a “cinematic” edit, and I’ve had my own nose reshape itself by maybe 10% in a version that otherwise looked stunning. If you need pixel-perfect accuracy, for a product shot with real branding or a headshot that has to match your passport photo, ChatGPT image editing is the wrong tool and you want actual photo editing software with layers and masks, not a prompt.
Where ChatGPT earns its place is anywhere “close enough and better than the original” beats “technically accurate.” Social content, mood boards, blog headers, speaker page images where nobody’s checking your bag strap pixel by pixel. Know which category your photo falls into before you start, because I’ve watched people spend two hours chasing a “perfect” cinematic edit of a product photo that needed Photoshop from the start.
Step by step: doing this from your phone
- Pick a photo with a clear subject and a simple background. Busy backgrounds confuse the regeneration and you’ll get warped detail in the corners.
- Upload it to ChatGPT and open with a locking instruction: “keep the subject, face, pose, and clothing exactly the same, only change the lighting and colour grade.”
- Name one light source and one direction. Don’t say “moody lighting,” say “single hard light from above and slightly behind, like a bare bulb.”
- Add a lens or film reference: “shot on 50mm, shallow depth of field, subtle grain like 400 ISO film.”
- Name a colour grade in plain terms: “teal shadows, warm orange skin tones” beats “cinematic colours” every time.
- Ask for one version, look at it, then correct in the next message rather than starting over. “Keep this exact lighting but pull the grain back by half” gets you closer faster than a fresh prompt.
- Expect to run through several versions. Budget real time for this, ten to fifteen minutes minimum if you want it to look intentional.
Words that move the needle
Some phrases consistently produce better results than the generic “cinematic” request. From testing on my own content and on striking portrait prompts for client headshots, these are the ones that pull their weight:
- “Rembrandt lighting” for a triangle of light under one eye, good for a serious, thoughtful portrait
- “Practical light source” when you want the light to look like it’s coming from something in the room, a lamp, a window, a screen, rather than a studio setup
- “Anamorphic flare” for that horizontal streak of light you see in blockbuster night scenes
- “Bleach bypass” for a desaturated, high-contrast, gritty war-film look
- “Kodak Portra tones” for warm, slightly muted skin colours without going full orange filter
- “Letterboxed 2.39:1” if you want the black bars top and bottom that instantly read as “film” to most viewers
Naming these terms works because the model has seen thousands of images and descriptions tagged with these exact words, so you’re pointing it toward a specific visual pattern instead of a vague vibe.
Cinematic edits for content, not just headshots
This isn’t only about making yourself look moody for LinkedIn. I’ve used the same approach for seasonal content, giving a plain product photo a warm, low-key Christmas-film grade instead of the usual red and gold Instagram filter, the same logic sits behind the festive photo prompts I put together for December campaigns. The formula doesn’t change: name the light, name the grade, name the reference, lock the subject.
For anyone building out a wider style for their photos, it’s worth reading up on portrait style prompts so you have a consistent look across a whole set rather than one lucky cinematic shot sitting oddly next to ten flat ones. Consistency across a set matters more than any single image looking dramatic, particularly if you’re building out a personal brand where people scroll past six photos of you in one sitting.
Why this works the way it does, and where it breaks
It helps to understand roughly how image generators build a picture from text, because it explains why cinematic prompts sometimes go wrong in specific, predictable ways. The model is pattern matching against millions of tagged images, so if your prompt uses a term that’s ambiguous, “moody,” “aesthetic,” “epic,” it has too many patterns to choose from and picks an average, which is exactly the bland result most people get. Specific terms narrow the pattern space. That’s the whole trick, and it’s the same trick that makes text-to-video AI tools produce better motion when you name a camera move instead of saying “make it dynamic.”
Where it breaks is fine detail and text. Jewellery, watch faces, tattoos with actual words, product labels, anything with fine repeated pattern tends to distort across a regeneration. If your cinematic edit needs to keep a specific logo crisp, do the lighting and grade work in ChatGPT and then swap the label back in with real editing software afterward. Two tools, one result, faster than fighting the model for the twentieth version.
Common mistakes that waste your time
- Asking for “cinematic” with no other detail and being surprised when you get the same warm-orange look everyone else gets
- Uploading a photo with a cluttered background and expecting clean edges around your subject
- Not locking the subject at the start of the prompt, which lets the model drift your face or outfit further with every follow-up message
- Chasing perfection on a photo that needed real editing software from the start because it has text, logos, or fine detail that has to stay exact
- Giving up after one version instead of treating this as a five to ten pass process
Related reading: funny prompt chatgpt.
Related reading: fun chatgpt prompts.
Related reading: ai photo editing strengths.
For the rest of the ChatGPT prompts questions, see my ChatGPT prompts guide.
Related reading: how to create 80s look in chatgpt.
Related tool: my free tool that writes the prompt for you.
Frequently asked questions
Can ChatGPT edit my existing photo or does it just generate a new one?
It generates a new image based on your upload and your prompt, it doesn’t move pixels around like Photoshop, which is why small details like text, logos, and exact facial features can shift slightly between versions.
What’s the single best prompt phrase for a cinematic look?
There isn’t one magic phrase, but naming a specific light source and direction, a lens or grain reference, and a colour grade in plain terms beats any single word like “cinematic” or “moody” on its own.
How many versions should I expect to run before it looks right?
Budget for five to ten passes. My own headshot took nine tries before the lighting, grain, and colour grade all worked together without warping my face.
Is ChatGPT good enough for product photos with logos or text?
Not reliably. Fine detail like text and logos tends to distort during regeneration, so do the lighting and colour grade in ChatGPT and then fix the logo or text in real editing software afterward.
