The short version: ChatGPT doesn’t edit your photo, it repaints it, so your prompt needs to describe light, contrast, grain and subject separation as if you’re briefing a darkroom printer, not asking for a filter. Skip the word “monochrome” on its own and you’ll get flat grey mush every time. Add tonal range, a named film stock or photographer’s style, and one clear light source, and the difference is immediate.
Why most AI black and white edits look flat
I tested this last month with a set of headshots I’d taken for a client’s LinkedIn refresh. Nice photos, good light, nothing wrong with them. I typed “make this photo black and white” into ChatGPT and got back something that looked like it had been drained of colour rather than converted. Grey face, grey background, grey jumper, no separation between any of them. Technically black and white. Not striking. Not even close.
Here’s the bit nobody tells you: when ChatGPT “edits” a photo, it isn’t adjusting pixels the way Lightroom or Photoshop does. It’s generating a brand new image informed by your original and your words. That means a lazy prompt gives you a lazy, low-contrast desaturation. A specific prompt gives you something that looks like it belongs on a gallery wall. The gap between those two outcomes is entirely down to how much visual information you put in the prompt, because the model has nothing else to go on.
The four things a striking prompt always includes
After running maybe sixty variations on the same handful of photos over a few weeks, I narrowed it down to four elements that show up in every result I kept.
- Tonal range, not just “black and white”. Say “full tonal range from deep black to bright white” or “high contrast with true blacks and clean highlights”. Without this the model defaults to a narrow band of mid greys, which is the single biggest reason your edits look dull.
- A light direction. “Side lighting”, “window light from the left”, “harsh overhead sun”. Black and white photography lives and dies on shadow shape. If you don’t specify where the light comes from, the model flattens everything evenly and you lose the drama.
- A named reference point. This is the one that changed my results the most. Instead of “dramatic black and white”, I started writing “in the style of a 1970s Magnum Photos print” or “like a Peter Lindbergh portrait” or “Kodak Tri-X film grain”. Naming an actual visual tradition gives the model a texture and contrast level to aim for instead of guessing.
- Grain or texture instructions. Digital black and white without grain often looks sterile, almost clinical. Adding “fine film grain” or “subtle grain, not noisy” brings warmth back in.
A prompt structure that works
I use the same skeleton every time now, and I’d suggest you do too rather than reinventing it for each photo:
- Subject and what to keep unchanged (pose, expression, composition)
- Tonal range and contrast level
- Light source and direction
- Style or era reference
- Grain or texture
- What to avoid
Here’s the exact prompt that turned my client’s flat headshot into something she used as her main LinkedIn photo:
“Convert this photo to black and white, keeping the pose, expression and framing exactly as they are. Full tonal range from deep black to bright white, strong contrast. Light appears to come from the left, soft window light. Style it like a classic editorial portrait, similar to a 1980s Vanity Fair black and white shoot. Add fine film grain, subtle, not noisy. Avoid crushing the shadows on her hair and avoid making the skin look plastic or airbrushed.”
That last line matters more than people expect. Without it, roughly seven times out of ten the skin comes back smoothed in a way that reads as fake rather than moody. Telling the model what to avoid is doing half the work of telling it what you want.
Specific styles worth naming (and what each one gives you)
Vague style words waste your prompt. Specific ones earn their place.
- “Ansel Adams zone system contrast” gives you deep blacks, bright whites, very little mid-grey, dramatic landscapes with almost sculptural shadow.
- “Kodak Tri-X 400 film grain” gives visible, slightly gritty texture, closer to documentary photography than polished studio work.
- “Herb Ritts studio lighting” gives smooth, sculpted shadows on skin and bodies, good for portraits, less good for street scenes.
- “Film noir lighting, hard shadows” gives strong directional light, dramatic shadow shapes across the face or background, quite theatrical.
- “High key, mostly whites, minimal shadow” gives an almost overexposed, airy look, works well for beauty and product shots.
Pick one, not three at once. I made this mistake early on, stacking “Ansel Adams contrast, film noir shadows, high key lighting” into a single prompt, and the model just picked whichever instruction it liked best and quietly ignored the rest. One clear style reference outperforms three competing ones every single time.
Step by step: from flat photo to striking edit
This is the exact process I now run through with every photo:
- Start with a well-exposed original. Black and white forgives colour problems but not exposure problems, so a photo that’s too dark or blown out to begin with will stay that way.
- Write the base prompt using the six-part skeleton above.
- Generate one version and look specifically at the shadows, not the overall vibe. If the shadows are grey rather than black, add “deepen the blacks, increase contrast in the shadow areas.”
- Check the skin or main subject separation from the background. If they blend together, add a light direction that contrasts with the background tone.
- Ask for one small adjustment at a time rather than rewriting the whole prompt. “Same as before but add more grain” works better than starting from scratch.
- Save the wording that worked. I keep a running note of prompts by style because rebuilding a good one from memory wastes ten minutes I don’t have.
The mistake that ruins most black and white prompts
People treat “black and white” as the whole instruction. It’s the least useful word in the prompt because it tells the model almost nothing about mood, contrast or texture, and those three things are what make a black and white photo striking rather than just colourless. Colour photography and black and white photography are different disciplines with different rules, and a prompt that would work fine for a colour edit (“make it warmer, more vibrant”) is useless here because warmth and vibrancy don’t exist in monochrome. You need contrast words, light words, and texture words instead.
The other mistake, and this one’s harder to admit, is expecting pixel-perfect editing at all. If you need the exact original file with only the colour removed and nothing else touched, ChatGPT is the wrong tool full stop, and you’re better off with a proper editing app that does true desaturation. ChatGPT is brilliant for reimagining a photo in a new visual language. It’s not a replacement for non-destructive editing, and treating it like one is where people end up disappointed with faces that look subtly wrong or details that have quietly shifted.
Where this fits into a wider content workflow
I’ve started using AI-edited black and white images specifically for LinkedIn carousels and blog headers where a striking, slightly editorial photo does more work than a standard colour shot. It’s a small but useful piece of a bigger shift I’ve written about in what’s changing in AI marketing this year, where visual content generated or edited with AI is moving from novelty to normal practice a lot faster than most people expected. The tools are improving monthly. The prompting skill underneath them, understanding light, contrast and reference styles, stays useful regardless of which tool you’re using next year.
Related reading: ai photo editing strengths.
Frequently asked questions
Can ChatGPT edit my existing photo or does it create a new one?
It creates a new image informed by your original, it doesn’t do true pixel-level editing, which is why faces and fine detail can shift slightly and why specific, detailed prompts matter so much more than they would with real editing software.
Why do my ChatGPT black and white edits look flat and grey instead of striking?
You’re likely only saying “black and white” without specifying tonal range, light direction or a style reference, which leaves the model defaulting to a narrow, flat band of mid greys instead of true blacks and whites.
What’s the single best word to add to a black and white prompt?
“Contrast” paired with a specific level, such as “high contrast with deep blacks and clean bright whites”, does more to fix flat results than any other single addition.
Should I name real photographers or film stocks in my prompt?
Yes, naming something specific like “Kodak Tri-X grain” or “Ansel Adams style contrast” gives the model an actual visual target instead of a vague mood, and it consistently produces more distinctive, striking results than generic style words.
