Half Of Every One Of These Prompts Is A List Of Things Not To Do
The caption is the best line on the post: “Your photo doesn’t need a new face. It needs better direction.” Then seven prompts, posted in the comments, each turning an ordinary photo into something cinematic, editorial, DSLR-ish or studio-clean.
Read them and notice their shape, because it is unusual.
Here is prompt 2 in full: “Make my photo look like it was captured with a professional full-frame DSLR camera. Keep my face, body, expression, clothing, and identity exactly unchanged. Add realistic depth of field, natural skin texture, subtle film grain, and balanced exposure. Do not beautify, reshape, smooth, or alter my features.“
Four sentences. One says what to do. Two of them are defence. And that ratio holds across all seven – a short positive instruction wrapped in a long list of things not to touch.
That is not padding. It is the correct design, and it is the reason this card is worth more than the usual “make me look professional” prompt going round.
Because the failure mode of AI photo editing is not that the result looks bad. It is that the result looks wonderful and is no longer you. Softer jaw, straighter nose, symmetrical eyes, poreless skin – every one of those reads as “better” to a model optimising for attractiveness, and every one of them is a small deletion of the person. Nobody notices it in a single image. Everybody notices it when they meet you.
The post this article is about
Read the original post: “Your photo doesn’t need a new face. It needs better direction.”
The prompts are posted in the comments beneath. The tables below quote all seven as written, on 10 October 2026.
Count the prohibitions
Every prompt has three parts: the look it asks for, the preservation clause that lists what must stay, and the prohibitions that name specific ways it must not go wrong. Here is how much of each prompt is which.
Diagram: look, preservation, prohibition
| Prompt | The look it asks for | What must stay unchanged | Named prohibitions |
|---|---|---|---|
| 1. Cinematic Portrait | Lighting, contrast, colour grading, atmosphere | Face, identity, proportions, expression, clothing | “no artificial or overly stylized look” — 1 |
| 2. DSLR Look Upgrade | Depth of field, skin texture, film grain, exposure | Face, body, expression, clothing, identity | Do not beautify · reshape · smooth · alter — 4 |
| 3. Luxury Editorial | Lighting, composition, contrast, background separation | Facial structure, body proportions, identity, skin texture | “without making it look artificial” — 1 |
| 4. Golden Hour Outdoor | Warm sunlight, soft shadows, skin tones, depth of field | Identity, face, body, clothing, natural features | Avoid artificial glow · excessive warmth · oversaturation · obvious AI look — 4 |
| 5. Clean Headshot | Soft lighting, exposure, subtle background blur | Face, hair, expression, identity, skin texture | No beautification · face reshaping · excessive smoothing · artificial retouching — 4 |
| 6. Urban Street | Natural light, shadows, colours, background, grain | Face, identity, body proportions, expression, clothing | No face reshaping · artificial skin · animated effects — 3 |
| 7. Master Photographer | Exposure, lighting direction, skin tones, tonal range, sharpness, depth of field, composition, background separation | Identity, facial features, proportions, expression, hairstyle, clothing, skin texture, natural imperfections | Not retouched · plastic · AI-generated. Do not alter · beautify · reshape · replace — 7 |
Two things stand out. Column 2 is almost identical across all seven – every prompt asks for lighting, exposure and depth of field. The “styles” differ less than the names suggest; what actually changes between a cinematic portrait and a clean headshot is a handful of adjectives.
And column 4 is where the real work is. Prompt 7 alone names seven distinct prohibitions and is the only one that explicitly protects hairstyle and natural imperfections. That phrase – “keep natural imperfections” – is the single most useful instruction on the card, because imperfections are precisely what a model trained on attractive images will quietly remove.
All seven, and what each is permitted to touch
Column 3 is the one to read. The seven are not seven styles – they are seven different permissions, ranging from “change the colours” to “change the place I appear to be standing”.
| Prompt | Quoted as written | What it is actually allowed to change | Our note – read 10 October 2026 |
|---|---|---|---|
| 1. Cinematic Portrait Treatment | “Give my photo a premium cinematic treatment. Preserve my exact face, identity, proportions, expression, and clothing. Enhance the lighting, contrast, color grading, and atmosphere for a realistic camera-shot finish. Keep everything natural with no artificial or overly stylized look.” | Colour and mood only | Safest of the seven. Grading does not touch geometry. The weakest preservation clause, but it is also asking for the least |
| 2. DSLR Look Upgrade | “Make my photo look like it was captured with a professional full-frame DSLR camera… Add realistic depth of field, natural skin texture, subtle film grain, and balanced exposure. Do not beautify, reshape, smooth, or alter my features.” | Depth of field — which means inventing background blur | Watch the edges. Fake depth of field has to guess where you end and the background begins. Check your hair, glasses and shoulders against the blur |
| 3. Luxury Editorial Style | “Transform my photo into a high-end luxury editorial portrait. Preserve my exact facial structure, body proportions, identity, and natural skin texture… without making it look artificial.” | Lighting, composition, contrast | “Composition” is doing quiet work – changing composition can mean recropping or moving you in the frame. Only one prohibition on this one |
| 4. Golden Hour Outdoor Edit | “Transform my photo into a realistic golden-hour portrait… Add warm natural sunlight, soft realistic shadows… Avoid artificial glow, excessive warmth, oversaturation, or an obvious AI look.” | The light, and therefore the implied time of day | This one changes the scene, not just you. A photo taken indoors at noon now looks like an evening outdoors. See the disclosure section |
| 5. Clean Professional Headshot | “Turn my photo into a polished professional headshot… Use soft, realistic lighting, balanced exposure, and subtle background blur… No beautification, face reshaping, excessive smoothing, or artificial retouching.” | Lighting and background blur | The most practical one for a profile picture, and well guarded. “No beautification” is the clause doing the work |
| 6. Urban Street Photography | “Transform my photo into a realistic professional urban street portrait… Enhance the natural lighting, shadows, colors, and background while adding realistic depth of field and subtle camera grain… no face reshaping, artificial skin, or animated effects.” | “Enhance the background” – the place itself | The furthest-reaching prompt here. It is allowed to change where you appear to be standing. Fine for an avatar, not fine if the photo is evidence of anything |
| 7. Master Photographer Treatment | “…preserving my exact identity, facial features, proportions, expression, hairstyle, and clothing… Keep realistic skin texture, natural imperfections, and authentic details… not retouched, plastic, or AI-generated. Do not alter, beautify, reshape, or replace any part of my identity.” | Everything photographic at once | The superset. It contains the lighting, exposure, depth of field and composition work of the other six, with the strongest guard clause. Start here and add a style word |
A prompt is a request, not a guarantee
Here is the gap the card cannot close, and it is not the author’s fault.
“Do not reshape my face” is an instruction. An image model has no face-preservation lock that this sentence switches on – it has a strong steer, and steers are probabilistic. Writing the prohibition genuinely helps, which is why these prompts are good. It does not bind.
So the output needs checking, and there is no check anywhere on the card. Seven prompts, zero verification steps – the same shape we found in a build guide that assembled an agent and never asked it a question. The making is documented; the confirming is left to you.
Worse, you are the worst-placed person to judge this. You have seen your own face in mirrors – laterally flipped – your whole life, and you have a lifetime of opinions about which parts of it you would change. A model that quietly narrows your nose is handing you the version you half-wanted. You will like it. That is the problem.
So here is the check, and the last row matters most.
| What to compare | How | Why this one | Pass / fail |
|---|---|---|---|
| 1. Geometry | Flip between before and after at full size. Look at nose width, jaw line, eye spacing, hairline | These are the four things “improvement” moves first, and all four are identity | Fail if anything moved, even slightly in your favour |
| 2. Asymmetry | Real faces are lopsided. Check whether yours still is | Symmetrising is the most common silent edit and the hardest to spot without looking for it | Fail if you have become even |
| 3. The small marks | Moles, freckles, scars, a crooked tooth, laugh lines | Prompt 7 asks for “natural imperfections” by name. Check they survived | Fail if your skin is cleaner than it is |
| 4. Edges and extremities | Hair against the blurred background, glasses frames, ears, hands, jewellery | Invented depth of field has to guess where you end, and it guesses worst at fine edges | Fail if hair melts into the background |
| 5. The scene | Text, signage, other people, objects behind you | Prompts 4 and 6 are allowed to change the background. Make sure they only changed what you meant | Fail if anything behind you now says something different |
| 6. Somebody else | Show both versions to a person who knows your face. Do not say which is which | You cannot do this check yourself – you are flattered by the thing you are supposed to detect | Fail if they pick the edited one as “not quite you” |
Thirty seconds for the first five, one message for the sixth. And if a photo fails, the fix is usually not a better prompt – it is a weaker one. Ask only for the lighting, leave the rest alone, and run it again.
Do you have to disclose a relit photo?
If the output is going on a thumbnail or a channel banner, this is worth two minutes. We read YouTube’s own page on disclosing GenAI content earlier this week, and it names four triggers. Two of them are relevant here:
- content that “alters footage of a real event or place”
- content that “generates a realistic scene that didn’t actually occur”
The page also says explicitly that minor cosmetic edits do not need disclosing – colour correction, lighting fixes and similar adjustments are listed among the exemptions. And it makes one reassuring promise in writing: “Disclosing AI content won’t limit a video’s audience or impact its eligibility to earn money.”
Lay the seven prompts against that and the line falls in an obvious place. Prompts 1, 2, 3 and 5 adjust lighting, exposure, grading and blur – that is colour correction and lighting, squarely in the exempt category. Prompts 4 and 6 are different. Golden Hour makes an indoor noon photo look like an outdoor evening. Urban Street is permitted to “enhance the background”. Both move toward a realistic scene that did not occur.
None of this makes them wrong to use – a stylised avatar is nobody’s idea of a deception. But if the picture is doing any work as a record of a real moment, that is the moment to tick the box, and ticking it costs nothing.
Four honest caveats
The caption says six prompts. There are seven. They are numbered 1 to 7 in the comments. A small thing, and the kind of small thing worth noticing before you trust a bigger number on a bigger card.
Prompt 7 contains the other six. Master Photographer Treatment already asks for exposure, lighting direction, skin tones, tonal range, sharpness, depth of field, composition and background separation – which is the union of everything in prompts 1 to 6, with the strongest guard clause attached. You can treat the card as one prompt plus a style word: “…in a golden-hour outdoor setting”, “…as a clean studio headshot”. Fewer prompts, better protection.
We did not run them. Everything above is read off the prompts as written. We have not tested how any particular image model behaves with them, and behaviour varies enormously between models and between versions of the same model. The check in the table is there precisely because we cannot tell you the answer in advance.
And none of this applies to other people’s faces. Every one of these seven says “my photo” and “my identity”. Using them on somebody else’s portrait needs that person’s agreement, whatever the prompt allows – and if the subject is recognisable and the scene changes, you are into territory the prompt was never designed for.
The one-line version
These prompts are well made, and the reason is their shape: most of the words are defence. “Do not beautify, reshape, smooth, or alter.” “Keep realistic skin texture, natural imperfections, and authentic details.” Whoever wrote them understood that the risk in AI retouching is not an ugly result but a flattering one.
But a prohibition in a prompt is a request, not a lock. So the step the card is missing is the one that takes thirty seconds: flip between before and after, check geometry, asymmetry, small marks, edges and background – then show both to someone who knows your face and does not know which is which.
And if you only keep one: keep number seven, add a style word to the end of it, and delete the other six.
Learning a lot does not make anyone good. Doing a lot does. Saving seven prompts takes a second; running one and then actually checking whether the person in the picture is still you takes two minutes, and it is the only part that decides whether the photo is usable.
If you want to be held to that, we run a free challenge at BANI Academy. There is no barrier to entry – you commit a small fee that is refunded when you finish, and the only requirement is that you actually do the work for the length of the challenge.
Run one of these on your own photo, then show both versions to someone who knows you. Tell us in the comments under the original post whether they could tell – and whether they preferred the real one.
More on building a channel for a global audience at mmoyoutube.com.
One Comment