Good prompts get six things right: subject, lighting, a camera or technical reference, context, quality and style anchors, and negative instructions where the model supports them.
- Lead with the subject. Black Forest Labs says FLUX.2 pays more attention to what comes first; Google’s Veo formula opens with the camera instead.
- Name the lighting. Left unspecified, lighting is chosen for you, and often comes back flat and even.
- Quality tags cut both ways. Tags like 8K pull towards an advertising look; delete them for documentary realism.
- Check negative support first. If a model lacks it, describe the fix instead, like visible skin texture.
- For LLMs, name the format. Word counts or a markdown table constrain the output like a camera does for images.
The six elements hold. How to ask AI well covers the habits around them, and reusable context covers what to keep.
Across image generators, video models and LLMs, the difference between a prompt that works and one that does not comes down to six things. Not ten. Not twenty. Six. The reasoning for each is given below rather than asserted from experience — this site is anonymous and does not trade on anyone’s CV. Miss one and the output suffers; get all six right and results become repeatable rather than lucky.
These elements apply to every type of AI prompt. The specific words change depending on whether you're prompting an image model or an LLM, but the underlying structure doesn't.
A Clearly Defined Subject
The model needs to know exactly who or what it's generating. Vague subjects produce generic outputs. "A woman" gives the model infinite interpretations. "1woman, late 20s, Southeast Asian, long dark hair, angular jawline, confident expression" removes nearly all of them.
For image prompts: describe age, ethnicity, specific features, and expression. For LLM prompts: define the role precisely — "You are a senior copywriter at a luxury fashion brand targeting women aged 25–40" is a defined subject. "You are a copywriter" is not.
The subject should usually come early, because order matters on at least some models — Black Forest Labs states this outright for FLUX.2, that it “pays more attention to what comes first.”Black Forest Labs FLUX.2 prompting guide, read at source 17 Sep 2026: “Word order matters - FLUX.2 pays more attention to what comes first.” An earlier version of this note quoted the guide’s negative-prompt sentence instead, and said all models weight earlier tokens more heavily.
Not every vendor agrees. Google's published formula for Veo opens with the camera, not the subject: [Cinematography] + [Subject] + [Action] + [Context] + [Style & Ambiance]. If you are prompting video on Veo, lead with the shot. Formula per vendor.
Put a tightly defined subject first, unless the vendor’s own formula says otherwise, as Google’s does for Veo.
Specific Lighting
Lighting does a great deal of the work in an image, and it is often left out of prompts. "Good lighting" means nothing to a model. "Soft Rembrandt key light from upper left, warm fill from the right, subtle rim light" means something specific.
Named lighting setups the model understands: Rembrandt, clamshell, chiaroscuro, golden hour, blue hour, window light, practical neon. Describing the light works better than borrowing a name: a director’s or cinematographer’s name transfers a look unpredictably, and some tools restrict imitating living artists’ styles. See director styles in prompts for how to translate a name into light, lens and palette.
If you leave lighting unspecified, the model chooses for you, and the result is often flat, even illumination — close to a passport photo.
A Camera or Technical Reference
For image and video prompts: naming a camera and lens is a quick way to signal a photographic look. "Shot on Canon EOS R5, 85mm f/1.2" tells the model: high resolution, shallow depth of field, natural bokeh, the compression of a portrait lens. It reads the camera as a shorthand for an entire visual aesthetic.
You don't need an exact model number. A camera family works too: "Shot on medium format Hasselblad" or "ARRI ALEXA anamorphic" names a look that is common in photographs and film stills.
For LLMs: the equivalent is specifying format. "Output as a markdown table" or "write in exactly 80 words, no more" is the technical reference that constrains the output space.
Context and Environment
Where is this happening? What surrounds the subject? The environment frames the subject and determines how natural the output looks. A portrait without environmental context gets placed against whatever background the model decides — and it usually decides wrong.
Environment also carries mood. "Rain-soaked alley at midnight with neon reflections" communicates color palette, lighting quality, atmosphere, and emotional register in one phrase. "Outdoor setting" communicates nothing.
For LLMs: context is the situation, audience, and background. "I'm pitching to a room of 40 retail buyers who know nothing about AI" is context. "Write a pitch" is not.
Quality and Style Anchors
Quality tags tell the model what kind of image to aim at. Terms like "8K, ultra-detailed, commercial photography standard" pull output toward polished, professional, advertising-grade imagery, because that is the material they are associated with in training.
Know which way you are pulling, because this cuts both ways. That polished look is the opposite of documentary realism — if you are trying to make something look like an actual photograph rather than a campaign image, these are the first words to delete. See why AI output looks fake, which covers the training objective behind the glossy default.First-hand: an earlier version of this element recommended these tags without qualification, which contradicted this site’s own guidance on realism. Both effects are real; the tags help for commercial polish and hurt for realism, and the page now says which is which. Corrected 31 Aug 2026.
Style anchors go further — they reference a specific aesthetic. "Vogue editorial," "A24 film," "campaign photography for a luxury brand" each invoke a rich visual reference that the model can draw on. These work because the model has processed vast amounts of content tagged with these references.
Choose quality tags for the look you want: they pull toward polished advertising imagery, so remove them when you want something that looks like a real photograph.
Negative Instructions
Telling the model what to avoid is often useful — but how you do it, and whether you can at all, depends entirely on the model. LLMs take explicit exclusions in the prompt. Video models mostly want identity-lock instructions rather than a ban list. Image models split: Stable Diffusion and ComfyUI pipelines expose a real negative-prompt field, while Black Forest Labs states that FLUX.2 does not support negative prompts and asks you to describe what you want instead.Black Forest Labs, FLUX.2 prompting guide, read at source 23 Sep 2026: “FLUX.2 does not support negative prompts. Focus on describing what you want, not what you don’t want.” First-hand: an earlier version of this element said avoiding was "as important as" including and presented negative prompting as a general image-model technique — which contradicted this site’s correction of 25 Aug 2026. Corrected 31 Aug 2026.
For models that do take them: CGI, plastic skin, airbrushed face, watermark, blurry, overexposed, cartoon. For models that do not, the same list still works as a checklist for the positive prompt — "visible skin texture and pores" does the job that "no plastic skin" was doing, and it works everywhere.
For LLMs: "No filler phrases. No exclamation marks. Don't start sentences with 'I'. Avoid corporate jargon." Explicit exclusions prevent the model from defaulting to its most common patterns, which are usually the patterns you're trying to avoid.
Check whether your model reads negative prompts before writing a ban list; if it does not, describe the result you want instead.
More prompts, when something changes.
Prompts, model guides and workflow notes, sent when there is something new. Free.
SUBSCRIBE FREE ↗