Models & Parameters

Prompt Engineering for Image and Video Models

Also called what is prompt engineering, prompt template, few shot prompting, system prompt, prompt weighting, prompting

Prompt engineering is the practice of writing and structuring model input so the output lands where you want it. For image and video models it is mostly about word order, specificity and weighting, because these models do not follow instructions the way a chat model does; they match a description.

What these models are actually reading

A diffusion model does not receive your sentence. A text encoder converts it into a vector, and that vector steers the denoiser toward images whose learned description is nearby. Everything practical follows from that.

It means the model is doing matching, not obeying. It has no way to hold "do not" as an operation, no way to keep a running plan, and no way to ask what you meant. So the whole craft is describing a finished frame precisely enough that only frames you would accept are nearby, then removing every word that does not narrow the set.

Chat models are the opposite, which is why prompt engineering advice from LLM work transfers badly. Techniques like few shot prompting and a carefully written system prompt are real and useful, but they belong to the language layer of a pipeline (an assistant that expands your idea into a shot description, or a captioner writing training data), not to the image model itself.

A prompt template that holds up

Order matters, because encoders weight earlier tokens more heavily and truncate the tail. Lead with what you are least willing to lose.

[shot size] of [subject] [action], [environment],
[lighting], [lens/optics], [grade/film stock], [motion for video]

Applied: Medium close-up of a fisherman mending a net, harbour wall, overcast diffused light, 85mm shallow focus, desaturated blue-grey grade.

What earns its place in a prompt:

  • Concrete nouns and verbs. "Mending a net" beats "doing something with rope".
  • Photographic specifics. Lens length, light direction and quality, time of day. These are heavily represented in training captions, so they land.
  • One clear subject. Two subjects with two descriptions bleed attributes into each other.

What almost never earns its place: stacked quality words (masterpiece, 8k, hyper detailed) which mostly shift style toward a generic look, emotional appeals, and instructions about the process rather than the picture.

Techniques ranked by how much they change the frame

  1. Cutting words. The highest yield edit available. Delete every clause you cannot see in the output and adherence on what remains improves.
  2. Reordering. Move the thing you care about to the front. Same words, different result.
  3. Negative prompt. Effective against recurring artifacts (watermark, extra limbs, text) and useless as a way to add control the model lacks.
  4. Prompt weighting. Useful for rebalancing two competing elements. Keep multipliers modest.
  5. Style anchors. A named medium, era, or process (Polaroid, cel animation, tungsten-lit interior) is far more reliable than a stack of adjectives, and more repeatable across a sequence.

When words are the wrong tool

The single most useful judgement in prompt engineering is recognising the ceiling. If the same problem survives three rewrites, no fourth wording will fix it, and you are burning credits proving it. Reach past the prompt:

ProblemThe move that works
Look is not what you picturedGenerate the frame as an image first, then use it as a first frame
Character changes between shotsReference image, fixed seed, or a trained LoRA
Composition keeps driftingA control input or a mask, not more adjectives
Text on signs is garbledRender it clean in post, or use a model with a strong text encoder
Motion is wrongOne camera move per clip, and separate shots that need two

The workflow habit that separates fast operators from slow ones: keep a small library of prompts that worked, with the model version and seed attached. A prompt template you have already validated on a given model is worth more than any general rule, because it encodes that model's quirks, and every model has its own.

The prompt for this

A starting point that reliably produces the effect. Adjust the subject and setting; keep the technical clauses.

Medium close-up of a fisherman mending a net on a harbour wall, overcast diffused light, 85mm, shallow depth of field, desaturated blue-grey grade, film grain

Try Prompt Engineering yourself

Open the generator with a starting point already filled in.

Frequently asked questions

What is prompt engineering in one sentence?
It is writing model input deliberately, with attention to what the model actually reads, so that more of what you asked for survives into the output. On visual models that means composing a description, not issuing commands.
Does prompt engineering work the same for image models and chat models?
No, and conflating them is the most common mistake. A chat model follows instructions and can reason about your request. A diffusion model matches a text embedding against images, so phrasing like please avoid or make sure that you does nothing. Describe the picture you want, not the task.
Do longer prompts give better results?
Only up to a point. Adherence degrades as clauses accumulate, and older text encoders truncate at 77 tokens, silently dropping the tail. A tight 30-word prompt usually beats a 120-word one. Add clauses one at a time and keep the ones that visibly change the frame.
How does prompt weighting work?
Some interfaces let you scale a term's influence, often with parentheses or a colon and a number, such as (neon sign:1.3). It multiplies that phrase's contribution to the conditioning. Small moves work; past roughly 1.5 you usually get artifacts rather than emphasis.
Why does the model ignore my negative instruction?
Because a text encoder has no reliable representation of negation. Writing no text on the sign puts sign and text into the conditioning. Use the negative prompt field instead, which is a separate conditioning path, or remove the concept from the positive prompt entirely.
Is prompt engineering still a useful skill?
For visual models, yes, because they have no other input channel for taste. The part that is fading is trickery like magic quality tokens. The part that keeps mattering is knowing what to specify, in what order, and when to stop writing and reach for a reference image instead.

Related terms