I tested the same prompt at low res and high res. The composition was identical. The credit cost wasn't.

Look at the two images below.

Same prompt. Same model. Same result.

One cost a fraction of the other.

Here's the thing most people don't understand about AI image generation — the model builds your composition first. The resolution just determines how sharp the final print is.

The workflow:

1. Write your prompt

2. Generate at low res — fast, cheap, done in seconds

3. If the composition is wrong, fix the prompt and test again

4. Repeat until it's right

5. Then — and only then — run it at full resolution

Test cheap. Finalise once.

More workflow tips like this in the Prompt King book 👉

Hands. The tell everyone sees.

You can nail the composition and get the lighting right. But then the hands blow it.

A misshapen grip, six fingers on one knuckle, skin merging into skin like a fever dream. Here's what people usually write:

The lazy prompt:

"two farmers shaking hands outdoors"

Now it's not wrong, it's just not accurate.

The fix: treat the hands like the subject.

The prompt that fixes it:

"Photorealistic agricultural stock photo, medium shot, two adult men mid-30s to mid-40s, casual introductory handshake. Man one: medium build, long sleeve green work shirt, light blue jeans, elbow slightly bent. Man two: slightly taller, navy blue shirt, short black work pants. Both angled toward each other, right hands in a calm web-to-web handshake — thumbs correctly interlocked, palms flush, four fingers distinctly separated wrapping the partner hand, correct knuckle compression, neutral wrist. New plantation, young crop rows, black plastic mulch sheeting foreground and background, subtle sheen on mulch. Afternoon, slight warm sun from camera-left, slight warm amber light on shoulders and hands, long shadows across mulch, rim light on shirt edges, slight warm sky. Canon EOS R5 85mm f/2.8, hands sharp focus, soft bokeh background."

Hands always fail when you're not specific—but they work when you treat them as the subject.

More tactics like this in the Prompt King book 👉

STOP DESCRIBING THE VIBE

Here's what people who just describe the "vibes" in image prompts write:

What people write:

“Create a premium, cinematic, high-end water bottle photo. Make it feel dynamic, sleek, luxurious and refreshing.”

Sounds pretty good right?

Wrong.

There's no specifics, and no details.

So the output becomes generic.

What you should write:

“Textured navy powder-coat finish, insulated metal water bottle, black plastic loop lid, studio macro photography, bottle suspended diagonally in motion, tiny dynamic water bubbles surrounding the bottle."

Don’t tell AI how you want the image to feel.

Tell it exactly what the camera should see, and name the details.

Stop Prompting With Text Alone: Why Your AI Images Drift And How The Dual-Anchor Method Solves It

Same product. Three prompts. Three very different results.

Here's exactly what I ran 👇

1. Reference Image Only

Colour held. Everything else is wrong. The model just combined the inputs without any direction. Complete failure.

2. Text Only

Got the product, lighting, and water right. Got everything else wrong. Colour oversaturated, wrong background. Looks unrealistic.

3. Text + Reference Image (the dual-anchor)

Product locked. Lighting locked. Camera angle locked. It matched the prompt with the style from the reference images — the right surface, the right mood, the right detail. Every element landed where it was supposed to.

The pattern is always the same:

Reference only → no direction.

Text only → scene drifts.

Both → correct product and scene.

That's the dual-anchor method. Full tactics in the Prompt King book 👉