GPT Image 2.6 prompt guide: 12 techniques for better images
Learn how to write prompts for GPT Image 2.6. Twelve practical techniques with before/after examples for subjects, lighting, text, styles and aspect ratios.
GPT Image 2.6 rewards clear writing. You don't need secret keywords or walls of comma-separated tags. You need a sentence or two that says what is in the frame, how it is lit, and what it should feel like. These twelve techniques will get you there.
1. Lead with the subject
Put the most important noun first. "A ceramic teapot…" beats "An image showing a scene where there is a ceramic teapot…". The model weights early words more heavily.
2. Describe the setting in one clause
Where is the subject? "…on a wooden windowsill in a sunlit kitchen." One clause is enough; the model will fill in believable detail.
3. Name the light
Lighting is the fastest way to change the mood. Try golden hour, overcast soft light, neon backlight, single window light, studio softbox. Say where it comes from if it matters: "lit from the left".
4. Put text in quotes
GPT Image 2.6 is excellent at text, but only if you are explicit. Write a sign that reads "Open Late" rather than a sign about being open. Keep it to a few words per text element.
5. Pick a style, not ten
One clear style anchor works better than a pile of adjectives. On gptimages26.app the Style chips (Photoreal, Cinematic, Anime, Watercolor, 3D, Flat, Pixel) append a tuned suffix for you, so you can leave style words out of the prompt entirely.
6. Use camera language for realism
For photographic results, borrow photographer vocabulary: 85mm portrait, shallow depth of field, wide-angle 24mm, macro, long exposure, 35mm film grain. These words carry strong, consistent meaning for the model.
7. State counts and positions
"Three ducks in a row, the smallest on the right" is honored far more reliably in 2.6 than in older models. Use it.
8. Match the canvas to the use
Wide (1536×1024) for YouTube thumbnails, hero banners and slides. Tall (1024×1536) for Stories, Pinterest and posters. Square for Instagram and product tiles. The model composes differently for each, so choose before you generate.
9. Iterate with Draft, finish with HD
Draft quality costs 4 credits and is perfect for testing a composition. When the layout is right, rerun the same prompt at HD (25 credits) for the final file.
10. Say what to avoid in positive terms
Instead of "no people", write "an empty street". Positive descriptions are more reliable than negations.
11. Give brands a fictional name
Real logos are both unreliable and a rights problem. Invent one: "a coffee roaster called Northwind". You get clean, usable branding you actually own.
12. Keep a prompt library
When something works, save it. Small variations of a proven prompt (swap the subject, keep the lighting and lens) produce consistent series for campaigns and brand assets.
Three prompts to try right now
- Minimal product shot of a matte black water bottle on a stone pedestal, studio softbox light, square.
- Retro concert poster, bold headline "SUMMER NIGHTS", sun gradient, 70s typography, tall.
- Misty alpine lake at sunrise, pine forest, long exposure, ultra wide, cinematic.
Paste any of them into the generator and adjust from there.