Most advice about AI photo prompts is about wording. That matters, but we kept running into a second problem: choosing the wrong aspect ratio or resolution and paying for a re-run. We build Omnely, a multi-model AI workspace that offers GPT Image 2. Disclosure: that's our product, and the numbers below are how GPT Image 2 is priced and constrained on Omnely, not in ChatGPT.
This post treats prompts like a shot list. First a checklist for each prompt, then the limits that decide your settings, then a worked budget.
A prompt is a brief, not a tag pile
The phrasing that works best is a photographer briefing an assistant: full sentences, one subject, one light source, and explicit "keep this, avoid that" rules. Here is the checklist we use. Our ChatGPT photo prompts library uses the same nine parts.
| Part | What to write |
|---|---|
| Shot type | "Overhead flat lay", "environmental portrait" |
| Subject | Concrete details, such as materials or age |
| Action / pose | One thing the subject is doing |
| Setting | Where, in one clause |
| Light | Direction, quality, colour temperature |
| Lens feel | "85mm look at f/2.8" |
| Colour and texture | White balance, skin or material detail |
| Keep / avoid | "No retouching, no text" |
| Use | What it's for, such as "4:5 feed post" |
The light line is the one to get right. Compare these two briefs for the same bakery photo. They're our own examples, so we aren't claiming any particular output.
Tag pile: bakery, croissant, 8k, masterpiece, ultra realistic
Brief: Overhead flat lay of three butter croissants on a cream linen cloth beside a small white cup of coffee. Soft daylight from a window at the top of the frame, faint natural shadows, neutral white balance, 50mm lens look. Visible flaky texture, no text, no watermark. Intended use: feed post.
The second gives the model a shot type, light, lens feel and exclusions. The first gives it nothing to disambiguate.
Settings are part of the plan
On Omnely, GPT Image 2 is priced by resolution only. It costs the same for text-to-image and for edits on an uploaded photo.
| Resolution | Credits |
|---|---|
| 1K | 10 |
| 2K | 15 |
| 4K | 20 |
Aspect ratio doesn't change the price, but it limits which resolutions you can pick:
- 1K only: auto, 4:5, 5:4, 3:1, 1:3 and 9:21.
- Up to 2K: 1:1.
- Up to 4K: 3:2, 2:3, 4:3, 3:4, 16:9, 9:16, 21:9, 2:1 and 1:2.
This catches people out. A 4:5 portrait looks like the natural feed format, but it can't go past 1K. If you need a larger portrait, you'd move to a ratio such as 3:4 or 2:3, which allow 4K.
A worked budget: four shots for a bakery launch
Suppose we need four images: a 4:5 headshot of the owner, a 1:1 feed post, a 9:16 story and a 16:9 web banner. Using the rules above, the highest allowed resolution for each is:
| Shot | Ratio | Final setting | Credits |
|---|---|---|---|
| Owner headshot | 4:5 | 1K (the cap) | 10 |
| Feed post | 1:1 | 2K (the cap) | 15 |
| Story | 9:16 | 4K | 20 |
| Web banner | 16:9 | 4K | 20 |
| One pass of all four | 65 |
The totals from here on are our own arithmetic, not figures from a page.
Strategy A: iterate at final settings. Say each shot takes four attempts. That's 4 × 65 = 260 credits.
Strategy B: draft at 1K, finish once. Do three 1K drafts per shot (10 credits each), then one final at the target setting.
- Headshot: it's already 1K, so 4 × 10 = 40.
- Feed post: 3 × 10 + 15 = 45.
- Story: 3 × 10 + 20 = 50.
- Banner: 3 × 10 + 20 = 50.
Total: 40 + 45 + 50 + 50 = 185 credits. That saves 75 credits, about 29%. The headshot saves nothing because there's no cheaper size to draft at.
One caveat of our own: a 1K draft is good for composition, light and wording, but it won't tell you everything about fine detail at 4K. Budget for the occasional second final.
For a single image the Omnely guide gives a simpler rule of thumb: ten 1K drafts plus one 4K final is 120 credits.
Three habits that kept the budget down
- Put the aspect ratio in settings, not the prompt. Only mention the intended use in the text so the composition leans the right way.
- Change one thing between drafts. If you change the light and the lens together, you won't know which one helped.
- Use reference images for real products and people. GPT Image 2 on Omnely accepts up to 16 references per run, and references don't add to the credit price.
What to check before you rely on any of this
Rates and limits can change. The model page on Omnely shows the current options for GPT Image 2. Omnely image models also need a paid plan after the free sign-up credits, and we haven't covered plan prices here, so treat the credit numbers above as per-image costs only.
Top comments (0)