Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →The same handful of dials (aspect ratio, style strength, seed, reference images) mapped across every major image model.
.claude/skills/whatsuppiyush-ai-image-controls-compared/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -3% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 21% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 3% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 33% | 0% |
Every image model gives you the same small set of controls, they just spell them differently. Midjourney uses -- flags; GPT Image and Nano Banana want plain words; Flux and Stable Diffusion use fields. This is the translation table. Learn the six dials once and you can drive any model, and you will not be stranded when Midjourney changes or a new model arrives. Deliberately model-agnostic: the concepts outlive the syntax.
Works in: Midjourney, GPT Image, Flux, Nano Banana, Stable Diffusion.
When the user asks for an image prompt in this style, compose one using the formulas and vocabulary below. Fill each slot with a concrete choice, then return the finished prompt. Prefer naming a real camera/lens, a light, and a colour or film treatment, which is what makes the output read as a real photograph rather than an AI render.
The six dials
aspect ratio · style strength · seed · reference images · negative prompts · quality / upscaleIf you can set these six on a model, you can control it. Everything else is a variation on describing the image well, which the style guides on this site cover.
_"A 16:9 cinematic shot" in each model_
Midjourney: <prompt> --ar 16:9 --style raw · GPT Image / Nano Banana: <prompt>, 16:9 widescreen cinematic framing · Flux: set aspect_ratio 16:9 · Stable Diffusion: width 1344, height 768Aspect ratio (the shape of the frame)
--ar 16:9 (or 1:1, 9:16, 3:2…). A flag at the end of the prompt.Style strength (how hard it leans into your words)
--stylize / --s (0-1000). Low stays literal, high improvises. --style raw reduces the house look.Seed (repeat or vary a result)
--seed <number> reproduces a starting point; same seed + same prompt = similar result.Reference images (show it an example)
--cref for character and --sref for style reference.Negative prompts (say what to avoid)
--no <thing> removes elements (e.g. --no text).Quality & upscaling
--q affects detail; the Upscale buttons enlarge a chosen image.From God of Skills: a curated, hand-tested directory of AI skills, prompts, templates and image style guides. Source: https://godofskills.com/guides/ai-image-controls-compared?ref=claude-skill
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-15 | pass→pass | 5,582 | 1,918 | -66% | 1 | 1 | 0% | 1,076 | 1,504 | +40% | 0 | 0 | — |
case-22 | pass→pass | 14,927 | 11,050 | -26% | 1 | 1 | 0% | 2,678 | 3,282 | +23% | 0 | 0 | — |
case-01 | fail→pass | 13,764 | 7,478 | -46% | 1 | 1 | 0% | 2,330 | 2,538 | +9% | 0 | 0 | — |
case-02 | pass→pass | 8,171 | 4,480 | -45% | 1 | 1 | 0% | 1,659 | 2,062 | +24% | 0 | 0 | — |
case-03 | pass→pass | 13,534 | 6,738 | -50% | 1 | 1 | 0% | 2,240 | 2,371 | +6% | 0 | 0 | — |
case-04 | pass→pass | 12,477 | 3,739 | -70% | 1 | 1 | 0% | 2,169 | 1,786 | -18% | 0 | 0 | — |
case-05 | pass→pass | 9,811 | 6,570 | -33% | 1 | 1 | 0% | 2,022 | 2,406 | +19% | 0 | 0 | — |
case-06 | fail→fail | 10,406 | 5,716 | -45% | 1 | 1 | 0% | 1,766 | 2,129 | +21% | 0 | 0 | — |
case-07 | pass→pass | 14,656 | 6,271 | -57% | 1 | 1 | 0% | 2,656 | 2,300 | -13% | 0 | 0 | — |
case-08 | pass→pass | 9,298 | 3,094 | -67% | 1 | 1 | 0% | 1,588 | 1,744 | +10% | 0 | 0 | — |
case-09 | pass→pass | 8,818 | 3,252 | -63% | 1 | 1 | 0% | 1,559 | 1,760 | +13% | 0 | 0 | — |
case-10 | fail→pass | 8,799 | 2,393 | -73% | 1 | 1 | 0% | 1,635 | 1,581 | -3% | 0 | 0 | — |
case-11 | fail→pass | 9,016 | 4,215 | -53% | 1 | 1 | 0% | 1,562 | 1,894 | +21% | 0 | 0 | — |
case-12 | fail→pass | 10,909 | 5,301 | -51% | 1 | 1 | 0% | 1,989 | 2,057 | +3% | 0 | 0 | — |
case-13 | pass→pass | 15,165 | 11,870 | -22% | 1 | 1 | 0% | 2,687 | 3,004 | +12% | 0 | 0 | — |
case-14 | pass→pass | 10,366 | 3,527 | -66% | 1 | 1 | 0% | 1,518 | 1,696 | +12% | 0 | 0 | — |
case-16 | pass→pass | 9,049 | 2,934 | -68% | 1 | 1 | 0% | 1,669 | 1,736 | +4% | 0 | 0 | — |
case-17 | pass→pass | 16,580 | 8,458 | -49% | 1 | 1 | 0% | 2,696 | 2,469 | -8% | 0 | 0 | — |
case-18 | pass→pass | 5,682 | 1,912 | -66% | 1 | 1 | 0% | 929 | 1,515 | +63% | 0 | 0 | — |
case-19 | pass→pass | 9,383 | 5,267 | -44% | 1 | 1 | 0% | 1,597 | 2,054 | +29% | 0 | 0 | — |
case-20 | fail→pass | 9,069 | 5,617 | -38% | 1 | 1 | 0% | 1,713 | 2,276 | +33% | 0 | 0 | — |
case-21 | pass→pass | 16,271 | 8,615 | -47% | 1 | 1 | 0% | 2,639 | 2,737 | +4% | 0 | 0 | — |
case-23 | pass→pass | 16,330 | 14,758 | -10% | 1 | 1 | 0% | 2,850 | 3,700 | +30% | 0 | 0 | — |
case-24 | pass→pass | 13,548 | 9,185 | -32% | 1 | 1 | 0% | 2,716 | 3,069 | +13% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 24 cases were attempted. The headline lift of +21 percentage points is the difference between those two pass rates over the 24 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/9/2026 | +17% |
| gemini-3.6-flash | verified | 8/6/2026 | +26% |
Other measured skills in the registry, with their headline benchmark lift.