Cowboy Shot — AI shot size prompt
Cowboy Shot is a framing you can specify instead of leaving it to chance. Here is the phrase to use, what it changes, and which models render it reliably.
The prompt phrase
Cowboy Shot
cowboy shot framed from mid-thigh upShot size controls how much context the viewer gets. Left unspecified, models tend to land on a generic medium shot every time.
A worked example
A chef plating a dish in a busy restaurant kitchen, cowboy shot framed from mid-thigh upYour subject comes first, then the technique, then anything else. Models weight the front of a prompt most heavily, so keep the subject there and let the shot size modify it.
Models that render it
| Model | Credits | What it is for |
|---|---|---|
| Flux Schnell | 1 | instant drafts, lowest cost |
| Flux Dev | 2 | good quality without the top price |
| Nano Banana | 2 | follows instructions most closely |
| Recraft v3 | 3 | logos, icons and clean vector art |
| Seedream 4.5 | 3 | photoreal, and gets text right |
| Flux 2 | 4 | newest model, most realistic results |
| Flux Pro | 4 | sharpest detail of the Flux family |
| Ideogram 3 | 4 | posters and anything with words in it |
In video
The same phrase works in video prompts. These models accept it alongside a camera move:
| Model | Credits | Native audio |
|---|---|---|
| Veo 3.1 Lite | 6 | Yes |
| Veo 3.1 Fast | 12 | Yes |
| Wan 2.6 | 14 | Yes |
| Veo 3.1 | 22 | Yes |
| Hailuo 02 | 8 | Yes |
Related shot sizes
- Extreme Close-Up shot size
- Close-Up shot size
- Medium Shot shot size
- Wide Shot shot size
- Extreme Wide shot size
- Two Shot shot size
- Insert / Detail shot size
Common questions
What exactly do I put in the prompt?
The phrase above, word for word: "cowboy shot framed from mid-thigh up". Put it after your subject, separated by a comma.
Which models handle this best?
The 8 models in the table all support it. Flux Schnell is the cheapest way to test it at 1 credits, then move up a tier once the shot works.
Can I combine it with other shot sizes?
Yes. Stacking two or three related instructions is normal; stacking six competing ones is where results start to fall apart.