People hunt for the best AI prompt for image as if one magic sentence will work on every still. The miss is usually the sentence: two places, two lighting setups, and the word cinematic stacked on top.
This is how to write AI prompt images in Movmix Image so the file is closer to the picture you already have. Video prompts are a different post.
One formula
Write in this order. Do not start with mood.
Who or what → material or jacket → where the light comes from → lens → what to keep out
| Prompt | Output |
|---|---|
| Man, about 40, short dark curly hair, navy denim jacket, daylight garage, tools on a pegboard behind him, window light from the left, 50mm photo, real skin | ![]() |
That is a usable base. “Best” here means fewer variables, not more adjectives.
AI image prompts examples
Same subject. Only the writing changes.
| Prompt | |
|---|---|
| Off (That request paints a poster, not your garage | Epic cinematic portrait of a cool guy in a messy garage, ultra detailed, 8k, masterpiece, dramatic mood |
| On (Two more AI image prompts examples you can paste) | Same man in a navy denim jacket, daylight garage, pegboard with tools, window light from the left, 50mm, no extra people |
| Still life | Unopened glass bottle of orange soda on a metal workbench, daylight from a dirty garage window on the left, condensation on the glass, 85mm photo, no text on the label |
| Half body (Add one new fact per run: object or pose, not both) | Same man, denim jacket, leaning on the workbench, looking off camera, window light from the left, 50mm, real skin, no smile for the camera |
Image to image AI prompts: shorter than text to image
If the file already has the man, do not write his biography again. Image to image AI prompts only name the change and the ban.
Prompt:
Same man as the reference.
Same navy denim jacket and curly hair.
Move him to the open garage doorway.
Keep the face. Do not change his age.
The usual failure is uploading the reference and then asking for handsome, a new haircut, and a leather jacket. That is a second person.
| Job | Write | Do not write |
|---|---|---|
| New place | new place + keep the face | the whole face again |
| One garment | one garment only | jacket + hair + age |
| Cleaner frame | no extra people | cinematic, 8k |
Funny AI image prompts: a test, not the default
Funny AI image prompts are useful when you want to see if the model will follow a short, specific gag. Put the joke on one object. Do not write “make it funny.”
| Prompt | Output |
|---|---|
| Same garage. The orange soda bottle wears a tiny paper party hat. Daylight from the left. No people. 85mm photo. | ![]() |
If the joke needs five props at once, the frame falls apart before it gets amusing. One odd object first. Add a second only after the set still reads.
Then go back to the formula. The gag is practice.
When the still is wrong, edit one line
- Wrong person: drop handsome / beautiful / cinematic. Put back jacket color and light direction.
- Wrong place: delete the second location. Do not put garage and downtown in one sentence.
- Face drift on image to image: delete the appearance paragraph. Keep same face as the reference.
- Extra hands or logos: put them in the negative list. Do not explain the ban in the positive line.
Change one line. Generate once. Do not fix light, lens, and the joke in the same pass.




