Keep one character consistent across shots. Give one reference each for character, setting and style, and the model reads them separately: who acts, where, and in what texture.
The reference decides how it looks.The prompt decides what happens this second.



You can edit the reading. The more specific it is, the better it holds across shots.

Character A1, locked
Swap the reference and these four are parsed again. The prompt does not need to repeat the face.
Keep the reference, change only the prompt, and a whole short is covered.




One front facing reference with even light holds best.
Rainy night? Add a scene reference. Film grain? Add a style reference.
The character stays; only what happens in this scene changes.
It depends on the selected setup. The uploader shows as many slots as the current model accepts and hides the rest.
Keeping a character, mascot or product recognisable across shots. Instead of describing the subject in text, you point the model at the images it should follow.
If the provider rejects the job before accepting it, the credits are refunded automatically. Once the provider has accepted the job and started rendering, the run is charged.
Yes. Upload the same product images for each take and keep the prompt focused on the action. Product shape, colour and material come from the references, so the item stays recognisable between shots.
A clean image with one clear subject, an uncluttered background and even lighting. Front-facing or three-quarter views are more useful than dramatic angles when the subject has to stay consistent.
One clear image per subject is usually enough. Add a second angle only when the shape matters — a bottle, a shoe or a character seen from the side — and keep the rest of the set clean.
Yes, and you should. The references carry appearance while the prompt carries action, camera and setting. If a detail keeps drifting, add it to the reference set instead of repeating it in the prompt - text that fights the image usually loses.
Only as a style cue. State where the scene takes place in the prompt; otherwise the model tends to reuse the backdrop, lighting and colour of the reference image, which is why characters can end up in the same room in every shot.
Fix the reference set, the style words and the aspect ratio before the first shot, then change only the action per clip. Write the list of shots first; improvising the next scene is what breaks continuity in a series.
Use the sharpest version you have. A small or heavily compressed reference is the most common reason a face comes back soft, and a tight crop of the subject beats a wide photo where the subject fills a fraction of the frame.
No. Keep one set per subject. Mixing two faces in a single set asks the model to merge them, and the result resembles neither; run the characters in separate shots and cut them together in your editor. Separate sets also let you regenerate one performance without touching the other.
Try a generation, then decide whether to pay.