AI Prop & Item Generator

A prop is a reference image with a job

Objects that matter in a story — the letter, the blade, the phone with the message on it — need to look the same every time they appear, and generated video has no memory between shots. So each one gets a reference image, and each shot that uses it says what role that image is playing. Without the role, the model treats a picture of a dagger as a mood board rather than as “this exact dagger, in this hand, now”.

At a glance

Why the prop field is the only thing that counts

A dashboard panel once appeared in seven consecutive shots of an episode because the phrase sat in the location description and every shot inherited it. The fix was to stop treating mentions as references: images attach only from the shot's own prop list, and a prop that no line of the shot text actually mentions is dropped from that list before rendering. Set dressing describes the world; props are the objects the camera is asked to show.

Sheathed, drawn, broken

Objects with two physical configurations used to be drawn as one picture showing both — a blade out of its scabbard with the scabbard beside it. That is a good casting photo and a bad reference: the render kept snapping the drawn blade back into its sheath mid-shot. Now the base image is the object in its carried state, and drawn is a separate state image, chosen per segment like a costume change.

The same rule covers breakage. A cracked, snapped or burned version of an object is a different object structurally, so it gets its own image; an object that is merely dirty, wet or lit is the same object and is handled in text.

Describing an object so it can be drawn

The same rule that governs every visual prompt here applies: describe causes, not results. “Ornate”, “precious” and “masterfully crafted” give a model nothing to draw. Material, construction, wear, closure, proportion do.

• Instead of “an exquisite antique locket” — write a hinged oval case, tarnished silver, worn smooth at the thumb catch, hairline scratches across the front.
• Instead of “dramatic lighting on the blade” — write a single hard source from the upper left, so the forging pattern reads along the edge.
• Do not write what the light should land on. Written as a landing point, the render paints a bright patch exactly there.

Editing and replacing

Every prop can be renamed, re-described, regenerated or replaced with your own upload. Regenerating costs one image. Renaming is safe — shot text is rewritten to follow the new name, so references do not break — and uploading your own picture is the right move when the object is real and you have a photo of it.

What it will not do

• Render readable text on an object. Labels, letters and screens with legible writing are outside what current models do; a letter can be a letter, but not a letter you can read.
• Keep a prop in a shot that does not list it. By design.
• Turn a photo into a 3D asset. Uploads are reference images, not models.
• Track an object's exact position between shots. Continuity is carried by the written opening and closing state of each segment, not by simulation.

What it costs

Prop images are ordinary generated images: 7 credits each beyond the 15 free reference images in your first project, which cover characters, locations and props together. State images have their own separate free allowance in that first project. Attaching props to shots costs nothing, and so does renaming or re-describing. See location plates for the same approach applied to sets, and the pricing page for rates.

Frequently asked questions

Pricing·Novel to Video·Guides·Samples·Support·Privacy·Terms·Legal·Contact·© 2026 SceneMixer