Cinematic Camera Prompts for Seedance 2.5 Fan Edits

Key Takeaways

• Seedance 2.5 has no dedicated camera-movement parameter. Every shot type, lens choice, and camera move has to be written into the prompt itself.
• LTX 2.3 is different: its camera_lora field is an eight-value enum with a camera_lora_scale intensity dial. That structured control does not exist on Seedance 2.5.
• A camera prompt that reads consistently stacks four elements in order: shot type, lens and framing, movement, and movement speed, each one named rather than implied.

The AI Fan Edits of Costume Dramas: A Seedance 2.5 Guide covers the full production workflow for costume drama fan edits — reference sets, wardrobe consistency, clip length. This piece narrows to one part: directing the camera itself, shot by shot, using nothing but prompt text.

Why Camera Direction on Seedance 2.5 Lives Entirely in the Prompt

Seedance 2.5, available through Siray as bytedance/seedance-2.5-t2v, -i2v, and -ref2v, has no camera-movement field in its request schema. There's no enum to select, no slider to set. Whatever the camera does in the output comes from how precisely the prompt describes it.

That's a real contrast with LTX 2.3, which does expose camera direction as a structured parameter: camera_lora takes one of eight values — dolly_in, dolly_out, dolly_left, dolly_right, jib_up, jib_down, static, or none — paired with camera_lora_scale, a 0-to-1 dial for how strongly the move applies. That field is exclusive to LTX 2.3; Seedance 2.5 has no equivalent.

It's less a gap than a different division of labor: LTX 2.3 names the move for you, Seedance 2.5 asks the prompt to do it — so the four building blocks below carry more weight here:

1. Shot type — how much of the subject and setting is in frame.
2. Lens language — focal length, compression, and depth of field.
3. Camera movement — the move itself, plus its speed.
4. Framing and composition — where the subject sits, and what anchors it.

Shot Type: Naming What's in Frame

A close-up isolates an expression; a wide shot establishes the space around a figure. Naming the shot type explicitly beats describing the subject and hoping the model infers the right distance.

The reliable pattern names the shot type first: "medium shot," "close-up," "wide establishing shot," "over-the-shoulder shot." Testing across generic costume-drama scenes shows framing holds more consistently as its own phrase, not folded into a longer description.

Mini-example: "Medium shot of a court official standing in a palace hall, formal robes, hands folded, neutral expression."

Lens Language: Focal Length and Depth

A telephoto description compresses distance and blurs the background; a wide-angle description keeps more of the environment in focus, controlling how compressed or expansive the image feels.

Describe this like a cinematographer briefing an operator: "telephoto compression, shallow depth of field, background softly out of focus" for a portrait-style beat, or "wide-angle, deep focus, the full courtyard visible" for an establishing beat. Seedance 2.5 reads these as distinct instructions, not synonyms.

Mini-example: "Telephoto compression, shallow depth of field, a general in armor standing at attention, the training yard blurred behind him."

Camera Movement: Naming the Move and Its Speed

Camera movement is where prompts break down most often, and the failure is almost always the same: stacking movements that contradict each other, like a static shot and a dolly in one sentence. Seedance 2.5 needs one clear movement, not several competing instructions.

The vocabulary that reads reliably: dolly (toward or away from the subject), pan (a horizontal sweep), tilt (a vertical sweep from a fixed position), tracking shot (alongside a subject at constant distance), and crane (a vertical rise or fall). Naming just one, plus its speed, gives the model a tempo to follow instead of a vague "the camera moves."

Mini-example: "Slow dolly in on a court official seated at a low table, the candlelight dimming behind him as the frame tightens."

Framing and Composition: Keeping the Subject Anchored

Without an explicit anchor, a moving camera tends to drift toward whatever else is in frame — rarely the intended subject. Framing tells the model what to track through the shot.

Compositional language — "subject held in the left third of the frame," "headroom above the headpiece," "centered and steady through the pan" — gives the model something concrete to hold onto. This matters more in vertical 9:16 fan-edit formats, where push-ins, tilts, and orbits read more cleanly than a wide lateral tracking shot squeezed into a narrow frame.

Mini-example: "Subject held in the center third of a vertical 9:16 frame, headroom above her headpiece, steady through a slow tilt up."

A Full Combined Prompt

Stacking all four elements in order produces a prompt that holds together across a full clip: "Medium shot, telephoto compression with shallow depth of field, slow dolly in on a general in armor standing in a palace courtyard at dusk, subject held in the center third of the frame, headroom above the helmet, torchlight flickering across the armor as the frame tightens."

Shot type opens the sentence, lens language sets the depth, movement and speed follow, and framing closes with an explicit anchor. None of these four is optional for a shot that reads as directed rather than drifting.

Choosing between Seedance 2.5 and another model for a given segment is a separate decision, covered in Seedance 2.5 vs MiniMax H3 for Fan Edits: Which to Use. The structure above applies once Seedance 2.5 is the model in hand.

FAQ

Does Seedance 2.5 have a camera movement parameter?

No. bytedance/seedance-2.5-t2v, -i2v, and -ref2v have no camera-movement field — every shot type, lens choice, and move has to be written into the prompt as natural language.

Which endpoint does have a structured camera parameter?

LTX 2.3. Its camera_lora field takes one of eight enum values — dolly_in, dolly_out, dolly_left, dolly_right, jib_up, jib_down, static, none — with camera_lora_scale controlling intensity from 0 to 1. Exclusive to LTX 2.3.

What's the most reliable structure for a Seedance 2.5 camera prompt?

Shot type first, then lens and framing, then movement with its speed. Each as its own phrase reads more consistently than folding everything into one description.

What's the most common mistake in camera-movement prompts?

Stacking contradictory moves, like a static shot and a dolly at once. Naming exactly one movement plus its speed produces a cleaner result.

Create your free Siray account and start directing camera moves with Seedance 2.5 on Siray.