Seedance 1.5 Pro: how to write prompts the model actually understands

Where 1.5 Pro sits in the line

1.5 Pro is «1.0 on steroids»: the same fixed 5/10 second duration, the same missing sound, but noticeably improved motion physics, better prompt adherence, and basic support for reference images (1–2) to preserve style.

Key difference from 1.0 Pro: a simple storyboard («Start / Middle / End») arrived along with basic negative-prompt support for simple bans. The camera vocabulary is broader — combinations like orbit + zoom now work reliably (cap of two simultaneous moves).

What separates it from 2.0: no second-by-second storyboarding, no full @-syntax with multiple references, no sound control, resolution capped at 1080p (2.0 goes up to 2K).

• Duration 5 or 10 seconds (same as 1.0)
• Resolution up to 1080p, 24fps
• Input: text + 1–2 images (basic reference)
• Simple Start / Middle / End storyboard
• Basic negative prompts work, complex ones don't

Prompt structure

Base formula is the same as 1.0 Pro: `[Subject] + [Action] + [Scene] + [Camera] + [Style]`. But 1.5 Pro handles detail better — you can safely flesh out appearance, materials, and lighting in greater depth.

For Text-to-Video: «Subject + Motion + Scene + Camera + Style/Mood». For Image-to-Video: «Subject + Motion, Background + Motion, Camera + Motion + Style». Base parameters at the end: «Non-fixed camera, 1080p, 16:9, 10s».

For 10-second videos use the simple storyboard — this is a 1.5 strength: «Start: woman sits at a café table, reading a book. Medium shot. Middle: she looks up, sees someone, smiles. Camera slowly zooms in. End: close-up of her face, warm smile, golden hour light.»

Simple Start / Middle / End storyboard

1.5 Pro introduces a three-act structure without hard per-second timing. It gives baseline narrative control without the complexity of 2.0.

Each block describes: what is in frame, what the camera does, what the lighting is. The model handles transitions between blocks — no need to spell out «cut to» or «camera switching».

Example: «Start: detective sits at his desk, dim light, medium shot. Middle: phone rings, he picks it up, camera slowly pushes in. End: close-up of his eyes widening as he hears the news, warm desk lamp catches one side of his face.» This is more reliable than a single paragraph — the model sees a three-act shape and distributes motion across time meaningfully.

Basic negative prompts

1.5 Pro accepts simple bans — but not complex constructions. What works: «No text, no watermarks», «No cartoon style», «No blurry details». What doesn't: long negative lists, negative character descriptions, contradictory constraints.

Best approach: phrase as much as possible positively («photorealistic», «sharp focus», «cinematic») and reserve negatives for basic exclusions only. A handful of simple bans in one line: «No text. No watermarks. No motion blur.» — workable format.

Conflict between the negative and the main prompt is an anti-pattern. If the prompt says «detective with a mustache» and the negative says «no facial hair», the model breaks. Make the prompt self-consistent first, then add the negative.

Common mistakes

1. 1. Per-second timing in 2.0 style

   «0-3s: X, 4-7s: Y» does not work in 1.5 Pro — precise timestamp storyboarding only arrived in 2.0. Use a simple Start / Middle / End storyboard instead. If a user is used to per-second timing, convert it to the three-act shape.

2. 2. Full @-syntax from 2.0

   @image1, @image2, @video1, @audio1 do not work in 1.5 Pro. Basic reference on 1–2 images without a prefix is supported. A prompt with multiple @-tags leaks the tokens into the text as garbage and quality drops sharply.

3. 3. Complex negative lists

   «No text. No watermarks. No motion blur.» is the working format. Long lists of 10+ bans or complex constructions like «no facial hair on a man who has stubble» break generation. Keep the negative short and non-contradictory with the main prompt.

4. 4. Three or more simultaneous camera moves

   Orbit + zoom — works. Orbit + zoom + pan — the model chokes. The 1.5 Pro cap is two simultaneous camera operations. If you need a complex shot, split it across acts via Start / Middle / End («Start: orbit. Middle: zoom in. End: pan right.»).

5. 5. Asking for sound or dialogue

   Sound control arrived in 2.0 — 1.5 Pro doesn't have it. «Add dialogue "Hello"», «with epic music», «sound of footsteps» — the model simply ignores these. The clip will be silent; add audio separately in post.

Before / after examples

girl walking down a street and smiling

Start: a young woman in a beige trench coat walks down a rainy Parisian street, medium shot, soft overcast daylight, reflections on wet cobblestones. Middle: she notices something off camera and slows down, camera pushes in slightly, her expression shifts to gentle curiosity. End: close-up of her face breaking into a quiet smile, golden hour light just starting to cut through the clouds. Cinematic, 35mm lens, film grain, shallow depth of field. Non-fixed camera, 1080p, 16:9, 10s.

product shot of a watch on a table, view from all angles

Luxury silver chronograph watch on a dark walnut surface, soft three-point studio lighting catches the polished case. Camera slowly orbits around the watch over the first 5 seconds, then pushes in for a close-up of the dial and crown over the remaining 5. Subtle reflections on the sapphire crystal, sharp focus on the watch face, smooth depth of field falloff. Commercial product photography aesthetic. No text. No watermarks. Non-fixed camera, 1080p, 16:9, 10s.

animate this landscape photo with mountains

Aerial shot, the mountain landscape gradually reveals itself. Start: wide overview of the snow-capped peaks, clouds slowly drifting from left to right. Middle: camera descends and tilts forward, revealing a glacial lake in the valley below, the lake surface ripples gently. End: camera glides above the lake toward a small wooden cabin on the far shore, golden hour light catching the windows. Cinematic, drone aesthetic, 4K-quality, deep focus. Non-fixed camera, 1080p, 16:9, 10s.