Models / Wan / Start & end frame

Wan start & end frame prompting guide

How to prompt Wan in start/end frame mode. Every point links its source; community tips are marked. Your agent gets the same guide through the Atlas MCP (get_guide).

Prompt structure

motion between→camera

• Entity + Scene + Motion; advanced adds aesthetic control (light, shot size, angle, lens, camera move) and stylization. source ↗

What the text should describe

• wan3 keyframe mode: promptImage [{position:first},{position:last}]; ratio must be auto_480p|auto_720p|auto_1080p because aspect follows the first frame; cannot mix with reference images. source ↗
• Like I2V, describe Motion + Camera; the frames define entity, scene and style. source ↗

Camera vocabulary

• Push-in for intimacy/tension, pull-out to reveal, tracking alongside, orbit (<45°) for importance, fixed for stillness. source ↗

Responds well to

• Voice = line + emotion + tone + speed + timbre + accent; SFX = source + action + ambient; orbit arcs under 45°. source ↗

Duration and limits

• Runway wan3: 2–30 s, 480p/720p/1080p, native audio; video refs ≤15 s total. wan2.7-videoedit input 2–10 s; wan2.2-s2v audio <20 s. source ↗

Multi-shot

• 'Shot 1 [0-3 s] ..., Shot 2 [3-6 s] ...'; on Wan 2.7 'Generate single shot.' forces one shot. source ↗

Avoid and failure modes

• Don't prompt: real people by name, rapid scene changes, exact text legibility, long choreographed action, lip sync to specific words. source ↗

Negative prompt

Supported via .

Next

Sources

• docs.dev.runwayml.com/api/
• www.alibabacloud.com/help/en/model-studio/text-to-video-prompt