Image-to-video is the most practical AI video workflow. Start with a still image, then add motion with a video model. This guide walks through the full process in OpenVideoMaker.
Step 1: Generate a reference frame
Use an image model (GPT Image 2, Seedream, Imagen, or others) to create your starting frame. Focus on subject, composition, and lighting — the video model will handle motion later.
Step 2: Choose a video model
Upload the reference image to a video model. Each model has different strengths:
- Seedance: Up to 9 reference images, 3 reference videos, 6 aspect ratios including 21:9. Best for multi-reference and cinematic motion.
- Veo: Generates audio alongside video. Best for cinematic shots with sound cues and camera language.
- Kling: First and last frame control, reference video support, 3 quality tiers. Best for controlled transitions and reference-guided output.
- Wan 2.7: Prompt Extend for automatic prompt expansion, first and last frames. Best when you want the model to fill in prompt details.
Step 3: Write a motion prompt
Focus on motion, camera movement, and atmosphere — not on re-describing the image. The reference image already defines the frame.
Example prompts
Product:
The camera slowly pushes in, soft light moves across the product surface, subtle reflections shift, premium commercial mood
Character:
Gentle head turn toward camera, natural blinking, soft ambient light, subtle hair movement
Landscape:
Slow camera pan to the right, clouds drift gently, sunlight shifts across the scene, peaceful atmosphere
Key takeaway
The key to image-to-video is separation of concerns: image models handle the frame, video models handle the motion.