Image-to-Video AI: 8 Tips for Clips That Stay True to Your Photo
How to choose the source image, write motion prompts, use first and last frames, and stop faces and products from drifting when you animate a photo with AI.
Image-to-video is the most predictable way to get the clip you have in mind. The picture fixes how the subject looks, what the composition is and most of the lighting before the model starts, so it only has to work out how things move. That's a much easier job than inventing a whole scene from text.
It still goes wrong in familiar ways: faces that drift, products that warp, motion that's either frozen or chaotic. These tips are about avoiding that.
1. Start with the right image
The source image matters more than the prompt.
- Use a sharp, well-lit image. Blur, heavy compression and noise get amplified once things start moving.
- Leave room for the motion. If the subject is going to turn, walk or raise an arm, don't crop tightly around them. A little space around the subject gives the model somewhere to move into.
- Match the aspect ratio to where the clip will be posted. A vertical 9:16 image for Reels, TikTok and Shorts, 16:9 for YouTube and websites, 1:1 for some feeds. Reframing after generation always costs resolution.
- Keep it simple. One clear subject against a readable background animates more cleanly than a busy crowd scene.
If you don't have a good still yet, create one first. In Oraclip you can generate or restyle an image in the same chat and then animate it. The Product Photo Studio and Photo Restyle skills are useful starting points.
2. Don't describe what's already visible
A common mistake is to re-describe the whole image: "a red sports car on a coastal road at sunset with mountains in the background". The model can see all of that. Spend your words on what should change:
The car accelerates away from the camera along the coastal road. Slow, steady tracking shot from behind at car height. Dust lifts slightly from the road; the sun flares at the edge of frame. End with the car small in the distance.
3. Say what must not change
Tell the model what to preserve, especially for faces, logos and product shapes:
Keep her face, hairstyle and outfit exactly as in the image.
Preserve the bottle's shape and label; the label stays readable throughout.
This doesn't guarantee perfect consistency, but it noticeably reduces drift, particularly in longer clips.
4. One subject move, one camera move
The cleanest image-to-video clips usually combine:
- one subject action, such as a turn, a smile, a step forward or a product rotating, and
- one camera move, such as a slow push-in, a gentle orbit or a static shot.
Add restrained secondary motion if you want the scene to feel alive: hair moving in the wind, steam rising from a cup, leaves drifting. Keep it small. If everything in the frame moves, nothing reads as the main action.
5. Use a last frame when the ending matters
Some models in Oraclip can take two images, an opening frame and an ending frame, and animate between them. This is the most reliable way to control where a shot lands: a before and after, a product that opens, a character who turns from profile to facing the camera. Make sure both images share the same subject, lighting and framing, or the model will invent a transition you didn't ask for.
If you only have one image, describe the ending in words instead: "ends on a centered close-up with the logo facing the camera".
6. Reference image versus first frame
These are two different jobs, and mixing them up causes a lot of confusing results:
- First frame: the clip literally starts on your image and moves from there. Use it when the composition is exactly what you want.
- Reference: the image guides the character, product or style, but the clip can start from a different framing or setting. Use it when you want the same person or product in a new scene.
In Oraclip you don't have to pick a mode. Attach the image and say what you want, for example "animate this photo" or "put this product on a kitchen counter in a new shot", and the agent chooses whether to treat it as a first frame or a reference. If your intent is genuinely unclear, it may ask a quick question first.
7. Keep clips short at first
Start with a short duration while you're testing the motion. Short clips render faster, use fewer credits and drift less. Once the motion is right, lengthen the duration or raise the resolution for the final version.
8. Iterate one variable at a time
If a clip is close but not right, change one thing:
- Motion too strong? Add "subtle" or "slow", or remove the secondary motion.
- Face drifting? Shorten the clip and add a preservation sentence.
- Camera feels random? Replace the move with "static tripod shot" and see whether the subject motion is right on its own.
Try it
The quickest way to practise is the Animate a Photo skill: upload a still, pick the skill, and describe one action. For more ideas, browse community examples on the Prompts page, or read how to write AI video prompts for a full prompt structure. Failed renders are refunded, and the credit plans are on the pricing page.