Photographers start from the strongest position in AI video and mostly use it wrong. You already have the thing everyone else is trying to generate: a real, well-lit, correctly composed frame. So the job is not text-to-video at all, it is image-to-video, and the discipline is asking for the smallest motion that adds something rather than the most impressive motion available.
Why image-to-video is the right route
A generated frame has to invent composition, light and subject. Your photograph already decided all three, correctly, on a real camera. Feeding it in removes every failure mode that comes from the model guessing, and what remains is motion, which is the one thing models do adequately.
The practical consequence: your output quality ceiling is set by your photograph, not by the model. That is a much better place to stand than most people generating video occupy.
What to ask for, in order of reliability
Most reliable: a slow push or pull, subtle parallax, drifting atmosphere (smoke, dust, steam), water surface movement, fabric moving slightly in air, light shifting.
Workable: a subject's small gesture, hair moving, a slow pan across the frame.
Unreliable: anything the photograph does not contain. A person walking out of frame, an object being picked up, a new element entering. The model has to invent it and invention is where stills-to-video breaks.
The rule that holds: ask for motion that is implied by the frame, not motion that changes what the frame is about.
Model picks and costs
Canon 8frame prices, a credit is $0.01 at pack rate:
| Model | Clip | Credits | Best for |
|---|---|---|---|
| Wan 3.0 (480p) | 5s | 24 | Volume drafting from a contact sheet |
| Veo 3.1 Lite | 8s silent | 33 | Cheap tests of a motion idea |
| Kling v3 Standard | 5s silent | 57 | The workhorse for animating a still |
| Hailuo H3 | 5s at 2K | 88 | Higher resolution output |
| Seedance 2.5 | 5s at 720p | 157 | Heavy reference conditioning |
Animating a set of twelve portfolio images for a website runs roughly 700 credits on the workhorse tier, about $7.
Where motion hurts the photograph
When the still was about stillness. A quiet portrait animated becomes a lesser thing. Not every image improves by moving.
When the motion reveals the edges of the frame. Parallax pushes past what the photograph contains and the model invents the periphery, usually badly.
When faces are involved. A still portrait animated is the most likely place to produce the uncanny result, because viewers scrutinise faces in motion far harder than in stills.
When you are showing craft. For a portfolio judged on photography, motion can read as covering for the image rather than extending it.
Practical uses that work
Website hero backgrounds, social versions of print work, client presentations where a still deck feels static, behind-the-scenes texture, and product photography where a slow rotation or light sweep adds information rather than decoration.
The rights note
Animating your own photograph is your work extended. Animating a photograph you licensed is a derivative use and may exceed the licence you bought. Check before you animate client or stock images. See what is a usage licence.
FAQ
What is the best AI tool to animate a photo? For most photographers, Kling v3 image-to-video at 57 credits silent. Cheap drafting on Wan 3.0 at 24.
Will it ruin my image quality? Output resolution is typically lower than your original file, so treat the video as a derivative for screens rather than a replacement.
Should I animate portraits? Cautiously. Faces in motion get scrutinised hardest, and a still portrait is often the stronger piece.
Ask for motion the frame already implies. The 8frame canvas is free and unlimited, and generation is paid from $19/month.