The best photo for AI video has a clear subject, useful visible detail and a composition that supports the intended movement. A head-and-shoulders portrait can suit a small expression change; walking needs a fuller view; a product reveal needs its important edges and label visible. Image size matters, but it does not replace those decisions.
- Match the source to the shot
- Start with detail you need to preserve
- Separate upload eligibility from image quality
- Frame the subject for the intended action
- Choose movement consistent with the image
- Make the important surfaces visible
- Simplify only what the shot does not need
- Decide whether to use, prepare or replace the photo
Choose the motion before choosing the photo. A beautiful close-up is not automatically a good starting point for a full-body action, and a dramatic action photograph may conflict with an instruction to keep everything still.
The Dreamina examples below show how portrait detail and product framing affect practical starting choices.
Match the source to the shot
These are useful starting conditions, not guarantees. The same image can be suitable for one action and unsuitable for another.
Start with detail you need to preserve
Inspect the source at a useful viewing size. For a face, check the eyes, mouth, hair and outline. For a product, check the shape and identifying features. A large image with blurred important detail can be less useful than a smaller, sharper original.
Runway's image-to-video prompting guide warns that visible defects can become more noticeable in motion. It also explains the role of an image as visual guidance and, in its documented generation mode, a first frame. Other reference modes may use the image differently; inspect the selected workflow.
Avoid assuming that animation will fix a flaw. If a hand is already distorted, a face is heavily blurred or a label is illegible, repair the source or choose a different one. Enlarging a file can meet a size requirement without recovering the original information.
A clear portrait gives you recognizable comparison points
This synthetic portrait was used in an existing Dreamina example:
The original is 2048 × 2048. The subject is a fictional AI-generated adult, with visible facial features, short curly hair and a blue shirt.
The resulting five-second clip requested a blink, a small head turn and a relaxed smile.
Half-second frame samples show the requested kinds of movement and broadly recognizable features. The native file is approximately 5.055 seconds at 960 × 960.
The useful lesson is how to inspect a source and result together. The portrait has enough visible detail to compare hair, face shape, eyes and clothing.
Separate upload eligibility from image quality
Check your chosen tool's supported file type, dimensions and size. Those requirements determine whether the service accepts the input. They do not determine whether it will produce the action you want.
An existing check used a 256 × 256 version of the same portrait:
This input was refused with “Image width must be at least 300 pixels.” No low-resolution animation was generated, so there is no matching output-quality comparison.
Use the input requirements shown by the selected mode, and choose the highest-quality original you have. Clear detail and comfortable framing give you a more useful starting point than a small, tightly cropped thumbnail.
Use the best available original that fits the service and the task. If a file is rejected, read the exact message and correct that issue. If it is accepted but the subject changes, investigate the image content, movement and mode rather than concluding that upload acceptance proves the source was ideal.
Frame the subject for the intended action
Leave space where movement needs to happen. A hand gesture needs room beside the body. Walking needs feet and a plausible place to step. A slow pullback asks the model to reveal surroundings, while an orbit asks it to infer another view.
A close portrait is appropriate when the action remains close. It is a poor constraint for showing the person stand up and walk across a room without inventing much of the body and setting.
Match the source orientation to the intended output when practical. A horizontal image can work in a vertical project, but the necessary crop may remove something important. Inspect the crop before generating or assembling, particularly around heads, product tops and bases.
A product example: prepare the vertical view before motion
The fictional NORI fragrance concept began with a wide product shot. Replacing the opening with a portrait-friendly composition gave the bottle more room and brought its color closer to the rest of the sequence. Prepare for the destination format before directing the camera move.
A later input provided a vertical view with the full bottle visible:
This prepared source gives the bottle's cap and base room within the vertical frame. It still includes a leaf shadow and surrounding glass; it is not a blank-background reference.
The replacement motion used that image as its first frame in the recorded workflow.
The revised five-second vertical shot keeps the cap and bottle base visible in the preview frames. Pale amber color and a roomy composition help it connect with the later product shots.
This shows a practical way to address framing: prepare an appropriate input instead of relying on a later crop. It is not a controlled experiment proving that vertical images are inherently better. The source, framing and revision history differ, and the improved fit concerns this particular vertical project.
Choose movement consistent with the image
Look for what the photo already implies. Motion blur, a mid-stride pose, airborne dust or flowing fabric can suggest action. That may help if you want to continue the action, but create a conflict if the prompt asks the subject to remain completely still.
Runway's prompting guide illustrates this with a vehicle: motion cues in the input can resist a stationary-car instruction. The practical response is to change the conflicting source cue or choose a different action, rather than merely adding stronger wording.
Its photo-animation guide also pairs movement with composition. A landscape's width can suit a pan; a close portrait often benefits from restrained movement. Treat these as starting ideas, not a fixed word count or a promise that a certain number of retries will succeed.
For your first attempt, use one main action and make a camera decision. A locked view with drifting clouds is easier to inspect than a scene where the camera orbits, the person walks, several objects move and the lighting changes at once.
Make the important surfaces visible
A single photograph does not show every side of an object or person. Turning the camera around a bottle reveals surfaces the model must infer. A frontal face does not verify the person's exact profile or the back of the head.
If the hidden view matters, provide an appropriate supported reference or choose a shot that does not need to expose it. A slow push toward an already visible surface can be a more suitable brief than a large orbit when the label and geometry must remain dependable.
The same applies to hand contact. A photograph with a hand already holding an object supplies a different starting relationship from asking a hand to enter later, find the handle and lift it. Neither guarantees correct motion, but the second asks the model to invent more of the interaction.
For product work, list the details that cannot change before generating: shape, cap, label, color, number of handles or a distinct fitting. The product-photo video comparison develops that decision further.
Simplify only what the shot does not need
A plain background can help isolate a subject, but it is not universally better. A landscape needs its environment. A room's architecture supplies depth. Shadows, reflections and the surrounding setting may be important to the intended shot.
Remove a distraction when it competes with the target or makes an interaction unclear. Keep useful context when the motion depends on it. For a group photo, see whether the relevant faces and hands remain distinguishable before asking everyone to interact.
Avoid aggressive restoration that changes the very details you want to preserve. With an old family photograph, a faithful source may matter more than a newly invented sharp face. Keep the original available when evaluating any repair.
Free accounts in the US receive 120 credits per day; daily allowances vary in other regions. You can start developing images and short videos with Seedance 2.0 for video and Seedream 5.0 Lite for images. In Dreamina, you can prepare the image before moving to video: correct the crop, refine a distracting detail or build the intended starting composition. The partial-image editing guide helps choose controls when only one area needs correction. Recheck important details after the image edit; a cleaner-looking source can still contain an unintended redesign.
Decide whether to use, prepare or replace the photo
A photo is ready when it contains the information the chosen shot needs and meets the actual upload requirements. That is more useful than aiming for one universal resolution or background style.
Once the source is suitable, follow the single-photo video tutorial. For a face-centered project, the portrait-animation comparison distinguishes expressions, identity and speech. Choose the source for the movement, then judge the moving result against the details you intended to retain.