Animating a still, step by step

Animate a still image without losing the subject

Image-to-video works when the still is right, the prompt describes only motion, and you know what to change when the first try drifts. This guide covers all three, including the failures worth expecting. Video generation in ZeroTwo needs a paid plan.

Need a starting image first? Generate a source still from a text description.

One still in, one clip out

Your still
Frame 2
Frame 3
Last frame
Input
1 still
Prompt
Motion only
Output
Short clip
Illustration of the idea, not a real generation.

Prepare the still before you write a word

Most failed clips are failed pictures. Five checks that cost nothing and save generations.

  • The still is the first frame

    Image-to-video models treat your upload as the opening frame and invent everything after it. Runway's Gen-4 guide puts it plainly: the image sets subject, composition, colour, lighting and style, so the text prompt should be almost entirely about motion. A weak still cannot be rescued by a longer prompt, so spend your effort on the picture first.

  • Crop to the delivery ratio

    Decide where the clip will be posted and crop the still to that ratio before you upload. Runway's guide notes that choosing an output ratio different from your image makes you crop it anyway.

  • Leave room to move

    A subject jammed against the frame edge has nowhere to go. Headroom and negative space give a camera move or a gesture space to happen.

  • Start sharp and large

    Use the largest, cleanest version you have as JPEG, PNG or WebP. Heavy compression and tiny subjects are where faces start to smear.

  • Keep text and logos out

    Models regenerate pixels instead of copying them, so lettering tends to warp. Add logos and captions after, in an editor.

Source for the first-frame and crop notes: Runway's Creating with Gen-4 Video help article, read 5 October 2026. It describes one vendor's model; other models differ in the details.

Write the motion prompt: four parts, nothing else

Do not re-describe the picture. The model can already see it. Spend the prompt on what changes and what must not.

  1. What the subject does

    One action, in the present tense: turns her head, steam rises, the dog wags its tail. Two simultaneous actions double the ways it can go wrong.

  2. What the camera does

    Name a single move such as a slow push-in, a pan left or a locked-off shot. If you do not name one, the model chooses.

  3. What the environment does

    Wind in grass, drifting clouds, flickering light. Ambient motion makes a still feel alive without touching the subject.

  4. What must not change

    State the constraint: the background stays still, the product keeps its shape. It will not guarantee a result, but it narrows the guess.

Templates to adapt (illustrative, fill the brackets)

Gentle

Portraits and product hero shots

[Subject] slowly [one small action]. Camera is locked off. Soft [light or air] movement in the background. Everything else stays still.

Camera move

Landscapes, interiors, packshots

Slow push-in toward [subject]. [Subject] stays still. Subtle [ambient motion] in the [foreground or sky].

Stylised

Illustration and concept art

[Subject] [one expressive action] in the style of the image. Keep the palette and line quality unchanged. Camera drifts slightly to the [direction].

When the clip goes wrong, and what to change

Typical failure modes of still-to-video generation, with the cause we would suspect first. These are working heuristics, not measured rates.

Common image-to-video problems, the likely cause of each, and what to change on the next attempt.
What you seeLikely causeChange this
The face or product shape drifts during the clipToo much motion requested, a long clip, or a small low-resolution subjectAsk for one smaller action, pick the shortest duration, and start from a closer, sharper crop.
The framing jumps or the camera wandersNo camera instruction, or a vague oneName exactly one move, or say the camera is locked off.
Text and logos warpThe model repaints the pixels instead of copying themRemove lettering from the still and add it back in editing.
Everything moves, including the backgroundThe prompt never said what stays stillAdd an explicit line that the background and the rest of the scene do not move.
The result looks cropped or stretchedAspect ratio of the still differs from the output ratioRe-crop the still to the target ratio yourself before uploading.
Hands and limbs bend in odd waysArticulated motion is hard to hold togetherSimplify the action, or frame the shot so the hands are out of view.
  1. Run the smallest prompt first

    One action, locked camera, shortest duration. If this drifts, the problem is the still, not the prompt.

  2. Change one variable per attempt

    Camera, then action, then model. Change two things at once and you will not know which one helped.

  3. Stop at three tries, then change the still

    Our rule of thumb, not a measurement: after three misses on the same image, a different crop or a cleaner source usually saves more time than a fourth prompt.

What to look up before you choose a video model

Limits differ by model and change often, and a wrong number here can wreck a project plan. Here is how one vendor documents its image-to-video model, as a template for the questions to ask of any other.

Runway Gen-4 and Gen-4 Turbo specifications as published in Runway's help centre.
SpecGen-4Gen-4 Turbo
InputImage required, text prompt optionalImage required, text prompt optional
Clip lengths5 s or 10 s5 s or 10 s
Credits per second125
Frame rate24 fps24 fps
Native outputs16:9 1280x720, 9:16 720x1280, 1:1 960x960, 4:3 1104x832, 3:4 832x1104, 21:9 1584x672Same six ratios and sizes
Prompt length1,000 characters1,000 characters
4KSeparate upscale stepSeparate upscale step

Source: Runway, Creating with Gen-4 Video, checked 5 October 2026. Runway labels this article as an older generation and points to Gen-4.5 for current capabilities, so confirm against their current pages. This is not a list of what ZeroTwo offers.

Where animating a still works, and where it does not

Good fits

  • Short loops: a product hero with a slow push-in, steam off a coffee cup, a portrait that breathes
  • Concept and storyboard previews, where the point is to test a mood before shooting anything
  • Bringing an illustration or generated image to life with ambient motion

Poor fits

  • Anything where a logo, label or line of text must stay pixel-exact
  • Long continuous scenes: you will be stitching short clips and checking every join
  • Precise lip sync or choreography; use a tool built for that job

Video generation in ZeroTwo: which plans include it

ZeroTwo publishes this page and sells these plans. In the plan comparison, video generation starts at Plus; Free has limited image generation and no video. Which video models you can pick changes over time, so check the model list and current allowances before you commit to a project.

  • Free

    $0/mo

    Video: Not included

    Limited image generations only

  • Plus

    $14.99/mo

    Video: Included

    Image and video generation

  • Pro

    $29.99/mo

    Video: Included

    $26.99/mo billed annually

  • Ultra

    $120/mo

    Video: Included

    Image, video and deep research included

Before and after the clip

Image-to-video questions

Can I animate any image?

Image-to-video models generally take a still such as a JPEG, PNG or WebP and use it as the first frame of the clip. Check the file types, size limits and aspect ratios of the specific model you choose. A picture with one clear subject, good light and some empty space around it tends to animate more predictably than a crowded scene or an image with lettering in it.

How long can an AI-generated clip be?

It depends on the model. Runway's Gen-4 documentation, for example, lists 5 or 10 second outputs, not 60 seconds. For anything longer, the usual approach is to generate several short clips and edit them together. Consistency between clips is not guaranteed, so check every join.

What resolution will the video be?

It is model-specific. Runway's Gen-4 documentation lists native outputs such as 1280x720 for 16:9 and treats upscaling to 4K as a separate step. Look up the output size and any upscale option for the model you pick before you plan final delivery sizes.

Is image-to-video free on ZeroTwo?

Not on the Free plan. In ZeroTwo's plan comparison, video generation is included from Plus ($14.99/month) upward, and the Free plan does not include it. Pro is $29.99/month, or $26.99/month billed annually. See the pricing page for current allowances.

How do I stop the subject from changing shape?

No setting guarantees it. What helps in practice: start from a sharp, well-cropped still, ask for one small action, name what must stay fixed, and choose the shortest duration. If it still drifts, change the source still before you change the model.

What should I do when the first result is wrong?

Change one thing and run it again: the camera instruction, then the action, then the model. Every retry is another generation, so keep notes on what you changed. After about three misses on the same image, try a different crop or a cleaner source.

Bring one still and try the smallest prompt first

Video generation is included from the Plus plan.