The image-to-video workflow: from a still to a controlled clip
Why starting from an image gives you more control than text alone, and a step-by-step workflow in FlexPic: build the still, write a motion-only prompt, draft cheaply, then render the final clip.
Text-to-video asks the model to invent everything at once: the subject, the framing, the light and the motion. Image-to-video splits that job in two. You decide the picture first, then the model only has to decide how it moves. That is why many creators build their clips this way. You get a composition you already approved and a clip that starts exactly there.
In FlexPic the tool for this is Frame to Video.
When to start from an image
Choose image-to-video when:
- The look matters more than surprise: a product, a character, a logo scene, a location.
- You need the same subject in several clips. One approved image can be the first frame of every clip.
- You have a photo, illustration or design that you own and want to bring to life.
- Text prompts keep producing a composition you do not like and you would rather fix it as a still.
Choose plain Text to Video when you want the model to surprise you or when the subject is simple.
Step 1: make or choose the still
You can upload your own image or create one with Create Image. Stills are far cheaper to iterate on than clips. A GPT Image 2.5 image at 1K and Low quality costs about 8 credits, while a 5 second video costs hundreds. Do your composition work here.
Good start frames share three traits:
- A clear subject with room around it. If the subject touches the edge of the frame, the camera has nowhere to move.
- The aspect ratio you will publish in. Make a vertical image for a vertical clip.
- Light you like. The model usually keeps the lighting of the first frame, so fix it in the still.
Avoid images with tiny text, many small faces or complicated hands. They tend to shift once things start moving. Large images are reduced in your browser before they are sent, so do not rely on fine pixel detail surviving.
Step 2: load the start frame
In Frame to Video, under Set start frame, choose Add a start frame. Pick a file from your computer or from History. Frame to Video takes a single start image.
Step 3: write a motion-only prompt
The picture already says what is there. Your prompt should say what happens. This is the most common mistake: people re-describe the image and add nothing about movement.
For an image of a woman in a yellow raincoat on a wet street:
She walks toward the camera, rain falls, puddles ripple, neon signs flicker softly, slow push-inFor a product photo of a perfume bottle:
The camera orbits slowly around the bottle, light glints along the glass edges, a soft mist drifts across the backgroundKeep it to one subject action and one camera move. If you need more, make a second clip. The video prompt guide covers the five-part structure in detail.
Step 4: choose model, duration and resolution
Use the Model row to pick a model. The list shows what is available now; models marked Soon cannot be selected. Then open Output to set duration and resolution.
For a first draft:
- Use a short duration such as 5s.
- Use the lowest resolution offered.
- Leave Audio on if you want the model's sound, off if you will add your own.
Some prices for a 5 second clip at 720p, to help you budget:
| Model | Credits |
|---|---|
| Wan 2.7 | 625 |
| Hailuo 3 | 625 |
| Gen-4 Turbo | 313 |
| Seedance 2.0 | 957 |
| Kling 3.0 | 813 |
| Veo 3.1 Fast | 938 |
Gen-4 Turbo is an image-to-video model, so it is only offered here, not in Text to Video. Prices are computed from the same table the server uses. The button always shows the exact price for your current settings.
Step 5: draft, judge, adjust
Press Generate and watch the result twice. Ask yourself:
- Did the subject keep its identity? If the face or product drifts, shorten the clip or reduce the amount of action.
- Is the motion what I asked for? If it is too wild, add words such as "slow", "subtle" or "gentle". If it is too static, name the movement more clearly.
- Is the camera doing one thing? If not, remove extra camera instructions.
Change one thing and run again. Drafts at low resolution are cheap enough to repeat several times.
Step 6: render the final clip
When the draft is right, rerun the same settings at your final resolution. Most models give a similar result at a higher resolution, but not an identical one, so compare before you discard the draft.
If you need a longer sequence, use the last frame of a good clip as the start frame of the next. The editing and extending guide explains how to chain clips.
Common problems
- The image barely moves. Your prompt probably describes the image rather than the motion. Add verbs and a camera move.
- The subject changes shape. Use fewer actions and a shorter duration, or choose a different model for the same still.
- Faces look different in later frames. Keep the face large and well lit in the still, and avoid fast head turns.
- The aspect ratio is wrong. Recrop the still and upload it again. It is faster than fixing it afterwards.
A complete example
- Create a 9:16 image of a trail-running shoe on wet rocks at sunrise.
- Load it as the start frame.
- Prompt: "Slow orbit around the shoe, water drips from the sole, sunlight flares across the rocks".
- Draft with Wan 2.7 at 5s and 720p (625 credits).
- If you like it, render the final with a higher-priced model and compare.
Total cost of the exploration, including a handful of still images, stays small, and you end with a clip whose first frame you approved.