Compare the workflows
| Decision | Text-led video | Image-led video |
|---|---|---|
| Starting material | A written description of the scene and action. | A prepared still plus a description of what should move. |
| Useful starting point | Exploring a fictional place, mood or visual direction. | Developing a portrait, character image or approved composition into motion. |
| What to decide first | Subject, setting, framing and the main action. | Whether the still is suitable, then subject motion and camera movement. |
| What to review | Scene interpretation as well as motion and continuity. | Reference fidelity as well as motion, edges and background changes. |
| Common limitation | A written identity or composition can be interpreted differently. | The still can still drift or deform during motion. |
1. Choose by the unresolved creative decision
Use a text-led task when you are still deciding what the scene looks like. Describe one subject, one setting and one readable action. This is useful for a fictional establishing shot where a specific existing face or object is not the main constraint.
Use an image-led task when you already have an approved starting composition. A portrait can establish framing and visible wardrobe before you ask for a small turn or glance. Inspect the still first; animation does not repair an unclear face or a cropped hand automatically.
2. Distinguish a starting image from multiple references
A single image used as a starting point and a collection of reference images are different inputs. Some models offer a dedicated image-to-video mode; others offer a broader reference workflow for images, video or audio. Do not assume that every model uses the first uploaded image as an exact first frame.
In the video studio, review the selected model and the reference controls it actually supports. A model may accept one kind of input and reject another. Keep reference roles clear: character appearance, scene composition and motion direction should not accidentally contradict each other.
3. For a character short, approve appearance before motion
Write a short character brief containing only the details you need to preserve. Prepare a readable image, then decide one action that fits the framing. A close portrait can support a glance or slight head movement; a wider action generally needs a composition with room around the body.
Describe subject movement separately from the camera. For the first pass, try a steady camera and a restrained action. Watch the full returned clip and pause on the face, hands and clothing. A recognizable first frame does not establish consistency through the entire video.
4. Compare results with the same review checklist
Judge whether the shot communicates the intended action, whether the subject remains coherent and whether the beginning and ending fit your sequence. Compare against the actual brief and any source reference. An appealing unrelated result does not answer the same creative task.
If you compare workflows yourself, record the model, input, settings and revision count for each attempt. This guide is a decision aid, not a measured quality or cost ranking. Review the current task quote in the tool; a still-first workflow can involve separate image and video generation steps.
Prompt examples to adapt
These prompts illustrate a creative brief. They are not claims about a measured generation result; adapt them to your reference and the model controls available in the app.
Text-led scene exploration
A quiet fictional workshop at dawn. An adult ceramic artist carefully places a finished cup beside a window. Medium-wide framing, soft side light, one slow continuous action and a locked-off camera. Keep the background simple and avoid added lettering.
A scene brief for exploration; no specific source identity is being asserted.
Motion from an approved portrait
Use the approved portrait reference in the input mode supported by the selected model. The adult character briefly looks toward the window, then settles into a relaxed expression. Small natural movement, steady framing and consistent wardrobe. One continuous shot.
Review how the selected model interprets the image; this is not a guarantee of identity preservation.
Keep reference roles clear
The character reference defines appearance and clothing. The scene reference defines the setting and lighting. Show one calm head turn in a medium shot with a steady camera. Do not introduce additional characters, outfit changes or cuts.
Use only if the chosen model supports these separate reference inputs.
When a result needs another pass
The upload does not behave like an exact first frame
Check whether the selected mode uses a starting frame or a broader reference input. Read the current controls before revising the prompt.
The face changes during the clip
Inspect the starting still, simplify motion and review the entire sequence. Select a different approved take if the identity changes.
The clip tries to tell too much at once
Split the sequence into individual shots. Give each task one action and decide how adjacent shots should connect.
Before the next creation step
- The intended shot is clear.
- The selected model supports the reference input.
- Subject motion and camera movement are distinct.
- The whole clip will be reviewed against the brief.
Prepare a starting image · Plan a sequence
Reading this guide does not start a generation. Sign in and review the selected model, settings and quote in the creation tool when you are ready.
Workflow guidance is compiled from the creation tools currently available in Omni AiLab. Controls and availability can change; the current interface is the reference for a specific task.