Artlist AI Video Generator: Models, Frames and Settings

Artlist is best known for licensed music and stock footage, and its AI Tools add image, video, voiceover and music generation to the same subscription. For video, Artlist does not rely on a single model of its own: it offers a menu of third-party video models side by side, with one prompt box whose options change with the model you pick. That is useful, because you can choose the right engine per shot, and confusing at first, because settings, durations and supported inputs differ by model. This guide walks through the video generator as Artlist's help centre describes it, and how to get dependable clips from it.

Generating a video, step by step

1. In the side navigation, click Video, and check that Video is selected in the dropdown at the bottom of the prompt field.
2. If your plan includes Unlimited models, decide whether to toggle Unlimited on; Unlimited models carry an infinity icon.
3. Select the model. Each model supports different features and may limit characters or image uploads.
4. Add inputs (optional, model dependent): a reference image, or a start frame and an end frame. Images must have an aspect ratio between 0.4 and 2.5, and you can drag them into the prompt field on desktop.
5. Write the prompt, or use Auto-prompt to expand it; the back button returns to your previous version.
6. Set the options the model offers, then hover over Generate to see the credit cost before you click.

Finished videos are saved as MP4 in the folder where you created them.

Settings at a glance

Credits are charged per second of video and vary widely by model, resolution and whether audio is on; Artlist publishes the per-second rates in its help centre. A 4K clip with audio on a premium model can cost many times a 720p silent clip on a fast one, so match the model to the stage of the project.

Frames, ingredients and references

Inputs are where Artlist's model menu matters most.

• Start and end frames give smoother transitions and let you decide where a shot begins and ends. If the model you chose does not support end frames, Artlist prompts you to switch to one that does.
• Ingredients (Veo 3.1) combine a character, a specific object and a background as references. They only support 8-second clips.
• Video and audio inputs are accepted by some models: Kling 3.0 Motion Control, Kling O3 Video Edit and the Seedance models. Seedance 2.0 takes up to 3 clips totalling 15 seconds; Seedance 2.5 up to 10 clips totalling 30 seconds. Overlong clips can be trimmed in place with the Trim video link.
• References let you save a Character, Location or Prop once and reuse it; characters can be a single photo or a multi-angle Character Sheet. They currently work with models such as Veo 3.1, Gemini Omni Flash, Seedance 2.0 and Nano Banana for images.

Our guide to the Agent, Flows and Style Kit covers saving references and brand looks across a project.

Prompting with @ tags

On models that support tagging (Kling O3, Kling 3.0, Seedance 2.0 and 2.5), you refer to your inputs by tag, and the tag says what each input is for. Artlist's examples:

• First frame: "Set @Image1 as the first frame".
• Character consistency: "Reference the appearance of @Image2".
• Motion transfer: "Imitate the action of @Video1".
• Audio sync: "Sync movement to the rhythm of @Audio1".
• Camera control: "Reference the camera movement from @Video1".

Artlist's advice is that the more specific the tags ("@Image1 as character, @Video1 for motion"), the more consistent the result. Untagged inputs leave the model guessing what to take from each.

Writing prompts that hold up

Artlist's own prompt formula runs subject, style, details, context and, for video with audio, sound. Its practical tips are worth following because they target the failures AI video still has:

1. Frame people from the waist or chest up. Mid-shots reduce leg and foot errors; ultra-wide full-body shots increase them.
2. Keep motion simple. A glance over the shoulder or a light wave works; backflips and complex dance moves glitch.
3. Keep clips short and stitch them later; longer clips drift in clothing and limbs.
4. One idea per prompt, written in the present tense with active verbs.
5. Avoid relying on generated text. Add on-screen text in your editor.

For longer shots, Veo 3.1 Extend adds 7 seconds to an input video of 2 to 30 seconds. Otherwise Artlist suggests generating several short clips and joining them in an editor.

Choosing a model

Artlist's Model Picker lets you search by name or filter by need. Its AI Agent, when asked, suggests Veo 3.1 for strong audio synchronisation, Seedance 2.5 for high-end cinematic realism and Kling O3 for precise control over complex scenes. Treat those as starting points, test on a short clip, and keep a note of the model with every result: the same prompt behaves differently on each engine.

Try it

Slow camera tracking backwards as a woman in a linen coat walks towards the camera along a quiet harbour at dawn, a light breeze moving her coat, waist-up framing, soft warm light. Sound: gulls in the distance, water against the quay, no music.

Run it at 5 or 6 seconds on a fast model first. When the motion is right, rerun it on a premium model with audio on, or add a start frame from an image you have already approved to lock the look.