Prepare reference images for a video job

Select the mode and check the files

In the video tool, choose image-to-video when your request needs source images. The current reference workflow accepts 1–10 images in JPEG, PNG or WebP format. Text-to-video does not take those reference-image inputs. File checks, ownership checks and the tool’s current upload limits still apply.

Before uploading a reference, check its actual format, file size and pixel dimensions. Each Studio image must be no larger than 20 MiB (20 × 1,024 × 1,024 bytes), with neither edge above 8,192 pixels and no more than 40 million pixels in total. A file can pass the size limit and still fail the dimension check. Image-to-video uploads require sign-in, an active Starter, Creator Annual or Studio Annual plan, and the source-media confirmations.

Use clear source images whose important details are easy to see. If several references are necessary, choose images that support the same intended result. Additional images do not guarantee better output; unrelated references can make the desired framing or subject ambiguous. Only upload material you have permission to use, with adults and valid consent wherever people are depicted.

Describe a small, coherent change

Describe what should move and what should remain stable. Camera motion, subject motion and changes in lighting are different instructions. A neutral example is “A slow camera pan across the room; keep the furnishings and warm lighting consistent.” Avoid simultaneously requesting incompatible camera directions or major scene changes.

Reference images guide generation; they do not lock every pixel in place. Identity, geometry, hands, fine text and motion can change in an AI-generated result. Check the complete clip before deciding whether it meets your needs, rather than relying only on its opening frame or preview image.

Match the output to its intended use

The video workflow offers 16:9, 9:16, 1:1, 4:3 and 3:4 aspect ratios. Choose a wide frame for landscape presentation, a tall frame for portrait presentation, or a square frame when the destination calls for it. Framing can change the visible composition, so check that important details have room around them.

For both text-to-video and image-to-video generation, the request supports whole-second durations from 2–30 seconds and 480p, 720p or 1080p output. Account access is checked separately: Starter and Creator Annual permit up to 10 seconds at 480p or 720p; 1080p or a duration above 10 seconds requires Studio Annual. These generation limits do not describe the separate video-upscaling or video-face-swap tools. A paid plan and sufficient Neurons are required for video generation.

Review the available duration, resolution and audio controls in the current Studio. The live quote reflects the supported configuration; an option shown elsewhere is not a promise that it is available for every request. After adjusting a setting, wait for the refreshed quote before submitting. Choosing generated audio does not promise that the result will reproduce speech or sound from a reference image.

Resolve input problems before another submission

If a reference cannot be accepted, check its file format and the selected mode, then review the error shown by Studio. Repeatedly renaming an unsupported file does not convert it. Avoid re-uploading sensitive material simply to bypass a validation message.

If a submitted job is still queued, working or being recovered, follow its existing status in Recent creations. A browser refresh does not prove that generation stopped. When requesting help, provide the job reference if available, the tool, the approximate time and a description of the problem. Support does not need your password or a private access token.