Apatero, Veo and Sora: Character Video Workflow Updates

Correction, September 20, 2026: Earlier claims that Sora and Veo always change a character's face while Apatero guarantees identity were unsupported. This update separates availability, reference controls and a proposed evaluation workflow. It reports no comparative generation test.

Sora availability: OpenAI discontinued Sora's web and app experiences on April 26, 2026. Its notice schedules the API shutdown for September 24, 2026. Do not start a recurring production workflow assuming continued Sora access. Existing users should follow the official export instructions and check the notice for updates. OpenAI discontinuation notice

Compare An Actual Workflow

A saved-character workspace, a video model and a consumer application are different things. Saving reference images can make assets easier to reuse. It does not prove that a downstream model will preserve every facial detail during motion.

Apatero's saved characters organize named references. For a particular video job, check the selected model and the inputs its generation path accepts. Do not assume that selecting a saved character supplies every reference to every video model. Inspect the resulting clip.

Google documents up to three subject reference images for Veo 3.1 and 3.1 Fast in the Gemini API. Support varies by variant; do not apply that feature to every model carrying the Veo name. Reference support is a documented control, not a guarantee of perfect consistency. Google Veo documentation

This is an availability and workflow table, not a measured quality ranking.

Start From An Approved Still

Choose a source image that clearly shows the features you need to preserve. Check permission to use the face, clothing, product and other protected material. Inspect the still before animating it; motion is not a reliable repair for an incorrect source image.

Write a restrained motion brief. For example: “The fictional presenter turns slightly toward the camera while the camera remains fixed.” This is a proposed test prompt, not a completed generation. Supply the reference or image input supported by the selected model and record those settings.

Before generating a batch, define rejection criteria. A product-presenter clip might fail for changed facial proportions, a distorted logo, an extra finger, altered clothing or a discontinuous background. Match the criteria to the intended use: a distant background figure and a close-up presenter require different scrutiny.

Keep Failures In The Comparison

Create one record for every attempt, including rejected clips:

Watch the entire clip. Inspect transitions where a face turns, a hand crosses the body or the camera moves. Check product geometry and clothing as well as identity. If the output contains audio, check speech, voice and visible movement against the intended script.

Calculate cost per accepted clip from the actual attempts you paid for. A cheaper attempt can produce a more expensive result if most clips are rejected. The number needs your recorded charges and acceptance decisions; this article supplies neither a provider-price comparison nor measured win rates.

Decide From The Recorded Result

Choose the configuration that meets your defined quality threshold within your actual budget and available access. Keep the source and settings so another reviewer can understand the decision. Recheck when a provider changes its model, inputs or service availability.

There is no supported universal winner here. Apatero can be evaluated as an asset workspace, and Veo through the specific interface you can access. Sora's announced shutdown changes its suitability for new work. None of these facts removes the need to inspect generated output.