From the character who warped within eight seconds to models that keep them the same from scene to scene
In November 2025, animating a character with Veo 3 meant cutting the clip before it stopped looking like itself. Today Google promises to keep it consistent from scene to scene.
On 8 November 2025, in EduMedia-IAG, several teachers who were producing videos shared the same frustration: in Veo 3 Fast, the character prepared as a reference image warped during the eight seconds of the clip, and in the next shot barely resembled itself. The fixes were makeshift: a storyboard of images drawn up beforehand, the exact description of the character repeated in every prompt, a saved frame to start the next scene, or a reminder that the dialogue had to fit into eight seconds. Some had been through fourteen attempts.
Veo 3.1, unveiled on 15 October 2025, already allowed scenes to be extended in Flow, but in the group the drift continued. On 19 May 2026, Google brought Gemini Omni to Flow, which, according to its announcement, preserves the character's identity and voice in every scene and lets users edit the video through conversation. The Gemini Omni page says it will replace Veo in the Gemini app.
On 5 November 2026 came Omni 1.2 Flash, which the paper reported on 6 November: clips of up to 15 seconds and, according to Google, fewer distortions between the first and the last frame. What used to take hours, keeping the character the same, is now, on paper, the model's job.
From the character who warped within eight seconds to models that keep them the same from scene to scene
In November 2025, animating a character with Veo 3 meant cutting the clip before it stopped looking like itself. Today Google promises to keep it consistent from scene to scene.
On 8 November 2025, in EduMedia-IAG, several teachers who were producing videos shared the same frustration: in Veo 3 Fast, the character prepared as a reference image warped during the eight seconds of the clip, and in the next shot barely resembled itself. The fixes were makeshift: a storyboard of images drawn up beforehand, the exact description of the character repeated in every prompt, a saved frame to start the next scene, or a reminder that the dialogue had to fit into eight seconds. Some had been through fourteen attempts.
Veo 3.1, unveiled on 15 October 2025, already allowed scenes to be extended in Flow, but in the group the drift continued. On 19 May 2026, Google brought Gemini Omni to Flow, which, according to its announcement, preserves the character's identity and voice in every scene and lets users edit the video through conversation. The Gemini Omni page says it will replace Veo in the Gemini app.
On 5 November 2026 came Omni 1.2 Flash, which the paper reported on 6 November: clips of up to 15 seconds and, according to Google, fewer distortions between the first and the last frame. What used to take hours, keeping the character the same, is now, on paper, the model's job.
