What’s new
- Three generation modes, one endpoint: text-only, first/last-frame keyframing, or multi-reference from images/audio/video — pick a mode instead of switching products.
- Wider range of reference media: up to 8 images, one 2-12 second video and 3 audio clips per request, usable as content, style, rhythm or camera-motion references.
- Keyframing is its own mode: set just a first frame, just a last frame, or both, to control exactly how the clip opens and closes.
- More aspect ratios: six ratios (21:9, 16:9, 4:3, 1:1, 3:4, 9:16), so landscape, vertical and square come out directly; resolution also comes in several tiers (see the capability table above for what Nomi offers).
What it’s good for
- Turning a single sentence into a short clip with subject and camera motion.
- Controlling exactly where a clip starts and ends with first/last-frame keyframes, letting the model fill in the middle.
- Extending an existing image, audio track or video's style, rhythm or camera move into new footage instead of generating from nothing.
Prompting tips
- Cover six things in order: Agnes AI's own recommended sequence is subject/setting, action/change, camera language, visual style, sound/rhythm, then what must stay consistent.
- Name reference media with placeholders in reference mode: use
<Picture N>,<Audio N>and<Video N>in the prompt to say exactly what each numbered reference is for. - Duration is a string, in seconds: 4-12 seconds (e.g. "8"), defaulting to 5 if you don't set it.
- Fields don't mix across modes: text mode can't carry keyframes or references, keyframe mode can't add reference images/audio/video — pick one mode, since the wrong field combination is rejected outright.
- One result per request: the count parameter is fixed at 1 — run it again for another version instead of asking for a batch.
Camera: handheld tracking shot, slow push-in. Subject/setting: a rider weaving through a rain-soaked city street at night, neon reflected in the wet road.
Action/change: starts in a wide shot, gradually tightens to a side close-up. Visual style: cyberpunk tones, high contrast.
Sound/rhythm: low engine hum and rain, a pulsing beat underneath. Keep consistent: color grade throughout; avoid excessive camera shake.
Known limits
- The fields for text, keyframe and reference modes are mutually exclusive — you can't mix them in one request (e.g. keyframe mode can't also carry reference media).
- Only one result per generation (
nis fixed at 1); there's no batch-count control. - Duration is capped to the 4-12 second range; the faster 2.5 Flash variant is 720P-only and doesn't accept video references, but otherwise matches 2.5.
Using it in Nomi
On the generation canvas, add a video card, pick “Agnes Video 2.5” in the model picker and choose a mode (Text to video, First and last frame, Multi-reference). Add your references, write the prompt and generate. In the storyboard you can pick it per shot. The first time, connect any one of the providers above under Settings → Models.
