What’s new
- One model auto-routes between three modes: send a first-frame image and you get image-to-video, send 1–9 reference images and you get character reference, send neither and it's plain text-to-video — no mode switch needed.
- No special syntax required in prompts: the API docs state prompts can't contain special tokens, unlike some models that need @1/@2-style numbering to point at reference images.
- Drops the standalone "edit" mode from 1.0: version 1.0 had a separate video-editing mode for modifying existing footage; 1.1 currently keeps just the three generation modes — text, image, and character reference.
What it’s good for
- Workflows where you'd rather drop in assets and let the model figure out the mode than pick one yourself.
- Character-reference work, with up to 9 reference images per task.
- Standard-resolution shorts at 720P/1080P, with duration freely chosen from 3–15 seconds.
Prompting tips
- Describe the scene in plain language: don't add @-numbering or other special markers to point at reference images — the model handles that automatically.
- Keep prompts to roughly 2,500 characters or under (the stated limit).
- Spell out camera movement and mood directly, e.g. "slow push-in, cinematic, warm tones" — this works better than a purely static scene description.
- Reference images have a quality bar: a short side of at least 720px is recommended, and each file should stay under 10MB.
- First-frame and reference images are mutually exclusive: the two fields can't both be filled — pick one mode and leave the other set of assets empty.
1080P, 16:9, 8 seconds.
The young man in the reference image wears a khaki trench coat, same face throughout.
Slow push-in, cinematic, warm tones, a city rooftop at dusk, wind catching his coat as he looks out toward the skyline.
Known limits
- The first-frame (image-to-video) and reference-image (character reference) fields are mutually exclusive — they can't be used together.
- The aspect-ratio parameter has no effect in image-to-video mode.
- Reference images are capped at 9 — trim your set if you have more.
Using it in Nomi
On the generation canvas, add a video card, pick “HappyHorse 1.1” in the model picker and choose a mode (Text to video, First frame to video, 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.
