Sam 3 Video Segment is Sam's video editing model on Versely. This page is its structured prompting reference: the 6 parameters its schema actually exposes, the video-edit technique that applies to it, copy-ready templates.
Everything here is grounded in the same sources Versely's agent reads — the model's input schema. Where a line is general craft advice rather than a documented fact about Sam 3 Video Segment, the page says so.
What Sam 3 Video Segment wants
The exact input surface, from the same schema the Versely agent fetches with get_model_input_schema before every generation.
| Parameter | What it does | Values |
|---|---|---|
video_urlreq | Input video | string |
prompt | Text prompt selecting the object to segment | string |
point_prompts | Optional click points (fal field is `point_prompts` not `point_coords`) | array |
box_prompts | Optional bbox prompts (fal field is `box_prompts` not `box`) | array |
apply_mask | Apply segmentation mask to output | booleandefault: true |
video_output_type | Output codec | X264 (.mp4) · VP9 (.webm)default: X264 (.mp4) |
- — Verified against fal.ai/models/fal-ai/sam-3/video — fields are `point_prompts`/`box_prompts`, not `point_coords`/`box`/`point_labels`
Technique that applies here
Video-to-video editing/restyle/segmentation: instruction-style prompts over an input video
- Every model in this family operates ON an existing video, but 'prompt' means four different things across the four pages: an edit instruction (Happy Horse 1.0 Video Edit, Wan 2.7 Video Edit), an object selector (SAM 3 Video Segment — its prompt's schema description is literally 'Text prompt selecting the object to segment'), an optional guiding note on a motion transfer (Wan 2.2 14B Move), or nothing at all (Lucy Restyle, whose only published creative control is a style enum: Realistic / Artistic / Anime at 720p). Identify which of the four you're writing before you write anything.
- SAM 3 Video Segment is selection, not generation: the text prompt names the object ('the red car'), and when words are ambiguous the schema gives you geometric fallbacks — point_prompts (click points) and box_prompts (bounding boxes); its notes flag that the fal field names are point_prompts/box_prompts, not point_coords/box. apply_mask controls whether the mask is burned into the output, and video_output_type picks the codec: X264 (.mp4) or VP9 (.webm).
- The Wan pages here still get the Wan MODEL_TIPS rule from the prompt enhancer ('Wan models work well with straightforward scene descriptions and style keywords'), and Wan 2.7 Video Edit additionally exposes prompt_extend — a documented LLM rewrite of your instruction — so on Wan, short and plain beats long and cinematic twice over. Happy Horse, Lucy, and SAM have no MODEL_TIPS family rule; their guidance is entirely their schemas.
Copy-ready templates
Replace the bracketed slots; each template says when it's the right shape.
Change [ELEMENT TO EDIT — e.g. 'the jacket the presenter wears'] to match @Image1. Keep [WHAT MUST SURVIVE — framing, identity, background] unchanged. audio_setting: [auto / origin]
Use when: Happy Horse 1.0 Video Edit — an edit instruction (max 2,500 chars) anchored to up to 5 reference images via the schema's own @Image1...@Image5 tokens, on a 3-60s mp4/mov source under 100MB.
prompt: [THE OBJECT, NAMED PLAINLY — e.g. 'the red car'] | point_prompts / box_prompts: [add click points or boxes if the words alone are ambiguous] | apply_mask: [true/false] | video_output_type: [X264 (.mp4) / VP9 (.webm)]
Use when: SAM 3 Video Segment — the prompt selects an object to segment; it is not creative direction, and geometric prompts are the disambiguator when two similar objects are in frame.
How the Versely agent does this automatically
You can use this page by hand, or let the agent apply the same knowledge. Four real mechanisms — no more, no less:
get_model_input_schema— before generating, the agent looks up Sam 3 Video Segment's exact input fields, required fields, allowed values, defaults, and min/max bounds. The parameter table above is that same surface.- The prompt enhancer's family rules — 12 per-family rewrite rules (this model's family isn't one of the 12, so only general enhancement applies) shape how a rough prompt gets rewritten.
- The per-provider speech guide — for TTS scripts, the agent follows a provider-specific tag scheme — not relevant to this model, but it's why voiceover scripts come out marked up correctly.
expand_movie_scene— in movie flows, brief scene ideas are rewritten into detailed cinematic descriptions before generation.
Mistakes that waste generations
- Writing a full scene description instead of an edit instruction. The source video already defines the scene — general instruction-editing technique is to state only the change ('replace the background with...', 'make the shirt match @Image1') and what must stay fixed.
- Giving SAM 3 creative direction. Its prompt selects an object to segment — 'a moody cinematic shot of the car at dusk' isn't a selector; 'the red car' is, and point_prompts/box_prompts exist for when words alone can't single the object out.
- Feeding Happy Horse a source outside its input contract — the schema takes 3-60 second mp4/mov files up to 100MB via video_url; longer or heavier clips fail before any prompt quality matters.
Frequently asked questions
Does Sam 3 Video Segment support negative prompts?+
No — Sam 3 Video Segment's published schema has no negative_prompt parameter. Exclusions have to be phrased positively inside the main prompt, or dropped.
How does the Versely agent know Sam 3 Video Segment's parameters?+
Before generating, the agent calls its get_model_input_schema tool, which looks up the exact input fields, required fields, allowed values, defaults, and min/max bounds for the model. Nothing on this page is guessed — it is the same schema surface those tools read.
Related prompting guides
Generate with Sam 3 Video Segment
Sam 3 Video Segment is live in Versely — paste a template above, or just describe what you want and let the agent map it onto the schema for you.