Text to Motion: Generate FBX Motion from a Prompt

Text to motion on pixal3d.ai: describe duration, body part, and ending pose, then generate FBX motion plus JSON and review the clip in a free FBX viewer.

Text to motion is the lane on pixal3d.ai that produces movement instead of geometry. You write a description of an action, and the Hunyuan Motion lanes return animation data — an FBX file plus a motion JSON file — rather than a static mesh. Nothing is rendered into a video, and no character image is required: the prompt is the entire input.

That makes it a different tool from the image-to-3D lanes, and it is worth understanding where the output fits before you spend credits on it.

What the text-to-motion lane produces

Item	Detail
Input	A text prompt describing an action
Lanes	Hunyuan Motion fast and standard
Output	FBX motion file plus a motion JSON file
Cost	60 credits on the fast lane, 80 credits on the standard lane
Provider	Runs on the fal provider through this hosted workspace
What it is not	Not video generation, not image-to-3D, and not motion capture from footage

The FBX is skeletal animation data. The JSON carries the motion data in a structured form that a pipeline can read without parsing the FBX. Neither file is a rendered clip, so if your deliverable is a video, this lane is the wrong starting point; if your deliverable is movement that another tool will play, retarget, or evaluate, it is the right one.

How to write a motion prompt that works

Prompt quality decides most of the result, and the failure mode is not a bad model run — it is a prompt that leaves a decision to the model. The workspace names the four things to make explicit: duration, body part, action order, and ending pose.

Prompt element	Vague	Explicit
Action	"walking"	"walks forward at a steady pace"
Body part	"waves"	"raises the right arm and waves"
Order	"gets up and walks"	"stands up from a crouch, pauses, then walks forward"
Ending pose	(absent)	"finishes in a neutral standing pose"
Loop intent	(absent)	"returns to the starting pose so the clip can loop"

The workspace ships reference motion prompts along these lines, including an idle breathing loop that shifts weight and returns to neutral, and a walk that turns and waves with natural timing. Use them as a shape rather than as text to copy: the sequence and the ending pose are the parts that most often decide whether the clip is usable.

Step by step

Open the AI 3D Generator and switch to the Text to motion tab. Sign-in is required to run a job.
Choose the lane. Fast costs 60 credits and standard costs 80; the price is shown before you submit, along with your balance after the job.
Write the prompt. Name the action, the body part where it matters, the order of the beats, and the ending pose. Keep it to one continuous piece of movement; two unrelated actions in one prompt tend to produce a compromise between them.
Set the duration. Duration is part of the request, so decide whether you need a short loopable beat or a longer sequence before you run it.
Run the job and download both files. The workspace keeps the task in your history, so the FBX and the motion JSON can be re-downloaded without generating again.
Review the motion locally. The free FBX viewer opens FBX files in the browser for animation review before you import them into a DCC tool.
Hand the motion to your pipeline. Whether the clip is retargeted onto a rig, used as timing reference, or read programmatically from the JSON, that step happens in your own toolchain — the workspace generates the motion, it does not attach it to your mesh.

What this lane will not do

It will not animate your generated model. The image-to-3D output is static geometry with no rig, and the motion lane does not bind animation to it. Applying the FBX to a character is a retargeting step in your own animation tool.
It will not capture motion from a video. There is no video input on this lane; the only input is text.
It will not render or export a video. The output is data, not footage.
It will not produce a full performance with dialogue or facial work. Treat it as body movement and timing.
It will not guarantee an exact frame count. Duration and the action description set the shape of the clip, and fine trimming happens downstream.

Where text-to-motion is useful

Previsualisation: block out how an action reads before committing animation time.
Timing reference: check the beat structure of a sequence — stand, pause, walk — before an animator builds it.
Locomotion drafts: generate a starting cycle for a game character and refine the loop in the engine.
Pipeline testing: validate that your FBX import path, retarget rig, and JSON reader work end to end with real motion data.
Idea exploration: run several prompt variations cheaply, since this is the least expensive lane in the workspace, and keep the prompts that read well.

Common mistakes

Mistake	Result	Fix
Prompting a vibe, not an action	Ambiguous movement that satisfies nothing	State the action, body part, order, and ending pose
Packing two actions into one prompt	A blend of both, with neither finished	Run two prompts and choose the better clip
Expecting a video	The download is FBX plus JSON, not MP4	Use it as animation data or a timing reference
Expecting it to deform your mesh	Nothing is rigged by this lane	Retarget the motion in your animation tool
Not checking the loop	A clip that cannot repeat cleanly	Ask for a return to the starting pose in the prompt
Re-running instead of re-downloading	Extra credits for the same motion	Re-open the task in your history, as described on the workspace page

What it costs

Fast lane: 60 credits. Standard lane: 80 credits. The price appears in the workspace before submission, along with the balance that will remain afterwards.
Signup credits: new accounts receive 220 one-time credits valid for 30 days, which is enough for several motion runs or one default image-to-3D job — see what the free credits cover.
Comparison: the 3D AI guide sets out how the motion lane differs from the two image-to-3D lanes, and Pricing lists credit packs if motion becomes part of a regular workflow.

Describe the movement you need, then run it: open the AI 3D Generator, switch to Text to motion, and start with one clear action.

FAQ

What files does the text-to-motion lane return?

An FBX motion file plus a motion JSON file. The FBX is skeletal animation data for a pipeline or animation tool, and the JSON carries the same motion in a structured form. No video is rendered, and no static mesh is produced.

Can it animate the 3D model I generated from an image?

Not directly. The image-to-3D output is static geometry with no rig, and this lane does not bind animation to it. Applying the FBX to a character is a retargeting step in your own animation tool.

How should I write the prompt?

Name four things: the duration, the body part where it matters, the order of the actions, and the ending pose. Say whether the clip should return to the starting pose so it can loop. Two unrelated actions in one prompt tend to produce a compromise between them, so run them separately.

Can it capture motion from a video?

No. The only input on this lane is text. There is no video-to-motion or webcam capture mode in the hosted workspace.

What does it cost?

The fast lane costs 60 credits and the standard lane costs 80 credits, with the price and your remaining balance shown before you submit. It is the least expensive lane in the workspace, and new accounts get 220 one-time signup credits valid for 30 days.