HY-Motion 1.0 online means running a text-to-motion model through a hosted browser interface instead of your own machine: you type a description of a human action, the service returns a 3D human-motion draft, you preview it in 3D, and you download it as FBX or JSON. At hunyuanmotion.org you can do that without installing weights, provisioning a GPU, or wiring a node graph. The important caveat is that what you get is a draft: a strong starting point for an animator, not a finished, production-ready clip.
This guide explains what HY-Motion 1.0 is, what "online" actually changes, how to produce a reviewable draft in the browser, and when you still need Blender, Unity, or Unreal to finish the job.
What HY-Motion 1.0 Actually Is
HY-Motion 1.0 is a text-to-motion model. You give it a natural-language description of a human action — "a person walks forward, then stops and looks over their shoulder" — and it produces animated human motion: a skeleton whose joints rotate and translate over time. That motion can be exported as an animation file and applied to a character.
It is worth being precise about what it is not:
- It is not an image generator. It does not draw a person; it produces movement.
- It is not a video generator. The output is structured animation data, not pixels.
- It is not a rigging or modeling tool. You still need a humanoid character to carry the motion.
- It is not an auto-retargeter. Moving motion onto your specific skeleton is a separate step.
Because the output is structured motion data, the natural download formats are FBX (widely used by animation and game engines) and JSON (useful for pipelines, inspection, and custom tooling). That is exactly what the browser interface on this site offers.
Text-to-motion versus text-to-video
A frequent source of confusion is mixing up text-to-motion with text-to-video. They solve different problems:
- Text-to-video produces a rendered clip. You cannot easily change the camera, the character, or the timing, and you cannot apply it to your own rig.
- Text-to-motion produces the underlying movement. You can retarget it, re-light it, re-camera it, and blend it with other clips.
If your goal is to drive a 3D character in a game engine or a 3D animation, motion data is what you want. If your goal is a finished-looking video for a specific shot, a video model may be the right tool instead. Many pipelines use both.
What "Online" Actually Changes
"Online" here means the model runs behind a hosted interface that you reach through a browser. That single difference cascades into everything else.
When you run a motion model locally — for example inside ComfyUI — you own the whole stack. You download the weights, you install custom nodes, you manage GPU memory, you wait for the model to load, and you debug whatever breaks. In exchange you get deep control: custom sampling settings, batch runs, integration into a larger automatic pipeline, and no per-run cost beyond electricity.
When you use a hosted browser interface, you give up most of that low-level control and get a much shorter path from idea to draft. There is nothing to install, no drivers to match, no VRAM ceiling to hit. You describe the motion, generate, preview in the browser, and download. The tradeoff is that you operate inside the service's supported options rather than a fully open graph.
For most people evaluating an idea — a solo developer, an animator blocking out a scene, a technical artist testing whether text-to-motion fits a workflow — the hosted route is the fastest way to answer "is this motion usable?" That is the problem the online generator is built to solve.
The daily free credits model
On hunyuanmotion.org, signed-in accounts can claim free credits once a day, and no card is required. That is a deliberate design choice: you can test the workflow across several days, refine your prompts, and decide whether it fits before spending anything. When you need more volume, credits are sold as one-time packs rather than a subscription, so there is no recurring charge to cancel. You can see the current options on the pricing page.
How to Produce a Reviewable Draft in the Browser
A "reviewable draft" is motion that you or a teammate can actually evaluate — a clear action, played back at a sensible speed, that answers whether the movement reads correctly. Here is a repeatable process.
Step 1: Write one action per prompt
Describe a single, coherent human action in plain language. "A person jogs three steps and jumps" is one action. "A person jogs, then dances, then sits down while holding a phone" is three actions crammed into one prompt, and the model may blend or drop parts of it. Keep prompts under roughly sixty words and focus on movement, not appearance: describe what the body does, not what it wears. Appearance words are wasted on a motion model.
Step 2: Generate and watch the whole clip
Generate the draft, then watch it end to end at normal speed before you judge any single frame. Timing problems — a motion that feels floaty, a foot that slides, a pause that is too long — are easiest to spot in playback, not in a still.
Step 3: Inspect it in 3D
Use a 3D viewer to orbit the motion and check it from multiple angles. This is where you catch a limb passing through the torso, a twisted spine, or a root that drifts across the floor. The 3D viewer is a good place to inspect a downloaded file without opening a full DCC tool.
Step 4: Download FBX or JSON
Download FBX if you are heading into Blender, Unity, or Unreal. Download JSON if you want to inspect the raw joint data, build tooling, or feed a custom pipeline. If you need to convert formats for a downstream tool, a dedicated 3D converter can help.
Step 5: Log what worked
Keep a short note of the prompt and the result. Text-to-motion rewards iteration, and a small library of prompts that reliably produce usable blocking saves far more time than re-deriving them from scratch.
Local ComfyUI vs Hosted Browser Interface
Both approaches run the same class of model; they differ in who owns the setup burden and how much control you get. This table is a fair summary.
| Dimension | Run it yourself in ComfyUI | Hosted browser interface |
|---|---|---|
| Setup required | Install ComfyUI, custom nodes, and model weights | None beyond signing in |
| Hardware | A capable GPU with enough VRAM | Any browser, no local GPU |
| Time to first draft | Minutes to hours, depending on setup | Minutes |
| Control over sampling | Full, including advanced settings | Limited to the interface's options |
| Batch and automation | Easy to script into a larger graph | Limited to what the UI exposes |
| Cost model | Upfront hardware plus your time | Free daily credits, then one-time packs |
| Maintenance | You update nodes and weights yourself | Handled for you |
| Best for | Deep customization and pipelines | Fast evaluation and quick drafts |
Neither column is "better" in the abstract. If you already have a tuned ComfyUI setup and want to automate motion generation at scale, local wins. If you want to know whether a prompt produces usable motion before you invest in that setup, the hosted route is the rational first move.
When You Still Need Blender, Unity, or Unreal
Text-to-motion does not remove the need for a 3D pipeline; it feeds it. Expect to open a DCC tool or engine in these situations:
- Retargeting. The draft's skeleton rarely matches your character's skeleton exactly. You map bones, adjust rest poses, and fix scale or orientation differences.
- Foot sliding and ground contact. Generated motion can drift relative to the floor. Fixing contact and root motion is standard cleanup work.
- Blending and layering. A full performance is usually several generated clips blended together, with transitions and additive layers.
- Timing edits. You will often retime portions to hit a beat, and that is an animation-editor task.
- Final polish. Hand-authored adjustments to the hands, spine, and weight shifts are what make motion feel alive.
Treat the generated clip as blocking, the same way an animator would treat a rough pass. For a broader look at how this fits into production, the use cases section covers game animation, film, and prototyping scenarios.
How This Compares to Other Motion Sources
Text-to-motion is one option among several, and it is honest to name the alternatives. Motion capture gives high fidelity but needs equipment, a performer, and cleanup time. Hand animation gives total control but costs skilled hours. Buying stock motion gives you polished clips but limited specificity. Text-to-motion's advantage is specificity without equipment: you describe exactly the action you want and get a draft in minutes. Its limitation is fidelity — you trade polish for speed and iterate from there. The AI motion capture alternative page explores that tradeoff in more depth.
Why hunyuanmotion.org
hunyuanmotion.org is an independent interface for turning text into 3D human-motion drafts. It is not affiliated with Tencent, and it does not host or train the underlying HY-Motion model. What it provides is a focused browser workflow: prompt, preview in 3D, download FBX or JSON, with free daily credits for signed-in accounts and no subscription. You can read more on the about page, and browse related write-ups on the blog.
FAQ
Is HY-Motion 1.0 the same as Hunyuan Video?
No. They are different tools aimed at different outputs. Hunyuan Video generates video frames; HY-Motion 1.0 generates human motion data — joint animation that you can retarget onto a 3D character and export as FBX or JSON. Confusing the two leads people to expect rendered clips when what they actually receive is animation data.
Do I need a GPU to use HY-Motion 1.0 online?
No. Because the model runs behind a hosted interface, your browser is the only thing doing work on your side. That is the main practical advantage over running the model locally in ComfyUI, where a capable GPU and sufficient VRAM are prerequisites. You can generate and preview drafts on a laptop that would struggle to load the model locally.
Is the output ready to ship in a game?
Generally no, and it is better to plan for that. The generated motion is a draft: a strong blocking pass that communicates the action. Before shipping you should expect to retarget the skeleton to your character, fix foot contact and root motion, blend clips, and hand-polish details. Teams that treat it as a fast first pass get the most value from it.
How many credits does a generation cost, and is there a subscription?
Signed-in accounts can claim free credits once a day with no card required, which is enough to test prompts and evaluate the workflow before paying. Paid credits come as one-time packs, not a subscription, so there is no recurring charge. Current pack options and prices are listed on the pricing page.
Can I use the downloaded FBX directly in Unity or Unreal?
You can import it, but plan on a retargeting step. Every engine expects a specific skeleton and humanoid definition, and the draft's joint proportions and rest pose may differ from your character. Import the FBX, map the bones to your rig, and expect to correct foot sliding and root drift before the motion looks right.
In what formats can I download the motion?
You can download FBX or JSON. FBX is the practical choice for Blender, Unity, and Unreal pipelines because those tools import it natively. JSON is useful when you want to inspect or transform the raw joint data, build custom tooling, or feed the motion into a pipeline of your own.