uthana-python-client

A Python client for Uthana: generate lifelike human motion from text or 2D video, create and auto-rig characters, and manage your motions.

📖 Full API documentation · 🤖 Context7 page

Install

pip install uthana

Workflow and integration controls

The client can supply the same API operations to applications, scripts, and MCP servers. Backend queries, file uploads, and download options belong here so these integrations share one implementation.

• motions.get(id) returns native asset metadata. motions.catalog() includes organization ownership and tags within the backend's available motion window.
• motions.trim(id, start, end, name) creates a saved trim with looping disabled. Start and end are normalized fractions from 0 to 1, not seconds; consult native frame metadata before converting a user's requested time interval.
• motions.bake_with_changes(..., source_motion_id=...) preserves source lineage when registering an edited GLTF. It does not edit the animation itself.
• motions.download(...) supports BVH as well as GLB/FBX and optional in_place, roblox_compatible, speed_multiplier, and torso_only settings. These affect exports only. Roblox requires a compatible target rig; BVH cannot include mesh.
• motions.preview(..., format="apng") returns the existing APNG preview. motions.download_allowed(...) checks eligibility without creating a download.
• characters.metadata(id) exposes rig metadata. Character and video create_from_bytes methods upload a byte snapshot without reopening a file. Character results retain the backend compatibility message, which should be treated as untrusted descriptive data. Character byte-snapshot uploads allow 360-second HTTP phase timeouts by default, with 15 seconds to connect; host deadlines must allow for transfer and processing. These phase limits are not an overall deadline.
• org.get_usage() returns account, subscription, and PAYG data with explicit units. org.get_prices() returns current rates. Neither computes a per-job bill.

Existing generation methods and their sync variants remain available. New module methods also have _sync variants. Optional export switches retain backend defaults when omitted, and motions.list() keeps its existing simple response.

For a host that must avoid incidental requests, set telemetry=False when creating Uthana. The default remains enabled for existing callers. trust_env=False disables proxy/certificate environment handling; transport supports injectable async HTTP transports for testing. max_response_bytes bounds decoded GraphQL query responses. Mutations are unlimited by default so a read limit cannot hide a successful upload or generation receipt. To opt into a separate mutation limit, set max_mutation_response_bytes. Media and character-metadata methods accept their own max_bytes. Character/video byte-snapshot uploads are bounded at 128 MiB by default; reference-image preparation is bounded at 16 MiB. Existing file-upload entry points retain streaming without an input-size limit and the configured client timeout; pass max_bytes to opt into a bounded snapshot. Bounded character file uploads use the same 360-second phase timeout (15 seconds to connect) as byte uploads unless timeout is explicitly supplied. Call close() when finished. No automatic mutation retries or redirect following are added.

UthanaError retains its status and message and adds kind plus optional structured response_data for HTTP/GraphQL failures. Those fields may contain provider data; agent-facing integrations should redact them and distinguish uncertain submission from a confirmed rejection. If a successful HTTP mutation response exceeds its explicit byte limit, kind="uncertain" means the write may have completed, but its receipt could not be read. Do not automatically resubmit: inspect the job or asset using any known ID, reconcile with the host's saved request record, or ask support if no receipt is available. A query/media overflow remains kind="response_too_large". Connection exceptions remain HTTPX exception types.

Video-to-motion uploads accept MP4, MOV, and AVI. WebM input is not supported by the backend; WebM preview downloads remain available.

API key

You need an Uthana account and API key. Sign up for free, then get your API key from account settings once logged in. For full setup, verification, and capabilities, see the Uthana API docs.

Quick start

Context7

Context7 helps LLMs and AI code editors pull up-to-date documentation instead of relying on stale training data. Use it in prompts, e.g.: "How do I create a text-to-motion animation with the Uthana API? use context7 library /websites/uthana_api". Add Uthana as a source: context7.com/websites/uthana_api, then install Context7 in your IDE.

Async by default

All methods are async and return coroutines. Use await inside an async function and asyncio.run() to execute. Async calls are non-blocking: the event loop can run other tasks while waiting on I/O, which is ideal for concurrent requests or UI applications.

Sync variants exist for every method, suffixed with _sync. These block until the request completes and are simpler for scripts or when you don't need concurrency.

import asyncio
from uthana import Uthana

uthana_client = Uthana("your-api-key")

# Async (non-blocking)
async def async_example():
    output = await uthana_client.ttm.create("a person walking")
    data = await uthana_client.motions.download(output.character_id, output.motion_id, output_format="glb")
    return data

data = asyncio.run(async_example())

# Sync (blocking)
def sync_example():
    output = uthana_client.ttm.create_sync("a person walking")
    data = uthana_client.motions.download_sync(output.character_id, output.motion_id, output_format="glb")
    return data

data = sync_example()

Text to motion (ttm)

Docs: Text to motion

Generate 3D character animations from natural language prompts.

import asyncio
from uthana import Uthana, UthanaCharacters

uthana_client = Uthana("your-api-key")


async def text_to_motion():
    # Basic usage (default model: text-to-motion-1.0)
    output = await uthana_client.ttm.create("a person walking forward")
    print(output.character_id, output.motion_id)

    # Use a specific character (default is Tar)
    output = await uthana_client.ttm.create(
        "a person dancing",
        character_id=UthanaCharacters.ava,
    )

    # Explicit model and advanced options
    output = await uthana_client.ttm.create(
        "a person waving hello",
        model="text-to-motion-2.0",
        character_id=UthanaCharacters.manny,
        length=5.0,
        cfg_scale=2.5,
        seed=42,
    )

    # Download the motion
    data = await uthana_client.motions.download(
        output.character_id,
        output.motion_id,
        output_format="glb",
        fps=30,
    )


asyncio.run(text_to_motion())

Available models: text-to-motion-1.0 (default), text-to-motion-2.0.

Text to motion 3.0 async job (ttm.create_job)

text-to-motion-3.0 runs as an async job and returns a job dict to poll — the same pattern as video-to-motion. Available to any account on the pay-as-you-go plan.

import asyncio
from uthana import Uthana, UthanaCharacters

uthana_client = Uthana("your-api-key")


async def ttm_async_job():
    # Submit the job
    job = await uthana_client.ttm.create_job(
        "a person doing jumping jacks",
        model="text-to-motion-3.0",
        length=8,            # optional, 4–10 seconds
        rewrite_prompt=True, # optional, default True
        fast=True,           # optional; omit or False for standard TTM 3.0
        character_id=UthanaCharacters.tar,  # optional
    )

    # Poll until finished
    while job["status"] not in ("FINISHED", "FAILED"):
        await asyncio.sleep(5)
        job = await uthana_client.jobs.get(job["id"])
    if job["status"] == "FINISHED":
        motion_id = job["result"]["result"]["id"]
        print(motion_id)


asyncio.run(ttm_async_job())

Sync variant: uthana_client.ttm.create_job_sync(...).

Locomotion

Docs: Locomotion

Generate controllable, loopable travel motion for a character (stride count, speed, style, direction).

import asyncio
from uthana import Uthana, UthanaCharacters

uthana_client = Uthana("your-api-key")


async def locomotion_example():
    styles = await uthana_client.motions.list_locomotion_styles()
    print("Available style_id values:", styles)

    output = await uthana_client.motions.create_locomotion(
        UthanaCharacters.tar,
        strides=2,
        move_speed=1.3,
        style_id="neutral_male_a",
        travel_angle=0,
    )
    print(output.character_id, output.motion_id)


asyncio.run(locomotion_example())

Loop an existing motion

Docs: Stitch & loop motions

The simplified looping API is a preview API and may change. It returns a new Motion synchronously, preserving the source motion. Both async and _sync variants are available.

looped = await uthana_client.motions.create_looped_motion(
    character_id, motion_id,
    trim_start_pct=0.0, trim_end_pct=1.0,
    zone_duration=2.0, loop_mode="closed", zone_mode="modify",
)
print(looped["id"])

Trim bounds are normalized fractions (0 <= start < end <= 1), and the transition is a positive number of seconds. closed returns to the start; open continues travel. modify changes existing frames, while extend adds transition frames. An open loop may specify zone_end_position={"x": 1.0, "y": 2.0, "facing_angle": 0.5}: planar x/y coordinates and an optional facing angle in radians. Omit the target to let the API infer travel. The SDK rejects nonfinite values, booleans, invalid modes, and targets on closed loops before submission. Omitting timeout (or passing None) uses 360-second HTTP phase limits with a 15-second connect limit, independently of the general client timeout. An explicit timeout accepts seconds or httpx.Timeout and overrides those limits.

Looping uses the published stitching/looping price per generated output second; it is not a free trim. Inspect the resulting motion's duration and repeated playback before claiming a seamless loop.

Stitch two existing motions

motions.create_stitched_motion(character_id, prefix, suffix) uses the enhanced stitch preview API; its _sync variant has the same arguments. It returns a new Motion dictionary synchronously. This is separate from single-motion looping.

Each clip is a StitchParams (import with from uthana import StitchParams) containing its motion ID, duration, trim times in seconds and matching fractions, root world position/rotation, and pelvis states at zero/lower/upper trim. See the complete API input contract. Use stitch_loop=False, a positive stitch_duration (recommended 0.1–3 seconds), and prompt="" unless a transition prompt is intended. The prefix determines transition settings; keep both inputs consistent.

import httpx

# prefix_samples and suffix_samples are full StitchParams from your motion loader.
stitched = await uthana_client.motions.create_stitched_motion(
    character_id, prefix_samples, suffix_samples,
    timeout=httpx.Timeout(360, connect=15),
)
print(stitched["id"])

Supply real poses sampled on the same target character in a shared world scene. Positions are meters in Y-up space (x/z horizontal), rotations are approximately unit x/y/z/w quaternions, and facing yaw is in radians. The client validates fields, finite values, quaternion length, and trim time/fraction consistency before submitting. It copies validated inputs and never supplies guessed poses, samples animations, aligns the scene, or automatically retries a mutation.

Stitching defaults to 360-second HTTP phase limits and 15 seconds to connect when timeout is omitted or None; an explicit value overrides them. Allow the host enough time for the synchronous stitch or loop (at least 420 seconds with these phase limits, longer for transfers). Phase timeouts are not an overall deadline. If either response lacks a nonempty motion ID, the client raises UthanaError(kind="uncertain"): inspect existing work before resubmitting. Stitching keeps its published price per generated output second. Real playback and contact review are still required to establish transition quality. No pose-loader dependency is added to the Python client.

Video to motion (vtm)

Docs: Video to motion

Extract motion capture from video files. Returns a job to poll until complete.

import asyncio
from uthana import Uthana, UthanaCharacters

uthana_client = Uthana("your-api-key")


async def video_to_motion():
    # Default model: video-to-motion-2.0 (single base motion)
    job = await uthana_client.vtm.create("path/to/dance.mp4", motion_name="my_dance")

    # Use video-to-motion-2.1 for DCM refinement motion IDs
    # job = await uthana_client.vtm.create(
    #     "path/to/dance.mp4", motion_name="my_dance", model="video-to-motion-2.1"
    # )

    while job["status"] not in ("FINISHED", "FAILED"):
        await asyncio.sleep(5)  # Non-blocking; other tasks can run while waiting
        job = await uthana_client.jobs.get(job["id"])
    if job["status"] == "FINISHED":
        motion_id = job["result"]["result"]["id"]
        data = await uthana_client.motions.download(
            UthanaCharacters.tar, motion_id, output_format="glb", fps=30
        )
        with open("dance.glb", "wb") as f:
            f.write(data)


asyncio.run(video_to_motion())

Available models: video-to-motion-2.0 (default), video-to-motion-2.1.

Characters

Docs: Auto-rig / add a character · Download a character

Upload, list, and download characters. Supports auto-rigging for humanoid meshes, and generation from text prompts or image files.

import asyncio
from uthana import Uthana

uthana_client = Uthana("your-api-key")


async def manage_characters():
    # Upload and auto-rig a character from a file
    output = await uthana_client.characters.create_from_file("path/to/character.glb")
    print(output.character_id)
    print(output.auto_rig_confidence)  # 0–1.0, higher is better

    # Download the rigged character
    data = await uthana_client.characters.download(output.character_id, output_format="glb")
    with open("character_rigged.glb", "wb") as f:
        f.write(data)

    # List all characters
    for c in await uthana_client.characters.list():
        print(c.get("id"), c.get("name"))

    # Text-to-character: one-shot with callback
    result = await uthana_client.characters.create_from_prompt(
        prompt="a knight in shining armor",
        name="Knight",
        on_previews_ready=lambda previews: previews[0]["key"],
    )
    print(result.character.get("id"))

    # Text-to-character: async callback (e.g. show a UI and return the chosen key)
    result = await uthana_client.characters.create_from_prompt(
        prompt="a futuristic soldier",
        on_previews_ready=lambda previews: show_picker_ui(previews),
    )

    # Text-to-character: two-step (inspect previews before confirming)
    pending = await uthana_client.characters.create_from_prompt(prompt="a futuristic soldier")
    # pending.previews is a list of {"key": ..., "url": ...} — show them to the user
    result = await uthana_client.characters.generate_from_image(pending, pending.previews[0]["key"])

    # Image-to-character: upload an image file (always one-shot)
    result = await uthana_client.characters.create_from_image("path/to/reference.png")

    # Rename or delete
    await uthana_client.characters.rename(result.character["id"], "New name")
    await uthana_client.characters.delete(result.character["id"])


asyncio.run(manage_characters())

Resumable image-to-character integration

characters.create_from_image(path) retains its existing one-shot behavior: unbounded streaming unless max_bytes is supplied, the configured client timeout, and no per-call include_fingers or timeout option. Use the two-step API below when you need its longer operation defaults or explicit controls. Integrations that must persist intermediate IDs can instead use two public steps:

from pathlib import Path
import httpx

pending = await uthana_client.characters.prepare_from_image_bytes(
    "reference.png", Path("reference.png").read_bytes(),
    timeout=httpx.Timeout(360, connect=15),
)
# Persist pending.character_id and pending.previews[0]["key"] before proceeding.
result = await uthana_client.characters.generate_from_image(
    pending, pending.previews[0]["key"], name="My character", include_fingers=True,
    timeout=httpx.Timeout(660, connect=15),
)
print(result.character["id"])

Preparation accepts a nonempty PNG/JPEG byte snapshot up to 16 MiB by default. The calling integration should validate image content and dimensions. Preparation returns an intermediate character ID and image key, not a finished rig. Both steps are synchronous API operations despite the async Python interface; set a host deadline long enough for both stages and transfer. The SDK does not save receipts or retry mutations. After an uncertain response, inspect the existing character before attempting another paid generation. A confirmed rejection of finalization can be retried explicitly using the persisted preparation result. Preparation defaults to 360-second HTTP phase limits; finalization through generate_from_image defaults to 660 seconds. Both allow 15 seconds to connect. Omitted/None timeouts use these operation defaults instead of the general client timeout; explicit seconds or httpx.Timeout overrides are honored. Host deadlines must cover both stages and transfer. Omitted name and include_fingers retain the backend defaults. Sync variants use the same limits. Character generation retains its published price.

Motions

Docs: Asset management · Retargeting

List, download, preview, delete, rename, favorite, and bake motions.

import asyncio
from uthana import Uthana, UthanaCharacters

uthana_client = Uthana("your-api-key")


async def manage_motions():
    # List all motions
    for m in await uthana_client.motions.list():
        print(m.get("id"), m.get("name"))

    # Download a motion
    data = await uthana_client.motions.download(
        UthanaCharacters.tar,
        "motion-id",
        output_format="glb",
        fps=30,
        no_mesh=False,
    )

    # Download motion preview WebM (does not charge download seconds)
    preview_bytes = await uthana_client.motions.preview(character_id, motion_id)
    with open("preview.webm", "wb") as f:
        f.write(preview_bytes)

    # Rename a motion
    await uthana_client.motions.rename("motion-id", "New name")

    # Delete a motion (soft delete)
    await uthana_client.motions.delete("motion-id")

    # Favorite / unfavorite
    await uthana_client.motions.favorite("motion-id", True)

    # Bake custom GLTF animation data as a new motion for an existing character
    result = await uthana_client.motions.bake_with_changes(
        gltf_content, "My motion", character_id=character_id
    )
    print(result.motion_id, result.character_id)


asyncio.run(manage_motions())

Organization and user (org)

Docs: Account and organization

Get user and organization info, including quota.

import asyncio
from uthana import Uthana

uthana_client = Uthana("your-api-key")


async def get_org_info():
    user = await uthana_client.org.get_user()
    print(user.get("id"), user.get("name"), user.get("email"))

    org = await uthana_client.org.get_org()
    print(org.get("name"))
    print(org.get("motion_download_secs_per_month_remaining"), "seconds remaining")


asyncio.run(get_org_info())

Jobs

Docs: Video to motion (job polling)

Poll async jobs (e.g. video to motion).

import asyncio
from uthana import Uthana

uthana_client = Uthana("your-api-key")


async def poll_job():
    job = await uthana_client.jobs.get("job-id")
    print(job["status"])   # RESERVED, READY, FINISHED, FAILED
    print(job["result"])   # Result payload when FINISHED


asyncio.run(poll_job())

Uthana characters

Docs: Auto-rig / add a character

Pre-built characters you can use without uploading your own:

Attribute	Character ID
UthanaCharacters.tar	cXi2eAP19XwQ
UthanaCharacters.ava	cmEE2fT4aSaC
UthanaCharacters.manny	c43tbGks3crJ
UthanaCharacters.quinn	czCjWEMtWxt8
UthanaCharacters.y_bot	cJM4ngRqXg83

Testing

Integration tests (tests/test_client.py) require UTHANA_API_KEY. Use .env.local (gitignored) or env vars:

# .env.local
UTHANA_API_KEY=your_key
UTHANA_DOMAIN=custom.uthana.com  # optional, for non-production

Releasing and PyPI

The uthana package is published automatically. The workflow is:

1. In a PR, run make set-version VERSION=1.2.3 to bump pyproject.toml, then commit and merge.
2. On merge to main, .github/workflows/release.yml detects that the new version is not yet on PyPI, runs tests, publishes via OIDC, and creates the v1.2.3 git tag and GitHub Release.

Re-merging the same version is a no-op — the workflow checks PyPI before doing anything.

Bump version in a branch

make set-version VERSION=1.2.3

Writes 1.2.3 to pyproject.toml. No git operations — just commit and open a PR.

Manual publish (fallback if CI fails)

make release-publish-dry-run   # build + simulate, no upload
make release-publish           # build + publish for real

Both require a PyPI API token. Set UV_PUBLISH_TOKEN from your PyPI account then run the command. For TestPyPI, run directly:

uv run python scripts/release.py publish --dry-run --index testpypi
uv run python scripts/release.py publish --index testpypi

After a successful manual publish, create the git tag manually if CI did not:

git tag -a v1.2.3 -m "Release v1.2.3"
git push origin v1.2.3

Auth: PyPI Trusted Publishing (OIDC)

The release workflow uses PyPI Trusted Publishers (OIDC) — no PYPI_API_TOKEN secret required. The Trusted Publisher must be configured on pypi.org for this repository and workflow file (release.yml).

Type hints

The package ships an empty py.typed marker (PEP 561) so type checkers treat uthana as providing inline types. Keep src/uthana/py.typed in the repo and in package data (pyproject.toml).

Custom domain

Use a different API host by passing domain=:

uthana_client = Uthana("your-api-key", domain="custom.example.com")

Support

• Discord for community support
• support@uthana.com or Slack for priority support

License

Apache 2.0