One to All Animation: Identity-Consistent Pose-to-Video Generation on playground and API | RunComfy

community/one-to-all-animation/1.3b

One to All Animation 1.3B creates identity-consistent animations from one image and a driving video with alignment-free, pose-driven motion transfer for realistic character animation and cinematic content production.

The resolution of the video to generate.
The number of inference steps to use for the video generation.
The image guidance scale to use for the video generation.
The pose guidance scale to use for the video generation.
Idle
The rate is $0.03 per second for 480p, $0.045 per second for 580p, and $0.06 per second for 720p.

Introduction to One to All Animation

One to All Animation 1.3B turns a single reference image plus a driving video into identity-consistent animations at $0.03 ~ $0.06 per second, delivering alignment-free pose-driven video-to-video motion transfer. Trading manual rigging and pose alignment for identity-robust motion retargeting and long-sequence coherence, it eliminates re-shoots and pixel-level masking for animation studios, game teams, and brand content operations using One to All Animation on RunComfy. For developers, One to All Animation on RunComfy can be used both in the browser and via an HTTP API, so you don’t need to host or scale the model yourself.
Ideal for: Identity-Consistent Character Animation | Alignment-Free Pose Retargeting | Rapid Previsualization for Social Video and Game Cinematics

Model Overview


  • Provider: One to All Animation (Shijun Shi, Jing Xu, Zhihang Li, et al.)
  • Task: video-to-video
  • Max Resolution/Duration: Up to 720p; short-form clips (seconds-long)
  • Summary: One to All Animation 1.3B generates identity-consistent character animations from a single reference image and a driving video with alignment-free, pose-driven motion transfer. It excels at preserving appearance under skeletal/layout mismatch and sustaining temporal coherence over longer clips, making it a strong fit for technical artists who need robust motion transfer and consistent character identity.

Key Capabilities


Alignment-free pose-driven motion transfer

  • Transfers motion from a driving video to any reference image without requiring spatial or skeletal alignment.
  • Produces stable character animation even when the reference and driving poses differ significantly in layout or scale.

Identity-robust control under difficult inputs

  • Maintains subject identity with hybrid reference fusion and identity-robust pose control.
  • Handles occlusions and missing parts through occluded-input/outpainting preprocessing to keep facial and costume details consistent.

Long-clip temporal consistency and cross-scale support

  • Uses a token-replacement strategy to preserve coherence across frames in longer sequences.
  • Supports cross-resolution setups between the reference image and motion inputs while maintaining motion fidelity.

Input Parameters


Core Inputs


ParameterTypeDefault/RangeDescription
promptstring""Required. High-level description of the desired style/look or scene guidance for the animation.
negative_promptstring""Optional. Text terms to avoid (artifacts, unwanted styles or objects).
image_urlstring (image_uri)""URL to the reference image (jpg/jpeg/png/webp/gif/avif). The identity is extracted from this image.
video_urlstring (video_uri)""URL to the driving video (mp4/mov/webm/m4v/gif). Motion is derived from this clip.

Generation Settings


ParameterTypeDefault/RangeDescription
resolutionstring (choice)480p, 580p, 720p (default 480p)Output video resolution. Higher values increase detail and cost.
num_inference_stepsintegerdefault 30Denoising steps. Higher values can improve quality at the cost of speed.
image_guidance_scalefloatdefault 2Strength of identity/style conditioning from the reference image. Increase to better preserve appearance.
pose_guidance_scalefloatdefault 1.5Strength of pose adherence from the driving video. Increase to tighten motion following.

How One to All Animation compares to other models


  • Vs Seedance 1.0 Pro: Compared to Seedance, One to All Animation delivers stronger identity preservation from a single image when the driving video’s skeleton/layout is misaligned. Seedance excels at prompt-to-scene generation and multi-shot storytelling; One to All Animation is superior for personalized motion transfer from images.
  • Vs Wan 2.1/2.5: Compared to Wan’s smaller text-first variants, One to All Animation is optimized for image-plus-video inputs and alignment-free pose transfer. It typically provides more reliable identity consistency and motion mapping when a driving video is available.
  • Vs Kling Video 2.6: Compared to Kling’s photoreal scene generation, One to All Animation focuses on accurate motion retargeting and identity maintenance under pose/layout mismatches. Kling may win on global cinematic realism; One to All Animation leads for character-centric, pose-driven personalization.
  • Vs One to All Animation 14B: The 14B sibling yields higher fidelity and finer textures, while One to All Animation 1.3B offers faster, lighter inference with strong alignment-free motion transfer and temporal stability.
  • Key Improvements: Better handling of misaligned reference/pose layouts, stronger facial and costume consistency, and improved long-clip coherence in the 1.3B lineage.
  • Ideal Use Case: Choose One to All Animation 1.3B when you need efficient, identity-consistent, alignment-free video-to-video character animation from a single image and a driving video.

API Integration


Developers can seamlessly integrate One to All Animation into pipelines using the RunComfy API with standard HTTP requests. The model accepts a reference image, a driving video, and optional prompt controls, making it straightforward to automate video-to-video animation in production workflows.


Note: API Endpoint for One to All Animation


Official resources and licensing


  • Hugging Face: https://huggingface.co/MochunniaN1/One-to-All-1.3b_1
  • Official Website: https://ssj9596.github.io/one-to-all-animation-project/
  • Paper (arXiv): https://arxiv.org/abs/2511.22940

Related Playgrounds

Frequently Asked Questions

What resolutions are supported by One-to-All Animation for video-to-video generation?

One-to-All Animation currently supports outputs up to 720p resolution for video-to-video tasks, with optional 580p and 480p modes for faster generation or lower compute environments. Higher output resolutions may be available in the 14B variant but are typically capped for the 1.3B model to ensure temporal coherence and consistent identity preservation.

Are there any technical limitations in prompt size or reference inputs for One-to-All Animation?

Yes. In One-to-All Animation video-to-video generation, prompts are typically limited to around 512 tokens, and only one reference image plus one driving video (pose sequence) can be uploaded at a time. Multiple ControlNet or IP-Adapter style inputs are not natively supported in the 1.3B variant for performance and memory reasons.

How can I move a test setup from the RunComfy Playground to production API usage for One-to-All Animation?

After evaluating results in the RunComfy Playground interface, developers can transition One-to-All Animation video-to-video pipelines to production via the RunComfy API. The API mirrors the playground parameters, including prompt, reference, and driving video fields. You’ll need to generate an API key with available USD balance, then call the REST endpoint documented on the RunComfy Developer Portal for automation or integration within larger workflows.

What makes One-to-All Animation unique for video-to-video character generation compared to competitors?

One-to-All Animation stands out for its alignment-free motion transfer, allowing arbitrary layouts between the reference and driving sequences. For video-to-video animation, it excels at identity retention and stable long-sequence generation, performing better than many text-driven competitors like Seedance when the source and target poses differ significantly.

How does One-to-All Animation maintain consistent facial identity across frames?

The One-to-All Animation model uses hybrid reference fusion attention and an appearance-robust pose decoder to separate identity from motion dynamically. In video-to-video mode, this ensures the character’s key facial and costume details remain coherent, even when the driving video introduces new or complex poses.

Is One-to-All Animation suitable for stylized or realistic animation outputs?

One-to-All Animation supports both, depending on the style of the reference image. For instance, a stylized 2D character reference in a video-to-video animation workflow will retain its drawn characteristics throughout motion transfer, while photorealistic references will yield more lifelike results. The model is optimized for cross-style pose replication without misalignment artifacts.

How does One-to-All Animation 1.3B differ from the 14B version?

The 1.3B version of One-to-All Animation targets accessibility and speed while maintaining moderate quality for video-to-video tasks. The 14B model supports sharper textures and higher resolutions (up to or beyond 1080p in some deployments), but it requires more compute and memory. For lightweight production pipelines, most developers use the 1.3B variant.

Can I use One-to-All Animation outputs commercially?

Yes, commercial usage of One-to-All Animation video-to-video outputs is generally permitted through licensed deployment on approved platforms like Fal.ai and RunComfy. However, you should review the specific license terms on the model’s official Hugging Face or Fal.ai page to verify rights for derivative content or resale.

How does One-to-All Animation perform when the reference and driving videos have different aspect ratios?

The One-to-All Animation model’s alignment-free pipeline tolerates varied aspect ratios between the reference image and motion-driving video-to-video input. It auto-normalizes poses spatially, ensuring smooth motion alignment and minimal distortion, though extremely wide or tall ratios might slightly reduce compositional fidelity.

What improvements in motion coherence does One-to-All Animation offer for longer videos?

One-to-All Animation introduces a token replacement mechanism that stabilizes long video-to-video sequences by progressively updating temporal tokens rather than re-encoding each frame independently. The result is fewer flickers and smoother transitions across complex motion arcs while retaining character details.

RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.