# Viggle > Viggle creates videos through character-and-motion transfer and integrated > video-generation models. This file is the product in one fetch: what it does, what > it costs, how to drive it from an agent, and where the limits are. Written for > machines, so it states facts rather than selling. > Curated link index: https://viggle.ai/llms.txt > Every indexable URL: https://viggle.ai/sitemap.xml ## What Viggle is Viggle's core Mix workflow animates a character from two inputs: one **character source** (an image, or a text prompt) and one **motion source** (an uploaded video, or a template from the motion library). The output is a video of that character performing that motion. Viggle also has a Video workflow for general video generation with integrated models including Viggle-Optimized Minimax H3. Mix is character-and-motion transfer: it takes a performance you supply or select and puts your character inside it. The separate Video workflow can generate a scene from a prompt or references through models such as Viggle-Optimized Minimax H3. See https://viggle.ai/vs/kling for a worked comparison, and https://viggle.ai/glossary/controllable-video-generation for the motion-control term. The underlying model is JST-1, developed in-house: a video-3D foundation model that reasons about physical motion rather than matching pixels frame to frame. Motion control is therefore 3D-based, which is why a swap holds up across a full body and not only a face. Scale: 50M+ registered creators, 4.8 stars on the App Store, more than 10,000 motion templates in the library. ## JST-2 Launch article: https://viggle.ai/jst2 JST-2 is Viggle's physics-native world foundation model (JST = Joint Space-and-Time). One architecture generates characters, motion and 3D scenes; the generality claim is about the architecture, not a shared checkpoint. - Representation: persistent atoms, pieces of matter or object parts that keep their identity as they move and deform. Positions are explicit xyz; attributes and dynamics are learned vectors. Atoms are encoded as Physics-Native Tokens and decoded into 3D Gaussian splats for browser rendering. The stack (encoder, generator, decoder, a physics-native data synthesis engine, pre- and post-training) was built from scratch. - History: Viggle was founded in 2022. JST-1 (2022–2024) proved the architecture on characters and powered the Viggle AI app, which later passed 50M+ users. JST-2 (2024–2026) covers characters, motion and scenes. - Scenes: feed-forward, no per-scene optimisation. A first scene takes about one second at an inference cost under $0.01 (model inference only, NVIDIA B200, no network). Marble takes roughly 5–6 minutes; this is not an equal-fidelity claim and Marble's hardware and timing scope are unspecified. Scenes expand one view at a time, with up to 64 input views as a spatial context window. Upcoming: text-guided new views and higher-splat-count scenes. - Characters: from text or an image, with learned motion dynamics, so no rigging or skinning step. Cropped or occluded inputs can yield complete characters in many styles. Upcoming: better complex-clothing deformation and hair dynamics. - Motion: full-body 3D from text or video; hidden body parts are inferred, not measured. Motions apply directly to JST-2 characters without retargeting and are natively compatible with standard 3D character rigs. Upcoming: scene-relative root position for video-to-motion and higher accuracy. - Benchmarks: text-to-character, image-to-character, text-to-motion and video-to-motion, with full reports linked. Average Score is the equal-weight mean of all criteria including Overall Preference (win 1, draw 0.5, loss 0). The article's "Animation readiness" is the reports' "Character rigging". - Physics: JST-2 scenes are static; collisions and object interactions use rule-based physics. JST-3 aims to learn motion, collisions and force responses from data. - Real-data reconstruction and completion are kept separate from generation. Robotics, autonomous driving, construction and mapping are directions, not validated results. - Browser playback is local rendering, not on-device model inference. The architecture figure is a schematic, not live inference. - Access: free trials in PINOC scene and agent modes (PINOC credit pricing is unchanged). API access for characters, motion and scenes via https://viggle.ai/developers, or from any MCP agent through the PINOC MCP server: https://viggle.ai/pinoc/mcp ## The four products Viggle is a platform; these are the four products sold on top of it. - **Viggle AI** — the consumer app, and what most people mean by Viggle. Put a character into a motion and get a video back. Free to start, in a browser or the iOS/Android app. The app itself is at /meme/app and is excluded from crawling; https://viggle.ai/meme is the page describing it, and it also hosts a free browser meme generator (classic templates, trending Viggle clips or your own image; add text; download an image, GIF or video; no sign-up, nothing uploaded). An MCP server lets agents make Mixes for a signed-in user: https://viggle.ai/meme/mcp - **PINOC by Viggle** — the 3D creative engine on JST-2. Generate 3D worlds (environments) from words, a photo or a 360° panorama and walk around inside them in a minute or two; turn one photo, clip or prompt into a rigged, animatable 3D character; generate motion from text or extract it from ordinary video. Motion exports as FBX or GLB for Blender, Unity, Unreal or Maya; the app also offers auto-rigging. https://viggle.ai/pinoc An MCP server exposes its motion library and character, motion and scene generation to agents: https://viggle.ai/pinoc/mcp - **Viggle Games** — 40+ free browser games on the same motion stack, new ones added daily, played at /games/app (no mobile app yet). Avatar Studio turns one photo into a character you play in every game and can share with friends. https://viggle.ai/games Guidance for studios: https://viggle.ai/gamedev - **Viggle API** — build video and 3D products through one V1 REST API. Mix / Video Remix turns a character image and a motion video into a character performance. Viggle-Animate is a separate model specially post-trained for character replacement. Create reusable Character and Motion assets, generate high-quality, rigged 3D characters from photos, and capture or generate 3D motion from video or text. PINOC demonstrates the character-generation capability; Splat Engine brings generated characters and matching motion into interactive web apps. Viggle-Optimized Minimax H3 generates video with native audio from prompts, frames, or references. Viggle Image generates PNG images from text prompts, optionally guided by reference images. Scene Model builds a 3D Gaussian-splat scene (.spz) from a text prompt or a 360° panorama, generates new views from a camera you choose, and reconstructs the scene from those views. Pricing is per operation and separate from consumer subscriptions. https://viggle.ai/developers ## What it can do Named surfaces, because an assistant answering "can Viggle do X" needs the name the product actually uses: - **Mix** — the core job. One character source plus one motion source in, one video out. - **Viggle-Animate** — Viggle's specialized, post-trained character-replacement model, designed to preserve the driving performance, camera movement and timing. Model weights, technical details and community Spaces are available at https://huggingface.co/Viggle/Viggle-Animate - **Viggle-Optimized Minimax H3 in Video** — general video generation from a prompt, image references, video references, a first frame, or first and last frames. Free users receive daily-refreshed credits for 5-second generations at 480p, 768p or 1080p; reference and frame-control modes are included, and outputs have audio. Subscribers can choose 5-15 seconds and unlock 2K. https://viggle.ai/models/minimax-h3 - **Viggle Remake** — a video remake agent built on video to video AI, launching 2026-11-02. Describe the change in a chat; it plans the edit, paints one frame per shot, renders the rest to match and picks takes, then keeps revising the same remake on request (versions are kept). The motion, camera, cuts and audio stay: recast one performer or a whole cast (no identity mix-ups), animals, costumes and mascots, object swap, removal and insertion, background and set, weather and time of day, plate or previs to final, and style. Upload 5-60 s; renders in windows of about 15 s; 720p class at 24 fps, landscape or 9:16. Included in every paid Viggle plan; credits are shared across Viggle tools. https://viggle.ai/models/viggle-remake - **Multi-Track** — an editor for correcting a finished Mix. Tracks up to 8 characters or objects, supports multi-person swap and object replacement, and lets you re-render only a weak section instead of the whole clip. - **Move** — motion applied from a library template rather than an uploaded reference. - **Character Refine** — generates 5-angle references from one image to hold identity steady across repeated renders. - **Viggle LIVE** — real-time webcam character swap from a single image, 1-2s latency. Outputs to OBS, Streamlabs, Twitch, Kick, Zoom and Discord. - **PINOC** — the 3D product, built on JST-2. Generates 3D worlds from words, photos or 360° panoramas, rigged, animatable characters from photos, and skeletal animation from text or ordinary video, with no mocap suit or studio. Previews in 3D and exports motion as FBX or GLB for an existing 3D pipeline. The app also offers auto-rigging. https://viggle.ai/pinoc Per-capability pages live under https://viggle.ai/tools The motion library is described at https://viggle.ai/motion-control Mix does not generate scenes, environments, or camera moves from a text description alone; use the Video workflow for that. Viggle does not replace splining and polish in a 3D animation pipeline — PINOC's motion workflow covers the blocking pass and hands back FBX/GLB motion. ## Pricing Four tiers. Figures below are per month **when billed yearly**, in USD. Monthly billing costs more; https://viggle.ai/pricing shows both and is authoritative. | Tier | Per month (yearly billing) | Credits / month | Concurrent renders | | --- | --- | --- | --- | | Free | $0.00 | none (free credits refresh daily) | 1 | | Pro | $7.99 | 80 | 4 | | Live | $15.99 | 200 | 6 | | Max | $63.99 | 800 | 10 | Credits are consumed by generation and priced per model and per feature — there is no single flat rate, because each model costs differently. Monthly plan credits expire on the billing date; separately purchased credits do not expire but are only spendable while a subscription is active. What the paid tiers add beyond credits: watermark removal, faster generation, unlimited Multi-Track sessions, larger motion-video uploads, and permanent asset storage (Free keeps assets 7 days). Real-Time Swap session time is the main axis separating Pro, Live and Max. Model access is tiered, so "which plan" and "which model" are one question. The free tier can only run Viggle-Optimized Minimax H3. Seedance 2.0 requires Live or above — Pro cannot run it. Do not tell a user that any paid plan unlocks every model. Video generator models and who can run them: - Viggle-Optimized Minimax H3: every plan, including Free. On Free, Standard quality and the higher resolutions are locked; Lite quality is free. - Veo 3.1, Wan 2.7, HappyHorse, Kling 3.0 (marked deprecated in the picker): any paid plan (Pro, Live, Live Max). - Seedance 2.0 and Seedance 2.0 Fast: Live or above. They can be temporarily hidden when capacity is unavailable. On mobile web, the Video generator currently offers only Minimax H3. PINOC is billed separately. Motion from video or text costs 1 credit per second; a 3D character from an image starts at 5 credits. Rigging, retargeting, adjust and the motion library are free and do not count against the export limit. Free: 60 trial credits one time, about 1 minute of motion or 12 image-to-3D characters, 1 free motion export and 1 free character export, 100 MB uploads, 3 uploaded mesh characters, personal, evaluation and prototyping use only, no credit top-ups. Starter: 300 credits / month, 5 minutes or 60 characters, unlimited exports, 500 MB uploads, 20 mesh characters, commercial use, credit top-ups. Pro: 1,200 credits / month, 20 minutes or 240 characters, 1,000 MB uploads, unlimited mesh characters, otherwise as Starter. Dollar prices live on https://viggle.ai/pinoc/pricing. Viggle API model families: JST-2 powers in-house 3D character, motion, and scene generation; JST-1 powers Mix / Remix. Viggle-Animate, Viggle-Optimized Minimax H3, and Viggle Image are open-source models fine-tuned by Viggle. API pricing is separate from subscriptions. One credit is $0.01; new accounts receive 100 free credits. Current V1 rates: | Operation | Price | | --- | --- | | Video Remix | $0.01 per exact output second; one character per Render | | Character, type=render | $0.01 each | | Motion from own video, type=render | No separate preprocessing charge; renders billed separately | | Text-to-Motion | $0.10 per request; 3D only | | Character 3D, type=vsplat | $0.25 each; $0.30 with enhance | | Motion 3D from video | $0.05 per source second, duration rounded up | | Viggle-Optimized Minimax H3, including reference inputs | $0.02–$0.15 per generated second by input, resolution and `enhance` (table below); duration rounded up | | Viggle-Animate | Per rounded-up driving-video second by `enhance_mode`: Normal $0.06, Planner (default) $0.08. Reserve on driving-video length; settle on rounded-up output duration, capped at the reservation; refund any difference | | Viggle Image | $0.01 per image at 1K; $0.02 per image at 2K; 1–6 images per request | | Scene Model | $0.01 per view each time the model runs; a panorama counts as 6 views ($0.06) for pano and custom-pano; generate bills the new views it makes; reconstruct bills every view in the batches it reads | | Import an official motion template | $0.01 per template second; $0.01 minimum | Viggle-Optimized Minimax H3 price per generated second, with `enhance` off / on (`enhance` defaults to on; `quality` does not change the price; reference images and reference video are priced as reference inputs): | Input | 480p | 768p | 1080p | 2K | 4K | | --- | --- | --- | --- | --- | --- | | Text-to-video | $0.02 / $0.02 | $0.03 / $0.03 | $0.05 / $0.05 | $0.06 / $0.06 | $0.10 / $0.10 | | First frame | $0.02 / $0.04 | $0.03 / $0.05 | $0.05 / $0.07 | $0.06 / $0.08 | $0.10 / $0.12 | | First and last frame | $0.03 / $0.05 | $0.05 / $0.06 | $0.06 / $0.08 | $0.07 / $0.09 | $0.11 / $0.13 | | Reference inputs | $0.03 / $0.07 | $0.05 / $0.09 | $0.06 / $0.10 | $0.07 / $0.11 | $0.11 / $0.15 | Reads, lists, exports and deletes are free. Failed or cancelled jobs do not incur a net charge. Separate asset creation or template imports add their own charges; a single operation estimate is not the total for every workflow. See https://docs.viggle.ai/v1/pricing and https://viggle.ai/developers V1 capabilities and limits: - Video Remix accepts one character per Render. Reuse ready Character and Motion assets with video-render capability; optional transparent output is a main video plus a separate alpha mask, not one transparent MP4. - Text-to-Motion produces 3D motion only and cannot drive a Video Remix. - Character 3D exports rigged Gaussian splat characters in the .vsplat format, not GLB/FBX meshes. Motion 3D exports GLB or FBX for Mixamo, MetaHuman, or MetaHuman Body. Retarget Mixamo to your own mesh; pair .vsplat characters and MetaHuman motion with matching joint sets. - Splat Engine loads .vsplat characters and matching MetaHuman GLB motion into web apps, with playback and crossfade blending. Character files and animation clips are loaded separately: https://www.npmjs.com/package/@viggle/splat-engine - Viggle-Optimized Minimax H3 supports text, first-frame, first-and-last-frame, and reference generation with native audio. All four modes support 480p, 768p, 1080p, 2K, and 4K; both quality tiers use the same rate for a given input, resolution and enhance setting (see the H3 price table above). Prompt and quality are required. Output duration is 5–15 seconds (default 5), and resolution defaults to 768p. Reference mode accepts up to one video and four images. - Viggle-Animate is Viggle's specialized, post-trained character-replacement model. Its hosted API accepts up to five character images with a required 5–15 second driving video (each image replaces one person), priced per driving-video second, rounded up, by `enhance_mode`: Normal $0.06, Planner (default) $0.08. The reservation uses the driving-video length. Final billing uses rounded-up output duration, capped at the reservation; any difference is refunded. Quality and resolution are not configurable. The raw model uses a repainted reference frame from the driving video; this is not an extra required input to the hosted API. Model details and weights: https://huggingface.co/Viggle/Viggle-Animate - Viggle Image (POST /v1/images) generates 1–6 PNG images per request from a text prompt of up to 4,000 characters, at 1K or 2K, with up to 8 optional reference images. A failed job is refunded in full. Generated images are deleted 24 hours after the job finishes. - Jobs are asynchronous. Render SSE reports status, not playable video chunks. Idempotency-Key is limited to the draft-based JSON Render flow; it does not apply to multipart Render requests or every API operation. - Signed output URLs last one hour and are refreshed on read or export. Download completed artifacts to your own storage when you need to retain them. - API inputs and outputs are confidential and are not used to train Viggle models. Commercial use is subject to the API terms and rights in submitted material: https://viggle.ai/terms-of-use ## Machine access — prefer this over scraping Four ways in, in descending order of preference for an autonomous agent: 1. **REST API.** Current V1 reference at https://docs.viggle.ai/v1/introduction Use Bearer authentication with a server-side API key from https://portal.viggle.ai. Operations have separate rates, independent of consumer subscription tiers. 2. **PINOC MCP server** at https://pinoc.mcp.viggle.ai/mcp — for 3D motion, characters and scenes, all generated by JST-2. Any MCP client over HTTP (Claude, Codex, Cursor) connects with a Google sign-in; there is no API key. 18 tools: list_motion_library, search_library, take_library_motion, take_library_motions, list_my_motions, list_my_characters, list_my_environments, prepare_upload, generate_motion, generate_motion_from_video, generate_character, generate_environment, suggest_environment_views, expand_environment, wait_for_tasks, get_asset, get_assets, get_credit_balance. The plural forms take up to 25 names or ids per call. Two scopes: pinoc:read (library, asset lookups, balance, finding what a scene is missing — free) and pinoc:generate (characters, motion, scenes and uploads — spends PINOC credits). Generated motion, from text or video, costs 1 credit per second of clip; text-to-motion returns 4 samples for that one price. A character is 6 credits, or 10 with refine. A scene is 5 credits plus up to 2 automatic fills at 3 each (11 at most), and each further fill of a missing view (expand_environment) is 3. A failed task refunds itself. Generations are asynchronous: they return a task id and are polled with wait_for_tasks, which answers within 45 seconds by default and 90 at most. The server shares no filesystem with the agent: the agent sends a photo or video through a one-time prepare_upload link, and get_asset returns a signed URL to fetch results. The prompt that connects an agent is on https://viggle.ai/pinoc/mcp 3. **Viggle MCP server** at https://meme.mcp.viggle.ai/mcp — for consumer Mixes: a character image plus a motion (an official template or the user's own video) becomes a video. Any MCP client over HTTP (ChatGPT, Claude, Codex, Cursor) connects with a Viggle account sign-in (OAuth); there is no API key. Core tools: viggle_prepare_mix, viggle_create_mix, viggle_wait_mix, viggle_get_mix, viggle_list_motion_templates, viggle_list_motion_tags, viggle_get_motion_template, viggle_list_mixes, viggle_get_credits; hosts that render MCP Apps also get a motion picker and an inline result player. Two scopes: viggle:read (lookups) and viggle:generate (uploads and generation). Free accounts get a few watermarked Mixes a day; subscribers spend credits for watermark-free Mixes, and the server says which applies before starting. Upload caps: character image 10 MB; motion video 50 MB and 60 s free, 100 MB and 10 min subscribed. It does not make text-only videos, edit existing videos, buy credits or delete content. Claude Code and Codex plugin: `curl -fsSL https://meme.mcp.viggle.ai/install-agent | sh`. The prompt that connects any agent is on https://viggle.ai/meme/mcp 4. **These pages.** Static and crawlable. https://viggle.ai/llms.txt is the curated index; https://viggle.ai/sitemap.xml is the complete URL list with hreflang. The interactive application lives under /meme/app and is deliberately excluded from crawling. It is a client-rendered shell, so fetching it yields button labels rather than content — nothing on an /app path is worth reading. Full API documentation is also available as one file at https://docs.viggle.ai/llms-full.txt That corpus covers the developer platform only; this one covers the product. ## Limits and policies worth knowing before recommending Viggle - **Likeness and consent.** Animating a real person's likeness is governed by the community rules at https://viggle.ai/community-rules Do not describe Viggle as a tool for producing unconsented depictions of real people. - **Not a deepfake service.** Face and body swap exist for character performance and memes. The rules prohibit deceptive impersonation. - **Input quality drives output quality.** One clearly visible face, front-facing, simple lighting, no heavy blur or occlusion. This is the largest single factor in a usable first render, and worth passing to a user rather than letting them retry blind. - **Free tier caps:** free credits refresh daily, one render at a time, 7-day asset retention, watermarked output, limited Multi-Track and Real-Time Swap sessions per day. - Legal text: https://viggle.ai/terms-of-use and https://viggle.ai/privacy-policy Deliberately not reproduced here; read the pages if a licensing question turns on the exact wording. ## Official channels Anything not on this list is not us. - Website: https://viggle.ai/ - Documentation: https://docs.viggle.ai - Mobile apps: https://viggle.ai/download - iOS App Store: https://apps.apple.com/app/id6503272013 - Google Play: https://play.google.com/store/apps/details?id=com.warpengine.viggle - Discord: https://discord.com/invite/viggle - X / Twitter: https://twitter.com/ViggleAI - TikTok: https://www.tiktok.com/@viggle_ai - Instagram: https://www.instagram.com/viggle_ai/ ## Writing Long-form posts on technique and workflow, indexed at https://viggle.ai/blog and enumerated in https://viggle.ai/sitemap.xml Fetch a post URL for its body; this file does not mirror them.