FramePackLoop
Free
FramePackLoop is an open source video loop generation tool based on lllyasviel/FramePack. It provides loop video generation capabilities through the Gradio interface and supports seamless loop video production from beginning to end.
FramePackLoop
Core parameters and statistics
| Project | Specifications |
|---|---|
| Product Name | FramePackLoop |
| Category | AI video generation |
| Delivery form | Desktop tool (Gradio Web UI) |
| Support Platform | Desktop |
| Supported languages | ja, en, zh-CN |
| Target users | VTuber, independent game developers, AI video creators |
| User scale | GitHub 85 Stars (niche professional tool) |
| Pricing model | Free (Apache-2.0 open source) |
Platform coverage and user scale data are based on the official real-time page and third-party statistics.
User and market recognition
FramePackLoop has 85 stars on GitHub. It is a niche professional tool, but it has gained some attention in the AI video generation community (especially in the Japanese market). Mainly used by VTuber and live broadcast hosts (used for live broadcast background and standby screen), independent game developers (produce looping background animation and UI animation), and AI video creators (integrated in ComfyUI/Gradio workflow). The project provides three operating modes: FramePackLoop (image to loop with video), FramePackConnect (video connection), FramePackVideo2LoopVideo (video to loop). It has been verified that loop playback is normal in CapCut and OBS Studio.
Cost advantage
| Cost Dimension | Description |
|---|---|
| Software License | Apache-2.0 open source, completely free |
| Hardware Cost | NVIDIA GPU required (6GB VRAM minimum, 12GB+ recommended) |
| Cloud GPU rental | About ¥1-5/hour (such as RunPod, AutoDL) |
| Storage cost | Output MP4 file, single video 10-50MB |
Compared with commercial loop generation services (such as Runway and Pika Labs subscription systems), FramePackLoop has zero software cost and no subsequent usage fees after one-time hardware investment.
Main functions
- Image to Loop Video: Input a single image and generate a seamless loop animation from beginning to end. It supports adjusting the main video length, connected video length, padding length and other parameters, and fine-grained control is effective.
- Video to loop video (Beta): Automatically detect and complete the differences between the beginning and the end of the generated FramePack video, and output the loop version.
- Video Connection (Connect, Beta): Smoothly connect two independent videos through AI tweening to generate a continuous video with natural transitions.
- LoRA support: Supports loading of FramePack-specific LoRA models, and is also compatible with some HunyuanVideo using LoRA to achieve stylized video generation.
- Latent preprocessing: Supports batch generation of Latent files (only about 5% storage overhead), and then decodes them into videos after filtering, saving GPU time.
Model and version evolution
| Version | Date | Key Changes |
|---|---|---|
| windows-v1.6 | 2025-06-08 | Added next latent video mode support |
| windows-v1.5 | 2025-05-14 | Added LoRA function support |
| windows-v1.4 | 2025-05-06 | Added Connect and Video2Loop functions |
| windows-v1.3 | 2025-05-02 | Initial Windows add-on package release |
Technical advantages
- Based on FramePack efficient architecture: Inherits the frame context packaging and drift prevention technology of lllyasviel/FramePack, and can run on 6GB VRAM.
- Connected Video Mechanism: The core innovation is to generate additional video clips that specifically merge the first and last frames of the main video, while referring to the context of "future frames" and "past frames" to make the transition natural.
- Latent preprocessing pipeline: First quickly explore different Seed effects in batches, and then perform complete VAE decoding after filtering to avoid wasting GPU time with invalid complete rendering.
- LoRA dynamic injection: Inject LoRA weights at startup through configuration files, support multiple LoRA overlays, and switch styles without modifying the code.
How to use
| Entrance | How to use |
|---|---|
| Python source code | git clone → Install dependencies → python demo_gradio_loop.py → Browser operations |
| Windows add-on package | Download run_loop.zip → Extract to FramePack directory → Run run_loop.bat |
Key parameters: Main Video Length (main video segment length), Connection Video Length (connection video segment length), Loop Num (number of loop repetitions), Seed (fixed random seed).
Product Pricing
| Package | Price | Contents |
|---|---|---|
| Open Source Core | Free | Full Features (Apache-2.0) |
There are no payment tiers or hidden charges. The cost of use is only the cost of bringing your own GPU hardware or renting a cloud GPU instance.
Application scenarios
- VTuber/Live Broadcast Background Material: Generate unlimited looping standby animation background and overlay elements, OBS Studio has verified compatibility. LoRA can control the picture style to be consistent with the character IP.
- Social Media Creative Video: The short video platform has backgrounds and dynamic covers, and CapCut has been verified to be compatible with the import.
- Game Development Materials: Loops in independent games have background animations and UI transition effects, and the video-to-loop function can convert existing rendered clips into seamless materials.
Applicable people
- Individual users: AI video creator and VTuber, free and open source to reduce material production costs, LoRA supports personalized style control.
- SME Team: Independent game developers need small teams with animation assets.
- Large Enterprise: Not suitable - No commercial-grade SLA and technical support.
- Not suitable for boundaries: Unable to be used by users without NVIDIA GPU; professional film and television scenes that require high-precision controllable loops have limited effects.
Comparison of competing products
| Compare Dimensions | FramePackLoop | Runway | Pika |
|---|---|---|---|
| Core differences | Xunyou video-specific tools | General video generation | General video generation |
| Price | Free (Open Source) | Starting at $15/month | Starting at $10/month |
| Covering Scenarios | Loop/Connect Video | Professional Video Editing | Social Media |
| User reviews | Effectiveness relies on "luck" | Stable and reliable | Easy to use |
| Technical threshold | Medium (GPU required) | Low | Low |
Summary and Outlook
FramePackLoop finds a differentiated positioning in the FramePack ecosystem by specializing in the "loop and connection" pain points in AI video generation. Functions such as connected video mechanism, Latent preprocessing, and LoRA support provide creators with a complete tool chain. The project is open source as Apache-2.0 and has no paywall. Current limitations: The tool is in the Beta stage, and the cycle effects are random; the documentation is mainly in Japanese; only supports FramePack ecological input.
Acquisition/Adoption Risk Assessment: As a personal open source project, there is uncertainty about long-term maintenance and upstream compatibility. Not available for users without NVIDIA GPU. It is recommended to pay attention to project activity and community feedback before deciding whether to rely on it as a formal production tool. | Product type | Open source video loop generation tool | | Basic Framework | lllyasviel/FramePack | | Platform Support | Desktop (Gradio Web UI) | | First Published | April 2025 | | Current version | windows-v1.6 (2025-06-08) | | License | Apache-2.0 | | Development Language | Python | | GitHub Stars | 85 | | Core Competencies | Image/Video→Seamless video generation |
FramePackLoop is an open source video loop generation tool created by developer red-polo based on lllyasviel/FramePack. Its core mission is to solve the "end-to-end connection" problem of AI-generated videos - allowing the beginning and end of the animation to be naturally connected to generate video material that can be played infinitely. FramePack itself is an efficient video diffusion framework for consumer-grade GPUs (minimum 6GB VRAM). Based on this, FramePackLoop has been specially extended for the continuous video scenario.
The tool's workflow is divided into four steps: first, generate a main video clip based on the input image, then generate a "connected video" that connects the beginning and end of the main video, then merge the main video and the connected video into a single-loop video, and finally output a specified number of long-loop videos through repeated splicing. The entire process is done within the Gradio web interface, no coding required.
User and market recognition
FramePackLoop has 85 stars on GitHub. Although it is a niche professional tool, it has gained some attention in the AI video generation community (especially in the Japanese market). The project is mainly used by the following groups:
- VTuber and live broadcast host: use looped videos as live broadcast background and standby screen Overlay materials;
- Independent Game Developer: Create UI dynamic elements with background animation;
- AI Video Creator: A supplementary tool that integrates the FramePack ecosystem in the ComfyUI/Gradio workflow.
The project comes with three operating modes:
- FramePackLoop (main mode): generate looped video from a single image;
- FramePackConnect (Beta): Smoothly connect two independent videos through AI tweening;
- FramePackVideo2LoopVideo (Beta): Convert existing videos to loop versions.
This tool has been verified to work properly in CapCut and OBS Studio, and the generated files can be directly used for live streaming and video editing.
Cost advantage
FramePackLoop is completely free and open source (Apache-2.0), with no paid tiers or hidden charges. The cost of use only includes:
- Hardware Cost: Requires NVIDIA GPU (minimum 6GB VRAM, 12GB+ recommended), or use a cloud GPU instance (e.g. RunPod, AutoDL).
- Power/Time Cost: Video generation time depends on GPU model, video length and number of iteration steps. Taking RTX 3060 12GB as an example, it takes about 3-5 minutes to generate a looped video of about 2 seconds.
- Storage Cost: Output is MP4 file, single video is usually between 10-50MB.
Compared with similar commercial services (such as the subscription system of Runway and Pika Labs), FramePackLoop's zero software cost model is very attractive to users with local GPUs - there is no subsequent usage fee after a one-time hardware investment. But for users who don’t have a GPU, the cloud GPU rental cost (about ¥1-5/hour) needs to be factored into the budget.
Main functions
- Image to Loop Video: Input a single image and generate a seamless loop animation from beginning to end. It supports adjusting the main video length, connected video length, padding length and other parameters, and fine control is effective. Core parameters include
Main Video Length,Connection Video Length,Padding Video Length,Loop Num, etc. - Video to Loop Video (Beta): Take the generated FramePack video as input, automatically detect and complete the difference between the beginning and the end, and output the loop version. Supports setting the length and number of loops of the connected video, which is suitable for secondary processing of existing video materials.
- Video Connection (Connect, Beta): Smoothly connect two independent FramePack-generated videos through AI tweening to generate a continuous video with a natural transition. It is suitable for scenarios such as game material splicing and multi-segment AI video combination.
- LoRA support: Supports loading of FramePack's dedicated LoRA model (can be trained through Musubi Tuner), and is also compatible with some HunyuanVideo's LoRA to achieve stylized video generation. Specify the model path and application weights by editing the
lora_setting.jsonconfiguration file. - Batch generation and Latent preprocessing: Supports the
Generation Countparameter to achieve multiple automatic generation;Without VAE Decodemode can first generate Latent files in batches (only about 5% storage overhead), filter them and then selectively decode them into videos, which greatly saves GPU time and storage space during batch exploration. - System Options and Preview: Support
system_setting.jsonconfiguration, which can switch next latent preview from static image to real-time video preview.
Model and version evolution
| Version | Release Date | Core Changes |
|---|---|---|
| Windows initial version | 2025-04-29 | The first Windows add-on package, providing basic functionality |
| Connect (Pre-release) | 2025-05-01 | Added FramePackConnect video connection function |
| windows-v1.3 | 2025-05-02 | Official version of Windows add-on package |
| windows-v1.4 | 2025-05-06 | Added Connect and Video2Loop functions |
| windows-v1.5 | 2025-05-14 | Added LoRA function support |
| windows-v1.6 | 2025-06-08 | Added next latent video mode support |
The main line of version evolution: from basic loop generation to a multi-functional tool set including Connect (video connection) and Video2Loop (video loop conversion), and the introduction of LoRA style control capabilities. The project is developed based on the 2025/04/28 main branch (commit 6da55e8) of lllyasviel/FramePack.
Technical advantages
- Based on FramePack efficient architecture: FramePack is a consumer-grade GPU video diffusion framework developed by the lllyasviel team. Through Frame Context Packing and Drift Prevention technology, a complete video can be generated on 6GB VRAM. FramePackLoop inherits this efficient bottom layer to ensure that loop generation tasks can also run on low-memory GPUs.
- Connection Video Mechanism: The core innovation of FramePackLoop is "Connection Video" - the additional video clips generated are specifically used to fuse the first and last frames of the main video. The implementation is to simultaneously refer to the context information of "future frames" (the beginning of the main video) and "past frames" (the end of the main video) when generating the connected video, so that the transition between the beginning and the end is natural and smooth.
- Latent preprocessing pipeline: supports lightweight generation mode that does not perform VAE decoding and directly outputs Latent image files. Users can quickly explore animation effects under different Seeds in batches and then perform complete VAE decoding on satisfactory results to avoid wasting GPU time with invalid complete rendering. Latent files only account for about 5% of full video storage.
- LoRA dynamic injection: Inject LoRA weight at startup through
lora_setting.json, support multiple LoRA overlays (control the weight through the scale parameter), and switch styles without modifying the code. LoRA takes effect on the entire generation pipeline, affecting visual output such as picture style and character characteristics. - Windows configuration-free package: Provides an independent
run_loop.zipadditional package. FramePack users only need to unzip it and place it in the specified directory to use it. No Python context configuration is required. Lowers the threshold for non-technical users.
How to use
FramePackLoop provides two ways of use:
Method 1: Run Python source code (recommended for developers)
- Clone the repository:
git clone https://github.com/red-polo/FramePackLoop.git - Install dependencies: Install Python dependencies according to FramePack’s official configuration requirements
- Start main mode:
python demo_gradio_loop.py - Open the Gradio interface in the browser (default
http://127.0.0.1:7860) - Upload the image, adjust the parameters, and click "Start Generation" to generate a loop video
Method 2: Windows add-on package (for users who have installed FramePack)
- Download
run_loop.zipfrom Releases - After unzipping, place the
run_loopfolder into the FramePack installation directory - Run
run_loop.batto start the tool - The output file is located in the
run_loop/output/directory
Other mode startup commands:
- FramePackConnect:
python demo_gradio_connect.py - FramePackVideo2LoopVideo:
python demo_gradio_video2loop.py
Key parameter description:
Main Video Length: Main video segment length (unit: section, about 1 second/section)Connection Video Length: The length of the connection video segment, ideally the same as the main videoPadding Video Length: starting padding length, 0 means starting directly from the input imageLoop Num: Outputs the number of repetitions of the loopGeneration Count: Number of batch generation (automatic random seed when ≥ 2)Seed: fixed random seed, used to reproduce specific animation effects
Product Pricing
FramePackLoop is completely free and open source (Apache-2.0 license) without any paid tiers. All features are open to all users.
The cost of using this tool is only at the hardware level:
- Bring your own GPU: One-time hardware investment (minimum NVIDIA GPU with 6GB VRAM, RTX 3060 12GB or higher recommended).
- Cloud GPU instance: Billed by the hour, about ¥1-5/hour (take RunPod, AutoDL and other platforms as examples, subject to real-time quotation).
- No Hidden Consumption: No subscription fees, no API call fees, and no forced embedding of watermarks.
Application scenarios
- VTuber/Live Broadcast Background Material: Generate unlimited looping standby animation background, transition screen and overlay elements. OBS Studio has verified playback compatibility. Creators can use LoRA to control the picture style to be consistent with the character IP, without having to regenerate it every time they broadcast.
- Social Media Creative Video: Produce short video platform with backgrounds, dynamic covers and creative short videos. CapCut editing software has verified import compatibility. The single image generation mode is suitable for quickly batch producing different styles of loop materials.
- Game development materials: Circular background animation UI transition special effects and ambient atmosphere animation in independent games. The video looping function can convert existing game rendering clips into seamless looping materials, reducing the repeated production of art resources.
- AI Video Research and Experiment: Study the loop generation capabilities of the FramePack video diffusion model, and explore the performance of frame context packaging technology in head-to-tail connection scenarios. The technical documentation provides detailed parameter descriptions and engineering pitfall records.
Applicable people
- AI Video Creator and VTuber: Individual creators who need high-quality looping background and standby animations. The free open source model reduces material production costs, and LoRA support enables personalized style control.
- Indie Game Developer: Small teams or independent producers who need to develop animation assets. Local GPU operation does not require dependence on cloud services and is suitable for offline development.
- AI/ML Research & Enthusiasts: Researchers interested in the FramePack video diffusion model and its derived applications. The project code is open source and can be used as the basis for experiments and secondary development.
- FramePack users: Users who are already using lllyasviel/FramePack can expand their video generation capabilities at zero cost.
- Not suitable for people: Users who do not have NVIDIA GPU and are unwilling to use cloud GPU (CPU mode is not supported); professional film and television production scenarios that require 4K ultra-high-definition output or complex multi-segment narratives (FramePackLoop is mainly for short loops and clips); enterprise users who require commercial-grade SLA and technical support (open source projects have no official customer service).
Summary and Outlook
FramePackLoop has found a precise and differentiated positioning in the lllyasviel/FramePack ecosystem by specializing in the "loop and connection" pain points in AI video generation. Its connection video mechanism, Latent pre-processing pipeline, LoRA support and other functions provide creators with a complete tool chain from images to seamlessly looping videos. The project is fully open source with Apache-2.0, no paywall, and extremely friendly to users with local GPUs.
Current limitations: The tool is in the Beta stage, and there is a certain degree of randomness in the looping effect (the developer admits that the smoothness of the start-to-end connection partly depends on "luck"); the documents are mainly in Japanese, and non-Japanese users may need to use translation tools; only videos generated within the FramePack ecosystem are supported as input.
Follow-up observation points: whether the project will keep up with FramePack upstream updates; whether the LoRA ecosystem will be further enriched; whether a NoGPU online version will be launched to lower the usage threshold.
Not suitable for boundaries: Users without NVIDIA GPU cannot use it; professional film and television scenes that require high-precision controllable looping (exact matching of specified first and last frames) have limited effects. Adoption Risk: As a personal open source project, there are uncertainties in long-term maintenance and upstream compatibility; it is recommended to pay attention to project activity and community feedback before deciding whether to rely on it as a formal production tool.
Related tools: runway, pika
Version Info
- Additional program for Windows ：Added next latent video mode support
- Additional program for Windows ：Added LoRA function support
- Additional program for Windows ：Added Connect and Video2Loop features
- Additional program for Windows ：Initial Windows add-on package release
User Reviews