Releases: intel/AI-Playground
Releases · intel/AI-Playground
AI Playground v2.6.2 beta
AI Playground Version 2.6.2
This is a minor release of AI Playground supporting bug fixes and code improvements
Important
This version supports additional NPU models. Be sure to update your NPU drivers and restart your PC before using.
This version also supports WAN 2.1 VACE advanced video generation models, which run best on discrete GPUs. While these can run on Intel Core Ultra with built-in Intel Arc GPUs, a discrete GPU is highly recommended for optimal performance.
Version 2.6.2b Features
- All features from the previous release (v2.6.1b) are included, unless noted in the "Fixes and Deprecations" section.
Version 2.6.2b Fixes and Deprecations
- fixes a conflict between OpenVINO and latest PyTorch 2.9 that prevented document embedding in OpenVINO
- restore last used inference device per backend after AI Playground restart
- avoid global PIP config file from being loaded during package installations
- remove obsolete device selection in basic settings
- remove additional DLLs from build
- update npm dependencies
- improve dll path resolution and fixes c10_xpu.dll dependency error for default and ComfyUI backends
- fix ollama PoC
Known Issues
- Some features may fail on the first run after downloading models and nodes, requiring a restart of AI Playground. If a model fails immediately after it's downloaded, restart AI Playground and try again.
- An interrupted download can block models from downloading properly. To fix this, go to the model location, delete the temporary file, and try again.
- The WAN 2.1 VACE inference preview may appear corrupted during steps, but the final output will be fine.
- The device selector in Image settings may default to the wrong device, requiring a manual setting after restarting AI Playground.
- Installing over a previous version of AI Playground without a clean install can cause performance issues, lag when typing in prompts, and errors.
- The Colorizer workflow is limited to Xe2-supported GPUs (LNL and BMG).
- The installer can fail due to environmental issues. See the Troubleshooting guide for help.
- A PC restart may be needed after installation for optimal functionality and performance.
- Reinstalling over an existing AI Playground installation will delete any ComfyUI models and custom nodes.
- Users can adjust settings beyond their specific system's ability to generate or complete a task.
Version 2.6.2b Supported Hardware
- Intel Core Ultra 200H Processors (ARL-H)
- Intel Core Ultra 200V Processors (LNL)
- Intel Core Ultra 100H Processors (MTL-H)
- Intel Arc B Series GPU Cards (BMG)
- Intel Arc A Series GPU Cards with 8GB+ Memory (ACM)
AI Playground v2.6.1 beta
AI Playground Version 2.6.1b
This release updates AI Playground with minor fixes and improvements
Important
This release is using a new signed key, which resets Microsoft's reputation algorithm metrics for the app. Downloads may give a warning the app is not trusted. Select "Keep" and that you trust the app to bypass.
This version supports additional NPU models. Be sure to update your NPU drivers and restart your PC before using.
This version also supports WAN 2.1 VACE advanced video generation models, which run best on discrete GPUs. While these can run on Intel Core Ultra with built-in Intel Arc GPUs, a discrete GPU is highly recommended for optimal performance.
Version 2.6.1b Features
- All features from the previous release (v2.6.0b) are included, unless noted in the "Fixes and Deprecations" section.
- Context size added for OpenVINO on NPU devices
- OpenVINO updated to 2025.3 for better management of Tokens/Context
- Added Nomic embeddings model to Llama.cpp
- Upgraded Llama.cpp version
- Improved conversation manager UX (@KyleHagy) - allows user to rename discussion
- Zoom UI enabled using CTRL+mouse wheel and CTRL 0 to reset zoom
- Huggingface endpoint configurable per issue 315
- via settings.json: Default is
"huggingfaceEndpoint": ["https://huggingface.co"](https://huggingface.co/)
- via settings.json: Default is
- Improve backend installation and startup error display
- add pip freeze output of the failing backend to error details
- capture error messages during backend startup and add to error details
Version 2.6.0b Fixes and Deprecations
- CopyFace and FaceSwap fixed by pinning the stringzilla dependency of insightface to 3.12.6
- BasicSR component removed as an unneeded dependency
- System level DLLs re-packaged to support installation requirements on some systems
Known Issues
- Some features may fail on the first run after downloading models and nodes, requiring a restart of AI Playground. If a model fails immediately after it's downloaded, restart AI Playground and try again.
- An interrupted download can block models from downloading properly. To fix this, go to the model location, delete the temporary file, and try again.
- The WAN 2.1 VACE inference preview may appear corrupted during steps, but the final output will be fine.
- The device selector in Image settings may default to the wrong device, requiring a manual setting after restarting AI Playground.
- Installing over a previous version of AI Playground without a clean install can cause performance issues, lag when typing in prompts, and errors.
- The Colorizer workflow is limited to Xe2-supported GPUs (LNL and BMG).
- The installer can fail due to environmental issues. See the Troubleshooting guide for help.
- A PC restart may be needed after installation for optimal functionality and performance.
- Reinstalling over an existing AI Playground installation will delete any ComfyUI models and custom nodes.
- Users can adjust settings beyond their specific system's ability to generate or complete a task.
Version 2.6.1b Supported Hardware
- Intel Core Ultra 200H Processors (ARL-H)
- Intel Core Ultra 200V Processors (LNL)
- Intel Core Ultra 100H Processors (MTL-H)
- Intel Arc B Series GPU Cards (BMG)
- Intel Arc A Series GPU Cards with 8GB+ Memory (ACM)
Contributors
KyleHagy
Assets 3
AI Playground v2.6.0 beta
AI Playground Version 2.6.0b
This release updates AI Playground with new models for the Create and Answer sections, and refreshes the backends and PyTorch versions to support the latest models, such as GPT-OSS.
Important
This version supports additional NPU models. Be sure to update your NPU drivers and restart your PC before using.
This version also supports WAN 2.1 VACE advanced video generation models, which run best on discrete GPUs. While these can run on Intel Core Ultra with built-in Intel Arc GPUs, a discrete GPU is highly recommended for optimal performance.
Version 2.6.0b Features
- All features from the previous release (v2.5.5b) are included, unless noted in the "Fixes and Deprecations" section.
- PyTorch backend updated to PyTorch 2.8.
- Flux.1 Kontext [dev] "Edit By Prompt" workflow added.
- WAN 2.1 VACE workflows added for image-to-video and video-to-video.
- Workflow dropdown categories added for a cleaner UI and easier access.
- A Launch ComfyUI button added to Image settings.
- Drag-and-drop support for adding history images to workflow settings.
- Acer VisionArt workflow added for upscaling and placing generated images on your desktop (Acer AI PC required).
- Demo mode added, which provides guided tips for kiosks, demos, and retail experiences.
- Llama.cpp backend switched to the Vulkan version.
- GPT-OSS 20B Q8 model added to Llama.cpp.
- Additional compressed weight (cw) models added for OpenVINO using NPU or GPU.
- Backend version control added, allowing users to adjust the versions of all installed backends (ComfyUI, Llama.cpp, or OpenVINO).
- Context Size setting added to Llama.cpp, allowing users to set the context with data on model memory usage displayed in the console.
- Markdown formatting improved.
Version 2.6.0b Fixes and Deprecations
- IPEX-LLM support has been removed from version 2.6.0.
- Markdown formatting in the Answer section is now fixed, providing better formatting for chat answers.
- The Flux Schnell VAE was switched to an ungated version to reduce download errors.
Known Issues
- Some features may fail on the first run after downloading models and nodes, requiring a restart of AI Playground. If a model fails immediately after it's downloaded, restart AI Playground and try again.
- An interrupted download can block models from downloading properly. To fix this, go to the model location, delete the temporary file, and try again.
- The WAN 2.1 VACE inference preview may appear corrupted during steps, but the final output will be fine.
- The device selector in Image settings may default to the wrong device, requiring a manual setting after restarting AI Playground.
- Installing over a previous version of AI Playground without a clean install can cause performance issues, lag when typing in prompts, and errors.
- The Colorizer workflow is limited to Xe2-supported GPUs (LNL and BMG).
- The installer can fail due to environmental issues. See the Troubleshooting guide for help.
- A PC restart may be needed after installation for optimal functionality and performance.
- Reinstalling over an existing AI Playground installation will delete any ComfyUI models and custom nodes.
- Users can adjust settings beyond their specific system's ability to generate or complete a task.
Version 2.6.0b Supported Hardware
- Intel Core Ultra 200H Processors (ARL-H)
- Intel Core Ultra 200V Processors (LNL)
- Intel Core Ultra 100H Processors (MTL-H)
- Intel Arc B Series GPU Cards (BMG)
- Intel Arc A Series GPU Cards with 8GB+ Memory (ACM)
AI Playground v2.5.5 beta
This is a minor release which updates AI Playground 2.5 by adding an NPU device option for chat, improves the device selector for image generation, and corrects some issues from 2.5.0
Important
This version allows for NPU to be used in chat. Be sure to update your NPU drivers then restart your PC before using.
Version 2.5.5b Features
- All features from previous release (v2.5.0b) including support for Intel Core Ultra 200H (ARL-H) Processors
- Device selector added to Answer tab and device selector for image generation moved to Image Settings tab
- OpenVINO backend in Answer tab exposes AI Boost (NPU) as an option for chat, with support for document embedding
- Additional embedding model, intfloat/multilingual-e5-small, listed for IPEX-LLM backend
Version 2.5.5b fixes
- iGPU can now be enabled when installing on dGPU systems, per the improved device selector
- Video workflow files now properly installed without hitting the cloud icon
- Text to Video file on main branch corrected- removing the erroneous image input field
Known Issues
- Some features may fail on first run after installation of models and nodes, and may require a restart of AI Playground
- Device selector on Image settings may default to the wrong device, and may require manual setting after restart of AI Playground
- Continuous installing over AI Playground without a clean install can lead to performance issues, lagging type in prompts, and errors.
- Flux.1 Schnell VAE model now requires a HuggingFace user access token to authenticate download. Visit the model card on HuggingFace - login, copy your user token, paste into the user token field in AI Playground models tab to clear the alert
- Colorizer workflow is limited to Xe2 supported GPUs (LNL, and BMG)
- Installer can fail due to environment issues. See Troubleshooting
- PC restart may be needed after installation in order for optimal functionality and performance.
- Reinstalling over existing AI Playground will delete in ComfyUI models and custom nodes
- Users can adjust settings beyond their specific system's ability to generate or complete the task.
Version 2.5.5b Supported Hardware :
- Intel Core Ultra 200H Processors (ARL-H)
- Intel Core Ultra 200V Processors (LNL)
- Intel Core Ultra 100H Processors (MTL-H)
- Intel Arc B Series GPU Cards (BMG)
- Intel Arc A Series GPU Cards with 8GB+ Memory (ACM)
AI Playground v2.5.0 beta
This release updates AI Playground with support for PyTorch 2.7, with minor fixes and feature updates.
Important
This version when used with recent Intel graphics drivers can lead to device selection issues on systems with more than one GPU. If you have a discrete Intel Arc GPU, it is recommended to disable your iGPU, if you have issue installing and running this release of AI Playground.
Version 2.5b Features
- All features from previous release (v2.4.0b) including support for Intel Core Ultra 200H (ARL-H) Processors
- Upgraded support for PyTorch 2.7
- Upgraded support for Python 3.12.10
- OpenVINO set as the default backend for chat
- Backend Manager and Installer now allow you to reinstall components
- ComfyUI Installer now defaulted to v0.3.30 and allows user to set a preferred version for ComfyUI
- Support for LTX-Video 0.9.5 and 0.9.6 models and workflows
Version 2.5b fixes
- Performance regressions with 2.3a and 2.4b are improved with this version
Known Issues
- LTX-Video workflows may fail on first run, requiring the ComfyUI backend to be restarted after the installation of video models and nodes
- Older LTX-Video workflow has a package incompatibility for Python 3.12. Use new LTX-Video workflows tagged for v2.5
- Flux.1 Schnell VAE model now requires a HuggingFace user access token to authenticate download. Visit the model card on HuggingFace - login, copy your user token, paste into the user token field in AI Playground models tab to clear the alert
- Some environments may erroneously set the device setting to iGPU instead of dGPU. Disable iGPU as a workaround.
- Colorizer workflow is limited to Xe2 supported GPUs (LNL, and BMG)
- Installer can fail due to environment issues. See Troubleshooting
- PC restart may be needed after installation in order for optimal functionality and performance.
- Reinstalling over existing AI Playground will delete in ComfyUI models and custom nodes
- Users can adjust settings beyond their specific system's ability to generate or complete the task.
Version 2.5b Supported Hardware :
- Intel Core Ultra 200H Processors (ARL-H)
- Intel Core Ultra 200V Processors (LNL)
- Intel Core Ultra 100H Processors (MTL-H)
- Intel Arc B Series GPU Cards (BMG)
- Intel Arc A Series GPU Cards with 8GB+ Memory (ACM)
AI Playground v2.4.0 beta
This release updates AI Playground with fixes and features for chat, such as RAG embeddings for all backend options and DeepSeek models support for OpenVINO Gen AI.
Version 2.4.b Features
- All features from previous release (v2.3.0a) including support for Intel Core Ultra 200H (ARL-H) Processors
- Document search and summary support for both OpenVINO and Llama.cpp back ends (RAG support)
- DeepSeek R1 Distill model support for OpenVINO Gen AI back-end
- Reasoning mode chat output is now collapsible or expandable.
- Chat back-end selection, and other chat settings move to the horizontal tool-bar above the chat prompt, for easier access
- Generate Image Number field now editable to override the value limit beyond 4 images (properly implemented in this version)
Version 2.4b Fixes
- All user installation option removed, eliminating the administrative permission launch error
Known Issues
- Version 2.3a and 2.4b with the PyTorch 2.6 backend can result in lower image generation performance compared to version 2.2.1b
- Some systems may erroneously set the device setting to iGPU instead of dGPU. Disable iGPU as a workaround.
- Colorizer workflow is limited to Xe2 supported GPUs (LNL, and BMG)
- Installer can fail due to environment issues. See Troubleshooting
- PC restart may be needed after installation in order for optimal functionality and performance.
- Custom changes to ComfyUI can break ComfyUI for AI Playground
- Reinstalling over existing AI Playground will result in ComfyUI models and custom nodes being deleted
- Users can adjust settings beyond their specific system's ability to generate or complete the task.
Version 2.4b Supported Hardware :
- Intel Core Ultra 200H Processors (ARL-H)
- Intel Core Ultra 200V Processors (LNL)
- Intel Core Ultra 100H Processors (MTL-H)
- Intel Arc B Series GPU Cards (BMG)
- Intel Arc A Series GPU Cards with 8GB+ Memory (ACM)
AI Playground v2.3.0 alpha
This release is an alpha preview of AI Playground with support for Intel Core Ultra 200H Processors (ARL-H). This release is mostly functionally the same as version v2.2.1b, but replaces the Intel Extensions for PyTorch (IPEX) with native PyTorch 2.6, for compatibility with Intel Core Ultra 200H processors. As this is our first release without IPEX, we are releasing this as an alpha version.
Important
This alpha preview will run on all AI Playground supported skus. If performance or compatibility issues arise, notify us in our issues section and consider returning to version 2.2.1 (for supported systems).
Version 2.3a Features
- All features from AI Playground 2.2.1b
- Support for Intel Core Ultra 200H (ARL-H) Processors
- Image diffusion backend switch from IPEX 2.3 to PyTorch 2.6, eliminating IPEX libraries as a dependency
- Generate Image Number field now editable to override the value limit beyond 4 images
Version 2.3a Fixes
- UI/UX adjustments including removing image history dimming during generation.
- Non admin permissions enforced on AI Playground.exe
Known Issues
- Version 2.3a with the PyTorch 2.6 backend can result in performance issues compared to 2.2.1b
- Colorizer workflow is limited to B Series GPUs
- Installer can fail due environment issues. See Troubleshooting
- PC restart may be needed after installation in order for optimal functionality and performance.
- Custom changes to ComfyUI can break ComfyUI for AI Playground
- Reinstalling over existing AI Playground will result in ComfyUI models and custom nodes being deleted
- Users can adjust settings beyond their specific system's ability to generate or complete the task.
Version 2.3a Supported Hardware :
- Intel Core Ultra 200H Processors (ARL-H)
- Intel Core Ultra 200V Processors (LNL)
- Intel Core Ultra 100H Processors (MTL-H)
- Intel Arc B Series GPU Cards (BMG)
- Intel Arc A Series GPU Cards with 8GB+ Memory (ACM)
AI Playground v2.2.1 beta
This release includes needed fixes for v2.2, which introduced OpenVINO as an early preview as an high-performance backend for chat, v2.2 also includes additional ComfyUI worfkflows such for video and colorization. This release provides a single installer for all supported hardware.
Intel Core Ultra 200-H (ARL-H) not yet supported.
Important
Fixes In This v2.2.1 Release - Fixes a video generation issue from a failed installation of a needed component. Fixes image generation failing due to installation outside the C drive. Colorize works, but is limited to Xe2 processors (ie B580, B570, Intel Core Ultra 200v)
BE SURE TO BACK UP VALUED CONTENT FROM A PREVIOUS VERSION BEFORE INSTALLING
Version 2.2 Features
- All features from AI Playground 2.0.4b
- OpenVINO chat backend early preview, using OpenVINO Gen AI, providing the most optimized inferencing solution for Intel hardware. OpenVINO when installed can be used by selecting the Gear icon, then selecting it from the LLM Backend list in the Basic tab
NOTE this early preview does not yet have RAG implementation. This feature will be added in an upcoming release - Text To Video Workflow. Powered by LTX Video this workflow allows you to create video clips from text prompts
- Colorize Workflow. This workflow feature allows you to colorize black and white images
Limited to B Series GPUs and Intel Core Ultra 200v. Works best on people and faces - Max token slider added, allowing you to set Answer response tokens up to 4096 tokens
- UX improvements for Create tab providing full session history in the viewer with video player added for videos content
- UX improvements for Answer section adding model information and performance summaries in the responses
Version 2.2 Fixes
- Image output directory moved to User Directory folder, avoiding deletion of image output when updating or uninstalling AI Playground
- Fixed ComfyUI workflow installation errors, incomplete installs, and erroneous online dependencies
- Re-enable markdown files as RAG input
- Installation timeout increased to reduce installation failures due to time-out
Version 2.2 Known Issues
- Colorizer workflow is limited to B Series GPUs
- Installer can fail due environment issues.
See Troubleshooting section of the project Readme for installation troubleshooting - PC restart may be needed after installation of AI Playground, otherwise can result in poor performance or functionality issues
- Custom changes to ComfyUI can break the installation or repair of ComfyUI when installing AI Playground over an existing version.
Users with custom ComfyUI set ups should backup and remove the ComfUI folder before installing a new version of AI Playground - Users can adjust settings for image and video beyond their systems ability to produce or complete the task.
Check for performance monitor to see if any usage is nearing your total GPU memory. If so you should decrease resolution and or frames for better performance.
Version 2.2 Supported Hardware :
- Intel Core Ultra 200v Laptop Processors (LNL)
- Intel Core Ultra 100h Laptop Processors (MTL-H)
- Intel Arc B Series GPU Cards (BMG)
- Intel Arc A Series GPU Cards with 8GB+ Memory (ACM)
Coming Soon
- Intel Core Ultra 200h Laptop Processors (ARL-H)
AI Playground v2.2 beta
This release introduces OpenVINO as an early preview and optional high-performance backend for chat, along with bug fixes, and additional ComfyUI worfkflows. This release provides a single installer for all supported hardware.
Intel Core Ultra 200-H (ARL-H) not yet supported.
Important
Two bugs sighted - affecting Video and Colorize Generation - fix in process, new release to come. See issue
BE SURE TO BACK UP VALUED CONTENT FROM A PREVIOUS VERSION BEFORE INSTALLING
In this release
- All features from AI Playground 2.0.4b
- OpenVINO backend early preview, using OpenVINO Gen AI, providing the most optimized inferencing solution for Intel hardware. OpenVINO when installed can be used by selecting the Gear icon, then selecting it from the LLM Backend list in the Basic tab
NOTE this early preview does not yet have RAG implementation. This feature will be added in an upcoming release - Text To Video Workflow. Powered by LTX Video this workflow allows you to create video clips from text prompts
- Colorize Workflow. This workflow feature allows you to colorize black and white images - (works best on people and faces)
- Max token slider added, allowing you to set Answer response tokens up to 4096 tokens
- UX improvements for Create tab providing full session history in the viewer with video player added for videos content
- UX improvements for Answer section adding model information and performance summaries in the responses
Fixes
- Image output directory moved to User Directory folder, avoiding deletion of image output when updating or uninstalling AI Playground
- Fixed ComfyUI workflow installation errors, incomplete installs, and erroneous online dependencies
- Re-enable markdown files as RAG input
- Installation timeout increased to reduce installation failures due to time-out
Sighted Bugs Since Release
- Colorize Fails on Alchemist GPUs: Intel A series GPUs (ACM) get an 4GB memory limit issue: Issue is hijack fixes are missing in our ComfyUI backend implementation - a new release with fix is soon to publish.
- LTX Video cannot complete: "Backend could not generate image" error to a required node not installing on some systems. Fix being added to AI Playground. Workaround posted here: #206 (comment)
Known Issues
- Installer may fail due to time, requiring a restart to finish.
See Troubleshooting section of the project Readme for installation troubleshooting - PC restart may be needed after installation of AI Playground, otherwise can result in poor performance or functionality issues
- Custom changes to ComfyUI can break the installation or repair of ComfyUI when installing AI Playground over an existing version.
Users with custom ComfyUI set ups should backup and remove the ComfUI folder before installing a new version of AI Playground - Users can adjust settings for image and video beyond their systems ability to produce or complete the task.
Check for performance monitor to see if any usage is nearing your total GPU memory. If so you should decrease resolution and or frames for better performance.
Supported Hardware :
- Intel Core Ultra 200v Laptop Processors (LNL)
- Intel Core Ultra 100h Laptop Processors (MTL-H)
- Intel Arc B Series GPU Cards (BMG)
- Intel Arc A Series GPU Cards with 8GB+ Memory (ACM)
Coming Soon
- Intel Core Ultra 200h Laptop Processors (ARL-H)
AI Playground v2.0.4 beta
This is an updated release of AI Playground 2.0 alpha with updates and fixes. This release provides a single installer for all supported hardware including the latest Intel Arc B580 GPU.
BE SURE TO BACK UP VALUED CONTENT FROM A PREVIOUS VERSION BEFORE INSTALLING
In this release
- All features from AI Playground 2.0a
- CopyFace workflow added providing an image to image solution similar to FaceSwap
- DeepSeek models added to drop down model list in the Answer tab for IPEX-LLM mode
-- deepseek-ai/DeepSeek-R1-Distill-Qwen-7B
-- deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B - Added High VRAM warning to workflows that usually are best run on systems with more than 8GB of VRAM
- Updated language translations
- By default Create is set to Standard resolution with Fast mode on, to provide the fastest initial experience for image generation
- Higher resolution options added up to 0.5 megapixel for Default Mode, Standard Resolution ( ie SD 1.5)
Fixes
- Missing language translations for parts of the UI have been added.
- Negative prompt fixed can now be applied to Default mode where appropriate - does not work in Fast mode
- Hyperlinks from Model Settings fixed
- Safe Check toggle feature added back to Settings. This fixes an issue where image generation returns black images, being censored. Turning off this feature will bypass it
- Broken FaceSwap and CopyFace installation fixed, due to a move the custom node repo being moved
- VC++ redistributable check and installation added back to the installation
- Llama.cpp default model set, matching the IPEX-LLM experience
Known Issues
- When running and installing workflows like CopyFace for the first time, the installation process may take a long time and not fully execute. A restart of AI Playground may be needed to complete the installation.
- PC restart may be needed after installation of AI Playground, otherwise can result in poor performance or functionality issues
- Performance degradation can happen after running a few different high memory models. Restarting backends through Backend Installation Manager will restore performance
- Network firewalls and protected networks can interrupt or block the installation process. Install using an open network.
- Alternate Python and Git installations on the PC may conflict with installation.
- VC++ may be detected and not installed by the installer but may be required to be installed if out of date for the hardware.