A.I. Image & Video Studio 1.14.3

Models now shows VRAM guidance on every model card, with All models, 8 GB VRAM and 16 GB VRAM filters. The 16 GB view also includes lighter models. Download selections are preserved when changing filters.

8 GB candidates include the SDXL models, FP8 FLUX.2 Klein and MuseTalk. Wan 1.3B is conditional on Low VRAM mode and small, short clips. Requirements are estimates for the packaged workflows; no Studio inference benchmark on an 8 GB card has been run. Models with an unknown fit remain under All models with explicit guidance.

Validation: all 287 Python regressions passed, plus real-browser filter and selection-preservation checks. Installer payload verified against source. No model downloads or generation were performed.

Update through Settings, or download AIImageVideoStudio-Setup-1.14.3.exe below. Existing settings, models, LoRAs and media are preserved by the installer.

A.I. Image & Video Studio 1.14.2

Adds two optional video models to Models and Video:

• MiniMax H3 · Singularity V2 INT8 — reference-to-video with up to nine image references, optional generated audio and the recommended four-step Turbo adapter. Uses the pruned INT8 checkpoint and reuses installed MiniMax encoders and VAEs. VSA is disabled for this fine-tune.
• WAMU · Wan 2.2 I2V Lightning 14B — starting-image generation with separate high/low-noise models, a four-step default, and silent output. Loads the original Diffusers shards directly and reuses the Wan encoder and Wan 2.1 VAE. Its model weights total about 57 GB.

Downloads are pinned and checksum-verified. Both models preserve queue settings and support compatible extra LoRAs. Model weights download separately and are not included in the installer or source archive.

Validation: 287 regression tests and UI checks passed; installer contents and application-only update preservation were verified. No video was generated during validation. The native compatibility probe stalled importing the installed PyTorch runtime, so actual generation and 16 GB GPU operation remain unverified. WAMU requires substantial system RAM, paging and disk space, and its publisher has not specified a merge license.

Windows: download AIImageVideoStudio-Setup-1.14.2.exe. Existing settings, models and Library media are preserved. The installer supports choosing the installation folder and optional desktop/Start-menu shortcuts.

A.I. Image & Video Studio 1.14.0

Reference-image generation and source-import fixes.

• Images: reference controls appear for FLUX.2 Klein (up to 3 images), FLUX 3 Klein (10) and Viggle (3). Choose Library or local files; remove or clear references. Each model keeps its own selection.
• Video: MiniMax H3 Ref2VA supports its primary reference plus 8 additional character references. These guide appearance rather than setting a first frame. Unsupported models hide extra-reference controls.
• References retain their order in queue jobs, restore when editing/reusing settings, and are recorded in output metadata. Duplicate, missing and excess references are rejected.
• Video source imports remain available during generation, including replacing an existing image. Imports use an independent CPU request, recover after errors and leave generation progress/results untouched.
• Updates preserve settings, models, LoRAs, Library media and Canvas projects. No model weights are bundled.

Validation: 268 Python regression tests passed, plus processing-disabled interface checks for conditional controls, imports, removal, restoration and queue payloads. Packaged assets and the installed app were verified. No AI image or video generation was run for validation.

Download the Windows Setup installer below. Source code and SHA-256 checksums are also attached. Third-party software and model licenses remain separate.

A.I. Image & Video Studio 1.13.1

Canvas and MOSS update, including the Canvas screen-fit fix.

• Canvas tab beside Image Editor: drag local or Library images onto an artboard; arrange image/text layers; resize, rotate, change opacity, reorder, lock, undo and zoom. Save editable projects or export to Library, Image Editor or Video.
• Canvas toolbar and bottom actions fit their contents. Side panels scroll independently, with page scrolling for short windows.
• MOSS-Image Base, SFT and Kromoss: compact INT8 and BF16 options, captions/tags/layouts, optional color/depth maps and compatible fast adapters. Canvas layouts transfer directly into MOSS controls.
• Video faceswap, optional SageAttention acceleration, improved memory handling, resource indicators, Library previews/metadata, completion sounds and generation/input fixes accumulated since v1.9.0.
• App updates preserve settings, downloaded models, LoRAs, Library media and Canvas projects. Model weights remain optional downloads.

Validation: 263 Python regression tests passed. Canvas/MOSS interface checks used processing-disabled fixtures; no AI generation was run for this release. The packaged and installed app assets were verified.

Download the Windows Setup installer below. Source code and SHA-256 checksums are also attached. Third-party software and model licenses remain separate.

A.I. Image & Video Studio 1.9.0

This update adds Talking Avatar, Faceswap, and Voice tools, plus new image and video model packs.

• Talking Avatar supports EchoMimicV3-Flash, SoulX-FlashHead, LongCat Avatar, InfiniteTalk, and MuseTalk, using imported recordings. Choose a Library image or a local file, and clear the selection with X.
• Video adds MiniMax H3 Turbo text-to-video and image-to-video, optional first/last frame guidance, and Ref2VA with optional audio and VSA controls.
• Images and Image Editor add Qwen-Image 2.1 Viggle Turbo six-step workflows. The new Faceswap tab uses the BFS v1.1 adapter with separate target and identity images.
• Voice connects to a separately running local Fish Speech or SGLang Drama3 server. Generated recordings can be sent to Talking Avatar. Voice server installation and model weights are not bundled.
• Library stores images and videos in separate folders under Library, offers File Explorer access, and opens images fullscreen with mouse-wheel zoom.
• New tools use the editable generation queue with priority and cancellation. Compatible models have their own LoRA scope.
• The desktop window stays open and reconnects during backend interruptions instead of silently closing.

Install with AIImageVideoStudio-Setup-1.9.0.exe, or use Settings > Check for updates in an existing installation. App updates preserve settings, downloaded models, LoRAs, recordings, and Library media. Optional model packs download separately from Models; their storage and hardware requirements vary.

The Studio source remains MIT licensed. Bundled components and downloaded models retain their own licenses; Viggle/Qwen components include research/non-commercial terms. MiniMax's local quantized workflow can produce different results from the linked hosted Space.

Validation: 178 Python tests passed, UI-script checks passed, and 16 MiniMax/Qwen workflow graphs passed engine schema validation. No AI image, video, or voice inference was run during these checks. GPU performance and output quality of the new workflows still require testing.

Downloads include the Windows installer, source archive, and SHA-256 checksums.

A.I. Image & Video Studio 1.6.0

Settings now includes Check for updates. New stable releases from this repository show a confirmation with version, size and release notes. On approval, the app verifies the installer's SHA-256 checksum, closes, replaces its application files and reopens. Settings, downloaded models, LoRAs, media and queued jobs are kept. Installation waits until processing is idle; the queue reopens paused. Program Files can require Windows administrator approval.

Also includes:

• An About block at the bottom right of Settings with the requested Codex / Jungle Jim credit and all five resource/source links.
• The version number directly below the logo in the app and installer.
• Models between Library and LoRAs.
• Bounded retries for transient Windows JSON file locks and transactional replacement with rollback of previous binaries on copy failure.

Validation: 123 Python regressions, Node UI regressions, isolated browser checks and a packaged installer handoff/restart check with test-only data. No image or video generation was performed. The unsigned installer retains selectable app/data folders and optional shortcuts. Model downloads and their licenses remain separate from the MIT application source.

A.I. Image & Video Studio 1.5.2

Fixes image enhancement failing on Windows when nested LoRA filenames used different separators. The engine now receives its exact advertised filename; detail strengths up to 2 remain supported.

• Upscale now sits between Video and Library.
• New seed for each job is enabled by default, with a fixed-seed option. Queued jobs, retries and saved results retain their exact seeds.
• Library previews show model, seed, prompts, generation settings, LoRAs, enhancement/upscale status and original source history in a compact layout.
• Continue after failed jobs is enabled by default. Failed items keep their error while remaining jobs run; disable it to pause on failure. Existing paused queues require Resume queue.

Validation: 106 Python regressions, Node UI checks, isolated browser checks and CPU-only native graph validation. No AI image/video generation was performed.

The unsigned Windows installer retains selectable app/data folders and desktop/Start menu shortcuts. App-only updates preserve models, LoRAs, settings, media and queue data. Model downloads and third-party licenses remain separate. SHA256SUMS.txt covers the installer and source archive.

A.I. Image & Video Studio 1.5.1

Correct the FLUX 3 installation instructions to say Enter a read token in Settings. The help text, access error and README now point to the existing token field in Settings. Return to Models to install the selected model. No generation or model changes.

JavaScript syntax and focused text/location checks passed. Windows installer and source ZIP are attached with SHA-256 checksums. Existing media, settings, adapters and queue are preserved by app-only updates.

A.I. Image & Video Studio 1.5.0

Add FLUX 3 Image Klein distilled INT8 to Images and Image Editor.

• Official separate model pack with fixed CFG 1, 16 starting steps and up to ten reference images in the editor. Masked instruction edits preserve the resized original outside the painted area.
• Existing queue, library, video handoff and optional image enhancement work with the new model. FLUX 3 has its own LoRA library and compatibility guidance.
• A pinned compatible ComfyUI engine and comfy-aimdo dependency install into separate folders. Other models retain the original engine and packages. The different FLUX 3 VAE uses a unique filename.
• Models links directly to the official Hugging Face access page and explains entering a read token. The gated model pack is about 11 GB; its separate engine adds about 14 MB. Tokens are not saved. The Base checkpoint and optional prompt upsampler are excluded.
• Fix a deliberate desktop/backend exit incorrectly opening a blocking startup-error dialog.

Validation: all 97 automated regressions passed. Four FLUX 3 graphs passed native typed validation without loading weights or initializing CUDA. The isolated engine install/CPU probe passed. App and installer builds passed; the local app-only upgrade preserved all 38 protected settings/queue/adapter/media hashes. Visual UI automation was unavailable because of a computer-use kernel error; no rendered UI pass is claimed.

No image/video generation test was run. Model weights require approved Hugging Face access and were not downloaded on the validation machine. Actual FLUX 3 quality, adapter compatibility, speed and GPU memory fit remain untested. Studio output sizes remain 256–1536 pixels; a graphical bounding-box editor is not included.

The app's original source is MIT. FLUX 3 weights use the FLUX Kommunity License; downloaded engine/dependencies retain their own licenses. The unsigned Windows x64 installer retains selectable application/data folders and optional desktop/Start menu shortcuts.

A.I. Image & Video Studio 1.4.0

Add mixed image/video scheduling, model removal and optional enhancement passes.

• New Queue tab: add jobs while another runs, edit waiting settings, cancel, pause/resume, change priorities and reorder equal priorities. Seeds, LoRAs, sources, masks and output folders are captured per job. Unfinished jobs reopen paused; errors pause the queue for review.
• Models can preview and uninstall their downloaded weights. Shared dependencies, LoRAs, runtime software, settings and library media remain.
• New Upscale tab and library handoff. Optional automatic enhancement after generation saves a separate copy and preserves the original.
• Optional RTX VSR integration uses tritant's pinned pipeline in an isolated worker with NVIDIA VFX SDK bindings. Install it from Models (about 467 MB). RTX hardware, driver and encoder support are required.
• Optional Elusarca Qwen 2.1 Detail Enhancer pack (about 17.4 GB). This is a Qwen 2.1 LoRA with instruction-conditioned enhancement; it can change fine details. Qwen 2.1 weights have the Qwen Research License, separate from this app's MIT license.
• Existing logo, compact sci-fi desktop layout, inline help, library and model-specific LoRA profiles are retained.

Validation: 86 automated tests passed, including serial mixed queues, cancellation, adapter isolation, SDK installation failure handling and file preservation. Seventeen workflows passed native ComfyUI typed-graph validation without loading weights. New tabs were checked at 900x640 and 1280x720 without page overflow or browser console errors.

No AI generation or GPU enhancement test was run for this release. Actual new-enhancer quality, GPU/driver compatibility and memory fit have not been established. No model weights are bundled, and the update does not automatically install the optional enhancers.

Windows x64 installer and source ZIP are attached with SHA-256 checksums. The installer is unsigned; it lets you choose application/data folders and optional desktop/Start menu shortcuts. Finish current work and close the studio before upgrading. Existing settings and library files are preserved by app-only updates.