Mix-Studio

Highlights

• Updated MiniMax H3 Frames Turbo to the creator-recommended v4 adapter with a six-step default, exact adapter metadata, and compact four-step quality guidance.
• Added experimental H3 Long Context generation up to 120 seconds, including sequential clip progress, audio/video bridge continuity, Turbo support, and final MP4 joining.
• Added Match video aspect for H3 references, reference aspect badges, and clearer icon-only compatibility cautions.
• Improved Wan Animate 2 reference and performance sizing and added stage-aware Revise Prompt progress.

Mobile responsiveness

• Kept Library navigation immediately interactive by reusing the mounted grid and replacing passive full-grid refreshes with lightweight revision checks.
• Replaced full-resolution animated gallery cards with cached, silent 480p H.264 preview clips capped at 12 fps.
• Limited touch-first devices to one centered preview decoder and serialized first-time proxy creation to prevent decoder and transcode spikes.
• Made video cards open on the first mobile tap, handed playback cleanly to the full-quality expanded viewer, and paused gallery decoding behind menus and sheets.
• Fixed mobile profile-menu actions, including Edit profile and Switch profile, by handling completed touch gestures without the follow-up synthetic click closing or intercepting the destination.

Updating

Open the Mix Studio side menu and choose Update app. Server changes restart Mix Studio automatically after both Mix Studio and ComfyUI queues are idle. Your settings, profiles, gallery, and generated media are preserved.

See the full changelog for every v1.2.1–v1.2.4 change.

Highlights

• Added MiniMax H3 Turbo generation with its recommended custom sampler, audio-safe workflow, configurable steps, and optional LoRA strength under Advanced settings.
• Added the official MiniMax H3 prompt guide, local no-LLM structure and dialogue formatting, and guide-aware Prompt Enhance while keeping all formatting optional at generation time.
• Made H3 Revise Prompt resilient to long ComfyUI waits with live queued/running status, explicit cancellation, bounded retries, orphan cleanup, and advisory guide warnings instead of speaker-format dead ends.
• Added SeedVR2 as a temporally coherent video post-upscale option alongside fast RTX Video Super Resolution.
• Added a centered animated update showcase with MiniMax H3 and responsive Mix Packs promos, full changelog and install actions, a desktop top-bar Update button, and a mobile profile notification and Update action.

Additional improvements

• Added critical web-asset integrity checks and startup recovery so partial desktop deployments cannot leave the app unstyled.
• Added one shared external prompt-AI provider for OpenAI, Gemini, or Ollama, with independent Image and Video switches.
• Added social-ready documentation-video aspect ratios and H.264 MP4 export with AAC audio when supported.
• Improved documentation-video prompt readability and removed duplicate MiniMax H3 reference labeling.
• Added compatibility-critical custom-node revision checks and repair support in Generation Setup.

MiniMax H3 and workflow updates

• Added MiniMax H3 text-to-video, image-to-video, first/last-frame, and native-audio generation through ComfyUI's day-zero native nodes.
• Added H3 Reference mode for prompt-addressable image, video, and audio inputs. Its separate Ref2VA model installs only when Reference mode is selected, so standard H3 users do not download it.
• Kept LoRA sliders and hold-drag strength controls within the practical 0–2 range while retaining exact manual entry from -100 to 100.
• Made Mix Packs available from Create, Edit, and video prompt fields, including preset-card persistence, gallery metadata, and image or video documentation exports.
• Fixed desktop Image Guide and Depth Guide actions so they open the complete Create Image workspace instead of entering an invalid partial view.
• Centered the Camera Motion, Face ID, video-model, and Director story-beat dialogs with balanced viewport margins and bounded scrolling for tall content.

MiniMax H3 requires ComfyUI 0.30.0 or newer. The curated H3 installer currently targets NVIDIA hardware.

Generation and workflows

• Added signed LoRA strengths from -100 to 100 across Image, Region, Edit, video, and Director workflows. Exact decimal entry is supported, and workflow graphs preserve the signed value.
• Updated Strength Hunt to step from zero toward a negative current LoRA strength. Hunts remain bounded to a magnitude of 2 so an extreme manual value cannot create an unexpectedly large batch.
• Updated all LTX 2.3 empty-audio graphs to submit the current frames_number, frame_rate, and batch_size inputs required by LTXVEmptyLatentAudio, including standard LTX, Face ID, and 10Eros routes.
• Added configurable Krea 2 Identity Edit sampling presets within the supported step and CFG range. App-managed workflow LoRAs no longer trigger misleading compatibility warnings.
• Preserved the editor prompt separately from the expanded graph prompt so gallery reuse restores the editable text without losing workflow metadata.
• Added Mix Pack preset identity to generation details and documentation exports while retaining the complete submitted prompt.

Model installation and setup

• Added resumable HTTP model downloads and isolated Hugging Face Xet acceleration, including byte progress reporting and automatic fallback when accelerated transfer fails.
• Added an optional trusted HTTPS Hugging Face-compatible download endpoint for networks that cannot reach huggingface.co. Hugging Face access tokens are never sent to a custom endpoint.
• Made the live ComfyUI model registry authoritative for compatible Krea 2 text encoders, including qwen3vl_4b_fp8_scaled.safetensors, so existing external models are reused and their exact registered paths persist.
• Expanded model discovery across configured model roots, manual subfolders, and ComfyUI extra model paths. Existing files are reused instead of downloaded again.
• Added support for ComfyUI Desktop adopted base directories when installing SAM3 custom nodes.
• Improved portable ComfyUI dependency installation with a uv fallback for missing or broken pip environments. Incompatible custom-node packages are isolated so one failure does not block unrelated selected workflows.
• Corrected the Qwen and SAM model download sources used by Generation setup.
• Prevented Generation setup from launching the ComfyUI Desktop installer when Mix Studio is already connected to a portable instance but still needs its local folder selected.

Mix Packs and prompt presets

• Replaced the legacy camera control wheels with a visual, searchable Mix Pack preset browser.
• Added a dedicated pack landing grid and detail pages with pack-scoped search, continuous category sections, scroll-aware category navigation, and smooth navigation to selected sections.
• Added responsive three-column desktop and two-column mobile layouts, full-thumbnail presentation, category rail scroll controls, and clearer selected and missing-thumbnail states.
• Added independent multi-preset selection, including multiple presets from the same category, with removable thumbnail cards in Applied Looks.
• Added immediate toggle-on and toggle-off behavior and color-coded preset phrases in the prompt composer.
• Restored preset cards from saved generation metadata and from exact prompt text when transient UI state or older saved generations do not contain card state.
• Increased the Mix Pack limit to 200 presets and expanded the validated thumbnail budget.
• Added atomic upgrades for newer pack versions, permanent deletion when removing a pack, and reviewed pack installation and management under Advanced Settings.
• Added profile preferences for prompt composition and whether applied preset cards are shown.

Mobile and remote access

• Made Mix Studio installable as a Progressive Web App with Mix Studio home-screen icons, maskable artwork, a service worker, and an offline fallback page.
• Added private HTTPS phone access through an existing or newly configured Tailscale Serve route. Existing Serve configuration