# Changelog --- ## [Unreleased] ### ⚡ Improvements - **Installation**: The pack now ships a **`requirements.txt`**, so ComfyUI Manager installs the dependencies automatically and manual/portable installs have a one-line command (`python_embeded\python.exe -m pip install -r ...` for ComfyUI Portable — see the new "Installing dependencies" section in the README). Because this pack targets **NVIDIA RTX 3060+ GPUs only**, every feature dependency is mandatory: `soundfile`, `jieba`, `huggingface_hub`, `ffmpeg-python`, `opencv-python`, `imageio` (smart split), `nvidia-vfx` (RTX VSR upscale), `openai-whisper` and `qwen-asr` (subtitle recognition), and `voxcpm` (speech). No optional extras remain — previously these were installed on demand with error messages naming the missing package. The file also routes `nvidia-vfx` to NVIDIA's own index (`--extra-index-url https://pypi.nvidia.com/` + `--only-binary=nvidia-vfx`): the PyPI copy is a stub sdist whose `wheel_stub` build backend can abort the whole install (`Cannot import 'wheel_stub.buildapi'`), while NVIDIA's index serves real prebuilt wheels. - **Installation**: Removed the bundled `download_models.sh` downloader. All model files live in the Hugging Face version of the repository — cloning it downloads the complete model set, so there is no separate downloader step (the README's Models section now explains this). - **MultiTrack Editor**: The editor frontend was **reverted to the pristine upstream v1.3.3 source** — the MiniMax/Moxie format display-label mapping and every workaround built around it were removed. The shipped editor bundle is now built from that pristine source, so its behavior is exactly the battle-tested upstream editor (Shot/Context continuity, reference-size, audio lock, and MiniMax prompt placeholders all key on the native `MiniMax` format value again). - **Editor delivery fix**: A fresh browser could previously end up without the editor bundle entirely. ComfyUI core lets a pack's `WEB_DIRECTORY` overwrite its extension web-dir registration, so the browser's `/extensions` listing only ever contained this pack's helper scripts — never the editor entry — and the editor only appeared from stale browser cache. A small static loader (`web/js/editor-loader.js`, served with the helpers) now imports the editor entry from the pack's own route, which serves `index.js` with `no-cache` so the entry is always revalidated while its content-hashed chunks stay cacheable. After updating: restart ComfyUI and reload the page once; no cache-clearing gymnastics required. - **MultiTrack Editor / MultiTrack Project**: **T2V resolves natively again** (`t2v` stays `t2v`, matching upstream) — the model now handles text-to-video with the updated prompting, so the earlier T2V→R2V normalization was removed from the editor type computation, `h3_task_type`, `_multitrack_task_type`, and the generation-mode mapping. The reference-mode-with-zero-media cloud fallback was removed accordingly (reference mode raises for missing media again, as upstream). - **Moxie Multimedia Loader**: The loader now **looks like it loads a single model**. Only the diffusion-model picker (plus the attention backend) is visible; the LoRA, video VAE, audio VAE, and CLIP widgets were removed and their files are auto-selected from the pack's `models/loras`, `models/vae`, `models/audio_vae`, and `models/clip` folders (first file in sorted order, matching the bundled model set). The LoRA is applied automatically with strength 1.0; remove the file from `models/loras` to disable it. Missing auxiliary models raise the same clear setup error as before. Workflows saved before this change keep working — extra widget values are ignored. - **MultiTrack Project / Moxie Project Video Combine**: Single-pass projects now condition and sample at half the editor-selected resolution internally. The project timeline is composed at the segments' own size, and Moxie Project Video Combine restores the exact editor-selected resolution in pixel space before any output leaves the node (RTX Video Super Resolution via the optional `nvidia-vfx` package when available, with an automatic Lanczos fallback otherwise) — so every output carries the full editor resolution regardless of template wiring. The editor resolution therefore stays the final output resolution while the sampling chain runs smaller and faster. Dual-pass, SelfLift, and passthrough modes are unchanged. - **Moxie Project Video Combine**: Removed the leftover `VIDEO` and `FILENAME_PREFIX` outputs. The node now exposes exactly the components consumed downstream — `images`, `audio`, and `fps`, matching the core `GetVideoComponents` layout — and the output file name is chosen on `easy saveVideo`. The full-featured `easy multitrackProjectVideoCombine` node keeps its `VIDEO` and `FILENAME_PREFIX` outputs. Workflows saved before this change that wired the removed outputs need those links reconnected once; the shipped Moxie-consolidated template is already rewired. - **MultiTrack Editor / MultiTrack Project**: The `format` combo shows `MiniMax` again (the `Moxie Multimedia` display label was removed together with the rest of the mapping). Workflows saved while the label was shown need the format re-picked once; queued legacy label values are still normalized to `MiniMax` at execution time. The per-run `Output upscale` console line was removed earlier; the resolution restore now happens silently. - **Moxie Multimedia Loader**: The `lora_name` combo now defaults to the bundled turbo LoRA (`MM3step.safetensors`) so a fresh clone of the pack works out of the box, with a `none` entry to disable LoRA. Previously the default was a `""` placeholder, which silently skipped LoRA application even when a LoRA was bundled. - **Moxie Preview Override**: The `tiny_vae` combo now lists files from the node pack's own `models/vae_approx` folder first (shipping `MM-preview.safetensors` for MiniMax H3 previews as the default), falling back to ComfyUI's standard `models/vae_approx` directory. The folder is rescanned on every schema refresh, so newly dropped files appear without restarting ComfyUI. - **Moxie Multimedia Loader**: Locked the loader to the MiniMax H3 model set. The `clip_type` dropdown (which previously exposed every CLIP recipe in ComfyUI) was removed; the H3 Qwen3-VL text encoder is now always loaded with the `minimax` CLIP recipe. Any older ComfyUI build without `CLIPType.MINIMAX` raises a clear startup error instead of silently loading the wrong encoder. - **Moxie Multimedia Loader / Moxie MultiTrack Project**: Removed the legacy `FAST_MODEL_LOADER` (`model_loader`) bundle. The loader now outputs exactly `model`, `clip`, `video_vae`, and `audio_vae` (with the attention backend and LoRA already applied to the model), and the Moxie MultiTrack Project takes those four outputs as required direct inputs instead of a bundled loader. `Moxie Preview Override` likewise dropped its `model_loader` pass-through slots and now patches the `model` input directly. The generic `easy multitrackProject` node keeps its `model_loader` input for non-Moxie workflows. ## [1.3.3] - 2026-09-25 ### ✨ New Features - **Multi-Image Loader**: Added the `easy multiImagesLoader` node, supporting loading up to 25 images. ### 🐛 Bug Fixes - **MultiTrack Project**: Fixed TE being executed an extra time in `Dual` mode. - **MultiTrack Editor**: When a single segment covers the entire task window, crop and resize it directly instead of recomputing through the reference path, fixing [#84](https://github.com/yolain/ComfyUI-Easy-Media/issues/84). The media loading timeout in this path is set to 60 seconds. - **MultiTrack Editor**: [Motion transfer & subject replacement] When the source video frame rate differs from the editor frame rate, smart-split segments now split using the editor frame rate, and when paused or stepping frame-by-frame the requested frame is returned via ffmpeg's `fps` filter as the preview image. - **MultiTrack Project**: [Motion transfer & subject replacement] Locked videos now use uniform temporal resampling. Previously, when the locked duration did not conform to the `17k+5` frame pattern, the value was rounded up and padded to fit the current frame count, which caused some frames to freeze early. Resampling now uses a uniform temporal distribution so added frames are spread evenly across the timeline rather than clustered at the tail, eliminating the freeze. - **Audio Lock**: Fixed redundant repeated encoding/loading when locking audio across multiple segments of long videos. Due to the multitrack project design, all audio locks are processed before the reference video generation node, so when video is present, ffmpeg is used to extract and cache the audio. Videos without audio and muted tracks no longer lock audio or repeat audio probing, while their video timeline timing remains unchanged. The `audio` input on `easy minimaxH3AudioLock` is now optional; leaving it disconnected passes the latent through unchanged instead of erroring. ## [1.3.2] - 2026-09-20 ### ⚡ Improvements - **MultiTrack Project**: Optimized project media, segment media, TaskOutput, and per-segment H3 conditioning caching to avoid media reloads and repeated first-pass conditioning encodes when re-executing without parameter changes. - **H3 Conditioning Cache**: Store up to five recently used segment caches in ComfyUI's temporary directory, reset the pool when its project, model, or upstream media cache scope changes, rebuild zero-filled initial AV latents from compact metadata, and report a per-category tensor size breakdown after each write. - **MultiTrack Project**: Enhanced action continuity in context mode and optimized lip-sync when audio is locked. - **MultiTrack Editor**: Added the ability to freely drag and adjust the width ratio between the image item area and the prompt area within segments. Optimized the UI display of user prompt highlight tags. - **MultiTrack Project**: Added `selflift` option to `sampling_mode`, adapted from [comfyui-SelfLift](https://github.com/facok/comfyui-SelfLift) - **Character Swap Context**: Switched to `drive_control` approach instead of the previous noise-based method. Testing shows `drive_control` is more stable and preserves more motion details. Although it may appear slightly less sharp visually than the noise approach, it is better suited for preserving source motion during character replacement and motion transfer. ### 🐛 Bug Fixes - **Context**: Fixed context loading latent preferring the previous segment's selected latent instead of the most recently generated latent by timestamp. - **Video Reference**: Fixed video reference frame alignment. Previous version had frame loss in motion transfer scenarios when timeline was locked, causing reference to fall back to image. - **MultiTrack Editor**: When segment A has a gap with segment B, adjusting segment A's duration should first fill the gap duration before propagating changes to subsequent segments. - **MultiTrack Editor**: Fixed some UI interaction compatibility issues with Node2.0, improving the overall operational experience. ## [1.3.1] - 2026-09-07 ### ✨ New Features - **MultiTrack Prompt Enhance to Project**: Added `easy multiTrackPromptEnhanceToProject` node, which supports directly applying the MultiTrack Prompt Enhancer's output to the MultiTrack Project. - **Character Swap Context**: Added the `context_swap` continuity mode for MiniMax H3 task segments. It preserves motion continuity while weakening the previous character's appearance with disposable tapered latent noise. - **Dual-Pass Swap Continuity**: Apply swap noise independently to the low-resolution first-pass context and the high-resolution second-pass context. Audio remains untouched, and only clean trimmed/re-encoded context latents are saved for later segments. - **Video-Track Audio Lock**: Lock a video track segment as the timeline constraint for MiniMax H3 generation, ensuring the output duration matches the video; once locked, the audio is synchronized and locked as well. To replace the audio lock, you can lock another audio track as the new audio source while keeping the video track unchanged for the visual. (Mainly used for motion transfer and character replacement.) - **Minimax Prompt Override**: Added the `easy minimaxPromptOverride` node, which supports assembling prompts and other parameters into the multitrack editor's prompt_override, suitable for explicit passing in. ### ⚡ Improvements - **Character Swap Context (drive_control)**: Switched from the previous noise-based approach to `drive_control`. Testing shows drive_control is more stable and preserves more motion details. Although it may appear slightly less sharp visually than the noise approach, it is better suited for preserving source motion in character replacement and motion transfer scenarios. - **MultiTrack Project Media Preparation**: Consolidated task-scoped image, audio, video, shared-media, and locked-audio preparation so project expansion loads and forwards only the media required by each segment. - Clear Redundant Subgraph Cache: Fixed an issue where execution_cache in multi-track project subgraphs was unnecessarily storing video objects, causing additional memory accumulation. ### 🐛 Bug Fixes - **Responsive Width Preservation**: Fixed an issue where all custom easymedia components would shrink in width after clicking the ComfyUI properties panel. - **Frame-Accurate Video Trimming**: Normalize source frame rates before trimming merged video inputs to keep segment boundaries aligned with the project timeline. The tapered context-noise approach used by Character Swap Context is adapted from [MacroSony/minimax-h3-chained-character-swap](https://github.com/MacroSony/minimax-h3-chained-character-swap). ## [1.3.0] - 2026-09-03 ### ✨ New Features - **MultiTrack Project**: Currently adapted for MiniMax, mainly used to simplify the pipeline. It handles multi-track loop segment encoding, sampling, and decoding/saving, supports specifying segment start position and maximum generation count. - **MultiTrack Project Video Combine**: Preview uncombined segment outputs, supports automatic and manual combining. ### ⚡ MiniMax H3 Optimizations - **Motion Context & Resampling**: Optimized the pairing of the original `MotionContext` with resampling. In the current default sampling configuration: `medium` produces smoother results, while `light` cannot sample sufficiently in the first sampling due to sigma separation. - **Motion Context Latent Memory**: Keep only the 22-frame high-resolution audio/video context tail on CPU between segments and in completed project artifacts, while retaining the complete low-resolution first-pass checkpoint for deferred second-pass runs. - **Multi-Track Editor (Audio Track)**: Added `Lock Audio` feature, only applicable under MultiTrack Project. When generating video, audio will no longer regenerate, but instead schedules video frame output based on this audio. - **Multi-Track Editor (Audio Track)**: Added `Reuse Audio` - all segments can reuse this audio (previously required cloning to each segment), supports up to 15 seconds without being truncated by segments. - **Multi-Track Editor (Shared Media)**: Images, audio clips, and video clips can be marked as shared references when context segments need to reuse the same media. Shared images are placed before each task's local images; shared audio uses up to the first 15 seconds of its source. - **Multi-Track Editor (Task Track)**: When multi-selecting task segments, you can uniformly adjust task mode, transition mode, and reference image size. - **Multi-Track Editor (Task Track)**: New feature added. - **Multi-Track Editor**: Adjusted many texts for better understanding. - **Multi-Track Task Output**: Adjusted minimum value of task_index to -1. When -1, it can output all track complete media information like the previous Multi-Track Editor. (Due to the previous optimization, the Multi-Track Editor cannot directly output complete content) - **Multi-Track Editor**: Defer file and URL media loading when no slot resources are referenced; task output nodes now materialize only the media they need, reduce unnecessary media resource overhead in long-form video tasks. (Now you just need to connect `TRACKS_INFO`; the `image`, `video`, and `audio` output by the multi-track editor will all be `None` in Slot-less resource mode.) ### 🐛 Bug Fixes - **Multi-Track Slot Media**: Added video slot selection and fixed image, audio, and video slot discovery and downstream delivery through `TRACKS_INFO`, including sparse lists that skip empty inputs and graphs whose link ID is `0`. ## [1.2.1] - 2026-08-22 ### ✨ New Features - **Multi-Track Editor**: Added right-click context menu to clone audio segments and video segments ### 🐛 Bug Fixes - **Multi-Track Task Output**: Fixed that segments without audio/video in `MiniMax` format should output as None ## [1.2.0] - 2026-08-21 ### ✨ New Features - **Compare Video**: Added `Side-by-side Compare` mode, support for directly selecting media for comparison, and `Watch Output History` mode - **Media Selector**: Added support for multi-selecting image items and selecting all images in current path - **Multi-Track Prompt Enhancer**: Added this node, supports prompt enhancement for models like h3-context-ir, supports local model usage - **Multi-Track Prompt Enhancer to Project**: Added `easy multiTrackPromptEnhanceToProject` node, supports directly applying the output of the multi-track prompt enhancer to the multi-track project - **Multi-Track Editor**: Added user prompt A/B output, supports selecting different user prompt outputs - **Multi-Track Editor**: Added user prompt reference functionality, supports referencing resources via ``, `