ChangeLog - Zuntan03/EasyWan22 GitHub Wiki

2025/08/22

• Fixed unintentionally changing the ColorMatch model.

2025/08/21

• Updated ComfyUI version to v0.3.51.
• Updated city96/ComfyUI-GGUF to the 8/20 version.
• Temporarily fixed the issue where SaveEndFrame was saving the start image instead.
  • The state becomes incorrect when toggling Enable SaveEndFrame.
• Added presets for AssertiveCowgirl, PovCowgirl, and OralInsertion.

2025/08/20

• Added support for past video generation from end images and video generation from start and end images.
  • SwapStartEnd treats the start image as an end image and generates a past video leading to the image.
  • UseEndImage additionally specifies an end image to generate a video from start image to end image.
    • Resource consumption is high due to brute force approach.
  • This is not VACE, so motion continuity depends on luck.
  • Added LoopFLF2V sample that generates a 15-second loop video from a single image.

LoopFLF2V sample generation procedure:

1. Enable SwapStartEnd to generate a 5-second past video from the end image.
2. Disable SwapStartEnd to generate a 5-second video from the start image.
   Enable AppendVideo and set Asset.webp from step 1 to PreviousWebpVideo to check motion continuity.
   • This becomes a simple 10-second video generation at this point.
   • AppendVideo is one of the post-processes, so Enable PostProcess must be enabled.
     • Settings like Refiner and Upscale should be matched between the videos to be combined.
     • The next update is expected to allow adjustment of the AppendVideo combination timing.
3. Enable UseEndImage, set the start image from step 2 and the end image from step 1's EndFrame.webp, then generate a 5-second video.
   Enable AppendVideo and set Asset.webp from step 2 to PreviousWebpVideo.
   This creates a looping video: step 1 > step 2 > step 3 > step 1.

2025/08/18

• Added a sample of the CumshotAesthetics preset assisted by prompts.
  • Japanese instructions are automatically converted to three-language prompts (Japanese, English, and Chinese).

2025/08/17

• Made it possible to toggle the entire post-processing on/off with Enable PostProcess.
• Made it possible to edit mosaics with Enable MosaicWork even when the entire post-processing is disabled.
  Here's an example procedure for editing mosaics at maximum speed:
  1. Disable Enable PostProcess.
  2. Load a generated Webp video with LoadWebpVideo in PostProcessSource.
  3. Enable Enable MosaicWork.
  4. Disable file saving like SaveMp4 in Output.
  5. Enable Enable PointMosaic, Enable MaskMosaic, Enable AutoMosaic to edit mosaics.
  6. When mosaic editing is finished, enable Enable PostProcess and file saving, then execute to generate a mosaic-processed video with post-processing.
• Made it possible to append generated videos to existing Webp videos with AppendVideo.
  • You can remove trailing frames from the previous generated Webp video with TrimEndFrames.
    • If you want to remove leading frames from the generated video, use TrimStartFrames.
• Added CumshotAesthetics preset and sample.

2025/08/16

• Released Wan22-FastMix and adjusted workflows accordingly.
  • Achieved ghost reduction and motion enhancement.
  • While there's compatibility with used LoRAs, you can increase or decrease motion size and speed with EnhanceMotion.
• Rolled back PyTorch version to 2.7.1 as some environments seem to have compilation issues or slower performance.

2025/08/14

• Updated ComfyUI version to v0.3.50.
• Upgraded PyTorch version to 2.8.0 and CUDA version to 12.8.
• Added samples demonstrating that GeneralNsfw presets can be used for various images.

2025/08/13

• Added a sample to verify the versatility of the general NSFW LoRA and the mass production capability of the I2V workflow.
  • Just loading past favorite images with folder specification and generating with presets as-is.
• Added Use Sigmas to enable disabling Sigmas.
• Added NSFW presets.
  • Download\loras\Nsfw\Footjob_v10-(H|L).bat

2025/08/12

• Fine-tuned the Refiner settings.
• Added support for notifications when generation is complete.
  • Can be disabled with the SystemNotification node.
• Changed the default TextEncoder from Native to Kijai.
  • While accuracy is reduced due to fp8 not being scaled, it has the benefits of avoiding the issue where GPU is sometimes not used and supporting disk cache.
• Asset Webp saving is no longer treated as a target for frame interpolation.
• Updated 'User Feedback'.
• Added 4 types of samples 20250812-GeneralNsfw_A~D to Sample.
  • These are simply generated using GeneralNsfw preset with random loading of past images thrown into folders via LoadImageFrom.
  • The highlights are the revival of kanji in speech bubbles through V2V Refiner and fluid simulation.

2025/08/11

• Changed to use GGUF with fast LoRA baked in for acceleration.
  • You can download higher accuracy models with Download/diffusion_models/FastMix/*.bat.
• Made it possible to use upscalers for enlargement in Refiner.
  • Quality improves when using modest rewrites like 2step.
• Adjusted parameters overall.
• Added RepeatFade that fades with the beginning at the end of the video.
  • Crops several frames from the beginning of the video and cross-fades them at the end.

2025/08/10

• Added functionality to generate videos using image files from the Input/Load/ folder randomly or sequentially.
  • Switch with LoadImageFrom below the start image loading.
  • You can also specify folders from different paths with LoadImageFolderPath.
  • Set random or sequential with randomize | increment & Set 0
• Added LoRA Presets. There were LoRA connection issues as of yesterday.
  • Increased presets to 5 types.
• Fixed a bug where ColorMatch in Refiner stopped working.
• Made it possible to choose between area-based and long-side specification for video size.

2025/08/09

• Currently revamping I2V workflow, but documentation updates haven't caught up.
  • Changed to specify video size by the side length of an area-based square.
  • Added support for Color Match. Available right after the first Trim in post-processing and after enlargement in Refiner.
  • Added support for Color Correct.
  • Made I2V-A14B LoRA easily accessible through LoRA Presets.
• Moved old workflows to Easy/Deprecated/.
• Added "Impact of Large AI Tool Models on SSD Lifespan" information.

2025/08/08

• Temporarily added 00_I2V_Experimental-Wan22Lightning workflow supporting Wan2.2-I2V-A14B-4steps-lora-rank64-Seko-V1.
  • This is a quick Experimental release as I haven't had much time for thorough testing.
• Added support for downloading the following files:
  • Download\loras\Fast\Wan2.2-I2V-A14B-4steps-lora-rank64-Seko-V1-H.bat
  • Download\loras\Fast\Wan2.2-I2V-A14B-4steps-lora-rank64-Seko-V1-L.bat
• Added explicit installation of PyYAML.
  • This may have resolved installation issues in Python 3.12 environments.
• Fixed a bug in SaveLastImage where the start frame image was being saved instead of the final frame image.

I forgot to set up GitHub repository Watch settings, so I wasn't noticing Issue registrations.
My apologies.

2025/08/07

• Continuing to update Easy/I2V workflow.
  • As the number of green nodes has increased, frequently used nodes are now colored red.
  • Added Trim functionality to PostProcess to remove specified number of frames from the beginning or end of videos.
  • Changed PointMosaic to be specified on the video before Refiner application.
    • Eliminates the need to respecify point positions after Refiner.
  • Changed to toggle function enable/disable with single toggles.
  • Adjusted video save filenames for specific purposes.
  • Made Refiner's Full Block swap an optional feature.
    • Resolves the issue of slightly increased VRAM consumption in LowNoise.
  • Added support for disabling start images as an experimental and temporary feature.

2025/08/06

• Easy/I2V workflow now supports Refiner.
  • Enlarges generated videos and refines them with LowNoise V2V that inherits from original video generation.
  • Added Refiner example samples to Sample/.
    • Generate with long side 640px for gacha, refine to 960px, then upscale 2x for FullHD+.
    • Refiner doesn't reduce VRAM consumption.
      Its feature is enabling faster gacha loops at lower resolutions.
• Added support for downloading Download\loras\Nsfw\GeneralNsfw22_v008a-L.bat.
• Updated ComfyUI and WanVideoWrapper versions.

2025/08/04

• Continuing to update Easy/I2V workflow.
  • Added support for Rewrite Prompt that rewrites prompts with local Qwen, separate from the prompt extension in ExtendPrompt.
• Updates to WanVideoWrapper may have resolved memory shortage issues.

2025/08/03

• Continuing to update Easy/I2V workflow.
  • Added support for Japanese prompt translation to English.
    You can choose between local translation and Google Translator.
  • Made text encoder selectable from Kijai, Native, NAG.
    Default is scaled fp8 Native.
  • Enabled concatenation of preset prompts and input prompts for negative prompts.
  • Added support for prompt generation from images.
    • The effect of calming movement is more powerful than Wan 2.1, so usage seems limited.
  • Added support for prompt extension by locally-running Qwen in WanVideoPromptExtender.
  • Added support for selective combination of multiple types of positive prompts.
  • Changed to set video length in seconds.
  • Made frame interpolation selectable from 60FPS, 60FPS 1.3x Speed, 30FPS.

2025/08/02

• Added Easy/00_I2V(ImageToVideo) workflow.
  • Supports three types of mosaic like EasyWanVideo's NsfwFast, upscaling, frame interpolation, and labels, eliminating the need for regeneration after post-processing from seed gacha hits.
• Civitai API key is now required for updates and installation.
• Added support for downloading Download\loras\Nsfw\GeneralNsfw22_v006.bat.
  • Updated to version Download\loras\Nsfw\GeneralNsfw22_v008a.bat.
• Added user feedback section.
• Added support for downloading examples to Sample/.
• Experimental introduction of unsupported Python 3.10 compatibility.
• Reverted to version pinning of numpy==1.26.4 and opencv-python==4.11.0.86 as safety measures.
• Added EndImage functionality to Easy/00_I2V(ImageToVideo) workflow, which was also available in 99_WIP_Kijai_LowSpec.

2025/07/31

• 99_WIP_Kijai_LowSpec workflow now supports LoRA loading and old-type end images.
  • 20:45 Fixed a bug where end images didn't work properly.
• Added support for downloading Download\vae_approx\taew2_1.bat.
  • With ComfyUI-Manager and Video Helper settings, preview is displayed in WanVideoSampler.
• Added support for downloading Wan2.2-I2V-A14B in Q4_K_M, Q6_K, Q8_0 formats.
  • You can download all at once with Download\diffusion_models\Base-Wan2.2-I2V-A14B.bat.
• Prompt Dance (Old WIP Workflow)
  • You can generate quickly by disabling upscaling and frame interpolation.
• Operate the green nodes above Easy/99_WIP_Kijai_LowSpec.json.

Copy ComfyUI/extra_model_paths.yaml.example and rename it to ComfyUI/extra_model_paths.yaml, then rewrite the content to:

comfyui:
	base_path: D:/EasyWanVideo/ComfyUI/
	loras: models/loras/

This allows you to reference Wan 2.1 LoRA from EasyWanVideo.
Replace D:/EasyWanVideo/ComfyUI/ with the absolute path to EasyWanVideo/ComfyUI/ where EasyWanVideo is installed.