Stable Video Infinity 2 1080p 60 FPS ComfyUI Workflow for Wan 2.2 I2V by Smittie — v6 — v4

Author notes from Smittie_SVI2_1080p_60FPS.json · selected release

### 1. If you ran out of RAM, deactivate the image upscaler first, then try again. | Source Resolution | Scale By | Final Resolution | |---------------------|---------------|---------------| | 480p | 1.5x | 720p | | 480p | 2.25x | 1080p | | 720p | 1.5x | 1080p | | 720p | 2x | 1440p | ### 2. The RIFE Frame Interpolation creates frames in-between the generated ones. Most AI models run at 16 FPS, this scales the frames by 3x and therefore to 48 FPS. Then the video will be saved with 60 FPS output to speed up the video by 1.25x, as most AI models default speed looks like a slow-motion. | Target FPS | Frame Rate | Thus Speed Up By | |---------------------|---------------|---------------| | 32 | 60 | ~2x | | 40 | 60 | 1.5x | | 48 | 60 | 1.25x | | 60 | 60 | 1 |

Author notes from Smittie_SVI2_1080p_60FPS.json · selected release

**Diffusion model** - Original: [Wan fp8 models](https://huggingface.co/Kijai/WanVideo_comfy_fp8_scaled/tree/main/I2V) - Recommended: [DaSiWa WAN 2.2 I2V 14B | Lightspeed](https://civitai.com/models/1981116?modelVersionId=2388548) It is a NSFW model, but it can be used for SFW too and has very good quality and performance. **LoRA** - Only needed for original Wan to speed up generation: [Lightx2v models](https://huggingface.co/lightx2v/Wan2.2-Distill-Loras/tree/main) - Always needed: [SVI 2.0 Pro models](https://huggingface.co/Kijai/WanVideo_comfy/tree/main/LoRAs/Stable-Video-Infinity/v2.0) **Text encoder** - Important as other CLIP's might not work: [CLIP Text Encoder](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/tree/main/split_files/text_encoders) **VAE** - wan_2.1_vae is sufficient: [Wan VAE's](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/tree/main/split_files/vae) **Model Storage Location** ``` 📂 ComfyUI/ ├── 📂 models/ │ ├── 📂 diffusion_models/ │ │ └── │ ├── 📂 loras/ │ │ └── │ ├── 📂 text_encoders/ │ │ └── │ └── 📂 vae/ │ └── ```

Author notes from Smittie_SVI2_1080p_60FPS.json · selected release

# Smittie's Workflow ## Description With this workflow you can generate and continue long videos in 1080p and 60 FPS without color or motion mismatch between the video segments and inject anchor images (not end images) as further detailed guidance. At least 16 GB of VRAM recommended, but with more quantized versions of the Wan 2.2 I2V diffusion models you can go even lower. This workflow was tested with 24 GB of VRAM. ## Install Instructions ### 1. [Install KJNodes](https://github.com/kijai/ComfyUI-KJNodes) ### 2. [Install Stable Video Infinity 2](https://github.com/vita-epfl/Stable-Video-Infinity) ### 3. Install all missing custom nodes via the ComfyUI Manager ### 4. Download all needed models ## Video Instructions If you don't know what to do, you can watch the [Stable Video Infinity Tutorial by AI Search](https://www.youtube.com/watch?v=-3DVJu72VhE). He works with another workflow I build upon. ## Acknowledgement Most credits goes to the Stable Video Infinity, Lightning, Wan, ComfyUI and CivitAI team, as well as kijai, AI Search, darksidewalker, Joviex, GACLove and Firetheft. Thank you guys!

Author notes from Smittie_SVI2_1080p_60FPS.json · selected release

### You can enable "Load Anchor Image Samples" node per segment individually, if you want to give further detailed guidance for the video generation. If bypassed it uses the start image/frame only. ### It could be used e.g.: ### 1. Give further details how the character looks from side/back. ### 2. What the character wears below its main clothes. (Matryoshka) ### 3. How another introduced character should look like. ### 4. How the area looks behind the camera/character. ### 5. How the building looks inside, when the camera dolly moves in. ### etc. ### It does not work like an end image! The provided anchor image is only strong if your prompt leads to it.

Author notes from Smittie_SVI2_1080p_60FPS.json · selected release

### Here you can add for each Video Generation Segment individually multiple LoRA's at once, change their model weight and, if the LoRA creator provided it, get the trigger words automatically. Otherwise, you can write them yourself by clicking on the pen symbol.

Author notes from Smittie_SVI2_1080p_60FPS.json · selected release

### Here you can write a general text prompt that is true for all video generation segments, like the background, lighting, mood, quality, etc. You can leave it empty, if you don't want to use it. ### For specific actions write the text prompt for each individual video generation segment in step 5.

Author notes from Smittie_SVI2_1080p_60FPS.json · selected release

### You can easily write a Text Prompt here for each segment. To change the CFG or Lenght per segment individually you have to remove the line from the Video Generation node.

Author notes from Smittie_SVI2_1080p_60FPS.json · selected release

### You can enable a preview for each video generation segment. The preview is set to 20 FPS as this will speed up the animation by 1.25x. This way you have the same speed as the end result will look like.

Author notes from Smittie_SVI2_1080p_60FPS.json · selected release

### Here you can load your image and crop or resize it to 720p.

Author notes from Smittie_SVI2_1080p_60FPS.json · selected release

### This LoRA gallery allows you to actually see a preview of each LoRA. You need to click on the cloud symbol that appears when you hover over a LoRA, that will get the information from civitai.com ### Write in the top filter "high" and in the bottom "low" to see only the correct high or low noise LoRA's. ### You can add this way multiple LoRA's at once, change their model weight and, if the LoRA creator provided it, get the trigger words automatically (which are set on the low LoRA). Otherwise, you can write them yourself by clicking on the pen symbol. ### The "SVI_v2_PRO_..." LoRA must be loaded always! Otherwise the long video generation will degrade very early.

Author notes from Smittie_SVI2_1080p_60FPS.json · selected release

### You can test and find the best setting by yourself. The following table is for reference. | Description | BasicScheduler Steps | SplitSigmas Step | Length Per Segment | |---------------------|---------------|---------------|---------------| | For not much movement (Best Performance) | 2 | 1 | 33 | | Good Quality, Performance, Prompt adherence | 4 | 2 | 49 | | Fast movement or new objects | 6 | 3 | 49 | | Model Default | 5 | 2 | 81 | ### As the model default is 81 frames and the result is speed up at the end by 1.25x, 101 frames are actually equivalent to 5s video. So with a length of 33 and 3 video generation segments, the final video will be ~5s long.

Author notes from Smittie_SVI2_1080p_60FPS.json · selected release

### You can enable more Video Generation Segment (Extension), if you want longer videos. More than 10 segments is not recommended as degradation is still incoming. ### The recommended usage to avoid long processing and a lot of unwanted saved videos would be like this: ### 1. Bypass the "7. Save Video" group to avoid processing time for upsampling and saving. ### 2. Setup steps 1-4 you wish to use for the final video. ### 3. Bypass all "Video Generation Segment (Extension)" nodes and their preview except the first one "Video Generation Segment (Initial)". ### 4. Run the workflow. Then check if the preview is to your liking or change your prompt/LoRA/Anchor Image and run again until you're satisfied. ### 5. Enable now the next video generation segment and its preview. If you run the workflow, only this new segment will process as the previous is already done. Do not change any previous settings anymore or it will re-run all previous segments again! ### 6. If you reached the desired video length, enable "7. Save Video" and run this workflow to save your result.

Author notes from Smittie_SVI2_1080p_60FPS.json · selected release

### The loaded video must be setup, that it has its original Wan model resolution of 720p and frame rate of 16 FPS. ### For this workflow, the generated videos were upsampled to 1080p and 3x interpolated from 16 to 48 frames (only the image count matters here, not the 60 frame rate!), therefore select only every 3rd frame and resize those to 720p. ### Change only the "Load Video" subgraph node to achieve this. | Source FPS | select_every_nth | |---------------------|---------------| | 16| 1 | | 32 | 2 | | 40 | 2 or 3 | | 48 | 3 | | 60 | 4 |

Author notes from Smittie_SVI2_1080p_60FPS.json · selected release

### Set the boolean to "false" for 3A Image to Video or to "true" for 3B Continue Video. A bypass is not necessary anymore. ### However, it could be that in 3B Continue Video an existing video must be chosen in order to run properly, even though it won't be used further when the bool is set to false.

Author notes from Smittie_SVI2_1080p_60FPS.json · selected release

### To spare some RAM, only the last 81 frames and the very first frame of the given video will be processed. It does not change the result to process more than that.