Wan 2.2 14B Fun Control ComfyUI template (video_wan2_2_14B_fun_control): control video wiring, bypassed Canny, 20-step default and 4-step LoRA group

Wan 2.2 14B Fun Control in ComfyUI: the official template Scope. This guide reads the official video_wan2_2_14B_fun_control.json in Comfy-Org/workflow_templates at commit 0cc0fba562efe9f68e48050c2b82ed6495050bc2 (SHA256 52752247...88ea, 61,018 bytes). All 44 nodes and 50 links were traced statically. The file has no subgraphs and no switch nodes. Nothing was run, and no model, image, video or linked page was downloaded. 1. Prerequisites - A ComfyUI build with the core nodes Wan22FunControlToVideo, LoadVideo, GetVideoComponents and Canny. Every node is comfy-core, saved at 0.3.45 or 0.3.49; those are saved version strings, not a tested minimum. The graph uses no custom node pack. Note 185 ("About Control Video") says you may need comfyui_controlnet_aux or ComfyUI-DepthAnythingV2 to preprocess other control types. It also says the model supports Canny, Depth, Pose and trajectory control. - Files, as declared by the loaders and the Model Links note (node 161, sizes as stated there): - models/diffusion_models/wan2.2_fun_control_high_noise_14B_fp8_scaled.safetensors and wan2.2_fun_control_low_noise_14B_fp8_scaled.safetensors (13.31 GB each), from Comfy-Org/Wan_2.2_ComfyUI_Repackaged. - models/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors (6.27 GB), from Comfy-Org/Wan_2.1_ComfyUI_repackaged. - models/vae/wan_2.1_vae.safetensors (242.1 MB). - Only for the 4-step group: models/loras/wan2.2_i2v_lightx2v_4steps_lora_v1_high_noise.safetensors and _low_noise.safetensors (1.14 GB each). These are the i2v Lightning LoRAs, applied here to the Fun Control experts. The default route needs about 33 GB of files, and the LoRAs add about 2.3 GB. No CLIP vision model is used. 2. The two groups Like the Wan 2.2 14B first-last-frame template, the file holds two complete copies of the pipeline: - "Default workflow" (active as saved). - "Wan2.2 Fun Control fp8 scaled+ 4 steps LoRA": every node in it is bypassed (mode 4) as saved. Note 183 says the Lightning LoRA reduces generation time but loses video dynamics, and that you enable one workflow as needed. To switch, toggle bypass on a group's nodes (Ctrl + B) and bypass the other group. Unlike the first-last-frame template, both groups here use the same prompts and the same example files. 3. Control video and preprocessing (default group) - LoadVideo 158 (example video_wan2_2_14B_fun_control_start_image_control_video.mp4) feeds GetVideoComponents 156. Its images output (link 285) goes to Canny 157 (saved 0.1 and 0.6, the low and high thresholds in the core node's order). Canny's output (link 286) reaches the control_video input of Wan22FunControlToVideo 160. PreviewImage 159 shows Canny's output. - As saved, Canny 157 and PreviewImage 159 are bypassed, even in the active group. A bypassed node passes its image input through, so the clip's frames reach control_video unprocessed. The example file name suggests it is already a control video. It was not viewed. - With ordinary footage, enable Canny 157 (and PreviewImage 159 to check the edges). For depth or pose control, use a preprocessor from the packs named in note 185 and feed its images to control_video. - GetVideoComponents also outputs audio and fps, but neither is connected. The output is written at a fixed 16 fps without audio. 4. Reference image LoadImage 145 (example video_wan2_2_14B_fun_control_start_image.jpg) feeds ref_image on Wan22FunControlToVideo 160 (link 284). 5. Size and length Wan22FunControlToVideo 160 holds width 640, height 640, length 81 and batch 1. Note 186 says the model supports multi-resolution video at 512, 768 and 1024, and was trained with 81 frames at 16 fps. The saved 640 is not one of those sizes. That is a note-versus-setting difference, not a tested problem. How the node handles a control clip that has more or fewer than 81 frames, or a different size, is not in this file. Prepare an 81-frame control clip at the output size and aspect ratio. 6. Default sampling - Prompts: positive CLIPTextEncode 99 (a girl with white curly ponytails dancing slowly before the Leaning Tower of Pisa, in a realistic 3D style) and negative 91 (the usual Wan Chinese quality list). Both use CLIPLoader 90 (umt5_xxl_fp8_e4m3fn_scaled, type wan). VAELoader 92 loads wan_2.1_vae. - High noise: UNETLoader 101 feeds ModelSamplingSD3 93 (shift 8), then KSamplerAdvanced 96: add_noise enable, seed 909895218860439 set to randomize, 20 steps, CFG 3.5, euler, simple, steps 0 to 10, leftover noise enable. - Low noise: UNETLoader 102 feeds ModelSamplingSD3 94 (shift 8), then KSamplerAdvanced 95: no added noise, seed 0 fixed, 20 steps, CFG 3.5, euler, simple, steps 10 to the end. Sampler 95 takes its latent from 96. - Output: VAEDecode 97, then CreateVideo 100 at 16 fps, then SaveVideo 98 (prefix video/wan2.2_fun_control, format auto, codec auto). Expect one 640x640 video of 81 frames, about 5 s, following the control clip's motion with the reference image's look. That is the template's purpose; no output was produced here. 7. 4-step LoRA group (bypassed as saved) - Models: UNETLoader 165 feeds LoraLoaderModelOnly 181 (wan2.2_i2v_lightx2v_4steps_lora_v1_high_noise, strength 1), then ModelSamplingSD3 176 (shift 8). UNETLoader 166 feeds LoraLoaderModelOnly 182 (the _low_noise LoRA), then ModelSamplingSD3 177 (shift 8). - Samplers: KSamplerAdvanced 169 runs 4 steps at CFG 1, euler, simple, steps 0 to 2 with leftover noise, seed 342522439403882 set to randomize. KSamplerAdvanced 175 runs steps 2 to 4 without added noise. - Own inputs: CLIPLoader 167, prompts 179 and 162 (same text as the default group), VAELoader 168, LoadImage 178 and LoadVideo 174 (same example files), GetVideoComponents 171, Canny 173 (bypassed), PreviewImage 172 and Wan22FunControlToVideo 180 (640x640, length 81). Output goes through VAEDecode 164 and CreateVideo 163 (16 fps) to SaveVideo 170, with the same prefix as the default group. 8. Run it 1) Put the four default files in the folders above, plus the two LoRAs only if you will use the 4-step group. 2) Open the template and keep the Default workflow group active. 3) Load your control clip in LoadVideo 158. If it is not already a control video, enable Canny 157 or add another preprocessor. 4) Load your reference image in LoadImage 145. 5) Edit prompts 99 and 91, and size and length on node 160. 6) Run. Expect one video under video/wan2.2_fun_control. For the 4-step group: enable its nodes, bypass the Default workflow group, and load the clip and image into LoadVideo 174 and LoadImage 178. 9. Failure cases and checks - Ordinary footage used as control: with Canny bypassed, raw frames go in as the control signal. Enable a preprocessor first. - Preprocessor not installed: Canny is core. The depth and pose preprocessors named in note 185 are third-party packs, which the graph does not use. - Both groups enabled: both pipelines run through their own loaders, so the time of both routes is spent and two videos are written under the same prefix. - Edits that do nothing: the groups do not share loaders, inputs or sizes. Edit the group you run. - Clip length, fps or size mismatch: see section 5. The output fps is fixed at 16, whatever the clip's own fps. - Missing LoRA files only matter for the 4-step group. Missing Fun Control experts stop both groups. These are not the i2v or t2v experts. - Lightning trade-off: note 183 says the 4-step LoRAs reduce time but lose video dynamics. 10. How it differs from the other 14B templates It uses the same two-group bypass pattern as the Wan 2.2 14B first-last-frame template, and the same 20-step default at CFG 3.5 as the T2V guide (first-last-frame uses CFG 4). Its 4-step group keeps shift 8, where first-last-frame's uses 5, and it applies the i2v lightx2v LoRAs to the Fun Control experts. The step split is half and half on both routes. Evidence and rights. Evidence status: source_inspected, not_run, unreviewed. This is shared-founder desk curation by the tutorial lane. The workflow repository declares the MIT licence; that covers this analysis of the template, not the Wan models, LoRAs, preprocessors, example media or outputs, which keep their own terms.

{
  "citations": [
    {
      "url": "https://raw.githubusercontent.com/Comfy-Org/workflow_templates/0cc0fba562efe9f68e48050c2b82ed6495050bc2/templates/video_wan2_2_14B_fun_control.json",
      "note": "Exact pinned template; node, link and group byte ranges and excerpt hashes are in the source pins."
    },
    {
      "url": "https://raw.githubusercontent.com/Comfy-Org/workflow_templates/0cc0fba562efe9f68e48050c2b82ed6495050bc2/LICENSE",
      "note": "Repository MIT licence at the same commit (existing native source)."
    }
  ]
}