<h3 id="extreme-speed-up-of-wan-2.2-t2v-and-i2v-with-just-4-steps">Extreme speed-up of Wan 2.2 T2V and I2V with just 4 steps</h3><a target="_blank" rel="ugc" href="https://huggingface.co/lightx2v">https://huggingface.co/lightx2v</a>Update: 25/10/30<ul><li>Refined HIGH lora, extracted from kijai from the most recent checkpoint of lighx2v</li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Kijai/WanVideo_comfy/tree/main/LoRAs/Wan22_Lightx2v">https://huggingface.co/Kijai/WanVideo_comfy/tree/main/LoRAs/Wan22_Lightx2v</a></li><li>not an official lighx2v lora</li><li>key warnings can be ignored (unused weights)</li></ul>Update: 25/10/23<ul><li>Added new release 1022 of Lighx2V LoRAs</li><li>Quality seems improved</li><li>Possible no ghosting anymore</li><li>Compatible to other LoRAs</li></ul>Update: 25/10/15<ul><li>MoE-Distill is the latest speed-up LoRA</li><li>Patched version from KJ (HIGH)</li><li>Restores some motions compared to other speed-up's</li><li>Quality seems high</li></ul>Needs more testing. Some situations are still slow motion.Use: Euler+simple, Sigma-Shift 5, CFG 1⚠️ Attention<ul><li>Some configurations seems to blur/ghost</li><li>It seems incompatible with some LoRAs (e.g. NSFW)</li><li>It seems incompatible some checkpoints</li></ul>Update: 25/08/21<ul><li>More Sampler+Scheduler info</li><li>ATM: The LoRAs cause a bit slowdown compared to native sampling without</li></ul>Update: 25/08/08<ul><li>The I2V Lora-Upload seem to be broken, re-uploaded it again</li><li>Changed some Usage descriptions</li><li>Mentioned scheduler/sampler combo</li></ul>Update: 25/08/07<ul><li>Added I2V LoRAs</li></ul>Update: 25/08/05<ul><li>Added more Infos and LoRAs fix1</li></ul><h3 id="usage">Usage</h3><h3 id="swarmui">SwarmUI</h3><ol><li>CFG: 1</li><li>Steps: 4</li></ol><ul><li>For T2V Use with this at the end of your prompt: <code>&lt;base&gt; &lt;lora:Wan2.2-Lightning_T2V-A14B-4steps-lora_HIGH_fp16&gt; &lt;refiner&gt; &lt;lora:Wan2.2-Lightning_T2V-A14B-4steps-lora_LOW_fp16&gt;</code> (adapt the lora filenames to whatever specific filenames you have locally)</li><li>For I2V, use <code>&lt;video&gt; &lt;lora:...i2v-high&gt; &lt;videoswap&gt; &lt;lora:...i2v-low&gt;</code> and of course use the i2v loras</li><li>the I2V Lightning LoRA appears to target 16 fps</li></ul><h3 id="general:">General:</h3><ul><li>1 strength on both</li><li>2+2 steps (total 4)</li><li>switch at 2 (50%)</li><li>Sigma Shift: 5</li><li>Sampler + Scheduler combinations:<ul><li>Euler + Beta</li><li>LCM + LTX</li><li>Euler + Simple</li></ul></li></ul><ul><li>You need to be very descriptive, short prompts leading to much more random things.</li><li>work best at 81x480x832, 81x832x480, 81x720x1280, 81x1280x720</li></ul>usage tip first loras (outdated):<ul><li>0.125 strength on both</li><li>4 steps</li><li>switch at 2</li></ul>Did not figured out optimal settings for comfyui or swarmui ... still testing. Let me know! :) Generation times &lt;3 minutes<hr /><span style="background-color:rgb(26, 27, 30);color:rgb(141, 141, 143);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">YOU are responsible for outputs as always! If you make <a target="_blank" rel="ugc" href="https://civarchive.com/content/tos"><span style="background-color:rgb(26, 27, 30);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">ToS</a><span style="background-color:rgb(26, 27, 30);color:rgb(141, 141, 143);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px"> violating content and I get aware I WILL report this.

Wan 2.2 Lightning LoRAs

Refined i2v lora, extracted from recent checkpoint of lighx2vby kijai

Wan_2_2_I2V_A14B_HIGH_lightx2v_4step_lora_v1030_rank_64_bf16.safetensors

high-r64-1030

low-r64-1022

high-r64-1022

i2v-Moe-Distill-Low

i2v-Moe-Distill-High bf16

i2v-seko-v1-low-r64

i2v-seko-v1-high-r64

t2v-low-fix1

t2v-high-fix1

T2V-A14B-4st-low

T2V-A14B-4st-high

<h3 id="2.2">2.2</h3>For I2V this motion helper node is extremely useful:<a target="_blank" rel="ugc" href="https://github.com/princepainter/ComfyUI-PainterI2V">https://github.com/princepainter/ComfyUI-PainterI2V</a>10/30 High lora was further refined.New I2V 1022 versions are out. They have by far the best prompt following / motion quality yet. (The lora key warning is fine, it just contains extra modulation keys that comfyui does not use. It does not matter.)<a target="_blank" rel="ugc" href="https://github.com/VraethrDalkr/ComfyUI-TripleKSampler">https://github.com/VraethrDalkr/ComfyUI-TripleKSampler</a>T2V versions just got updated 09/28. Probably still best to use a step or two with cfg / without the lora to establish motion with high noise as usual like:2 steps high noise without the low-step lora at 3.5 CFG2 steps high noise with lora and 1 CFG2-4 steps low noise with lora and 1 CFGIts definitely a big improvement either way.T2V:Using their full 'dyno' model for your high model seems best."On Sep 28, 2025, we released two models, Wan2.2-T2V-A14B-4steps-lora-250928 and Wan2.2-T2V-A14B-4steps-250928-dyno. The two models share the same low-noise weight. Wan2.2-T2V-A14B-4steps-250928-dyno delivers superior motion rendering and camera response, with object movement speeds that closely match those of the base model. For projects requiring highly dynamic visuals, we strongly recommend using Wan2.2-T2V-A14B-4steps-250928-dyno. Below, you will find some showcases for reference."<a target="_blank" rel="ugc" href="https://huggingface.co/lightx2v/Wan2.2-Lightning/blob/main/Wan2.2-T2V-A14B-4steps-250928-dyno/Wan2.2-T2V-A14B-4steps-250928-dyno-high-lightx2v.safetensors">https://huggingface.co/lightx2v/Wan2.2-Lightning/blob/main/Wan2.2-T2V-A14B-4steps-250928-dyno/Wan2.2-T2V-A14B-4steps-250928-dyno-high-lightx2v.safetensors</a><hr /><h3 id="2.1">2.1</h3>7/15 update: I added the new I2V lora, seems to have much better motion than using the old text to video lora on a image to video model. Example is 4 steps, 1 CFG, LCM sampler, 8 shift. I uploaded the new version of the T2V one also.I'm also putting up the rank 128 versions extracted by Kijai, they are double the size but are slightly better quality.I suggest using it with the Pusa V1 lora as well, it seems to improve movement even more: <a target="_blank" rel="ugc" href="https://huggingface.co/Kijai/WanVideo_comfy/tree/main/Pusa">https://huggingface.co/Kijai/WanVideo_comfy/tree/main/Pusa</a> No need for 2 sampler WF anymore IMO. Just plug it into your normal WF with 1 CFG and 4 steps or so. Could prob sharpen it with another pass if you wanted but it no longer hurts the movement like before for image to video.Full Image to video Lightx2V model: <a target="_blank" rel="ugc" href="https://huggingface.co/lightx2v/Wan2.1-I2V-14B-480P-StepDistill-CfgDistill-Lightx2v/tree/main/distill_models">https://huggingface.co/lightx2v/Wan2.1-I2V-14B-480P-StepDistill-CfgDistill-Lightx2v/tree/main/distill_models</a><hr />Old: <a target="_blank" rel="ugc" href="https://huggingface.co/lightx2v/Wan2.1-T2V-14B-StepDistill-CfgDistill">lightx2v</a> made a 14B self forcing model that is a massive improvement compared to Causvid / Accvid. Kijai extracted it as a <a target="_blank" rel="ugc" href="https://huggingface.co/Kijai/WanVideo_comfy/blob/main/Wan21_T2V_14B_lightx2v_cfg_step_distill_lora_rank32.safetensors">lora</a>. Example above was generated in about 35 seconds on a 4090 using 4 steps, lcm, 1 cfg, 8 shift, still playing with settings to see what is best.<hr />Please don't send me buzz or anything, give the <a target="_blank" rel="ugc" href="https://github.com/ModelTC/lightx2v">lightx2v</a> team or <a target="_blank" rel="ugc" href="https://www.myminifactory.com/users/kijai">kijai</a> support if anyone.

Files with SHA256: 2c4d41a22b0c16dc360570dfa12b53b2b5452338935eb8ea24bf40d2ada89bd0