Instructions to use Alex995647/loras-wan-2.2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use Alex995647/loras-wan-2.2 with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("Wan-AI/Wan2.2-T2V-A14B", dtype=torch.bfloat16, device_map="cuda") pipe.load_lora_weights("Alex995647/loras-wan-2.2") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Draw Things
Add 25 LoRA(s): realistic-fire-wan-2-2-i2v-5b, wan-2-2-i2v-unbuttoning-motion, female-superhero-costumes-wan, aether-blast-wan-2-2-5b-i2v-loraβ¦
6802706 verified Download female-superhero-costumes-wan/info.txt from Alex995647/loras-wan-2.2: direct link, hf CLI and curl.
- Browser
- Download file 28.7 kB
-
https://huggingface.co/Alex995647/loras-wan-2.2/resolve/main/female-superhero-costumes-wan/info.txt
- Command line
-
hf download hf://Alex995647/loras-wan-2.2/female-superhero-costumes-wan/info.txt
-
curl -L -o info.txt https://huggingface.co/Alex995647/loras-wan-2.2/resolve/main/female-superhero-costumes-wan/info.txt
28.7 kB
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| FEMALE SUPERHERO COSTUMES - WAN | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| IDENTITY | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| Base model Wan 2.2 | |
| Generation mode not declared by the author | |
| Checkpoint T2V-A14B (MoE 14B, high- and low-noise experts) | |
| Category Character / subject | |
| Author darkroast175696 | |
| Source https://civitai.com/models/2402924 | |
| Provenance CivitAI Β· model 2402924 Β· version 2701797 | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| WEIGHTS | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| File Female-Superhero-Costumes-Wan-v1-LowNoise-herostyle.safetensors | |
| Version Low Noise v1.0 | |
| Size 146.4 MB | |
| Format SafeTensor / unknown precision | |
| SHA256 131d9537d77f6d31bd7f307122d072b46f21fcfe7fe2d8713677a0437ef463e8 | |
| Published 2026-02-18 | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| TRIGGER WORDS | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| βΊ "herostyle" | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| RECOMMENDED SETTINGS | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| The author did not publish specific settings. | |
| No author settings were published. Community baseline: steps Official | |
| Wan2.2 reference code (wan/configs/): sample_steps 40 for T2V-A14B and | |
| I2V-A14B with the expert handover at the SNR boundary, and 50 for TI2V-5B. | |
| ComfyUI's CURRENT 14B t2v/i2v templates ship only the distilled | |
| configuration: 4 steps total, split 2 on the high-noise expert then 2 on | |
| the low-noise expert (two KSamplerAdvanced nodes, start/end 0->2 and 2->4). | |
| An undistilled 20-step configuration split 10/10 shipped as a second, | |
| bypassed node group in an earlier revision of the t2v template and still | |
| ships as the ACTIVE default in the current video_wan2_2_14B_flf2v template | |
| (20 steps, split 0->10 and 10->end). TI2V-5B: 20 steps in ComfyUI's | |
| template, 50 in the reference code; single sampler, no split., cfg Official | |
| reference sample_guide_scale is a per-expert pair: (3.0 low- noise, 4.0 | |
| high-noise) for T2V-A14B and (3.5, 3.5) for I2V-A14B β the two experts are | |
| allowed different CFG. TI2V-5B: 5.0 in both the reference code and | |
| ComfyUI's template. Undistilled ComfyUI values vary by template revision: | |
| 3.5 in the older t2v template's bypassed undistilled group, 4.0 in the | |
| current flf2v template. With ANY 4-step distill LoRA: cfg 1.0 on both | |
| samplers (verified in the shipped t2v and i2v templates)., sampler euler + | |
| simple scheduler on both A14B experts, with shift set via ModelSamplingSD3: | |
| 5.0 in ComfyUI's current distilled 14B t2v/i2v templates, 8.0 in the | |
| undistilled groups (older t2v revision, and the current flf2v template). | |
| Official reference sample_shift differs by task: 12.0 for T2V-A14B, 5.0 for | |
| I2V-A14B, 5.0 for TI2V-5B. TI2V-5B in ComfyUI: uni_pc + simple, shift 8.0., | |
| resolution A14B: 480P and 720P β 1280x720 or 832x480 (and portrait). | |
| Dimensions should be multiples of 16 (Wan2.1 VAE stride 4x8x8 combined with | |
| 1x2x2 patching, both read from the reference configs); ComfyUI's templates | |
| ship a cheap 640x640. Length 81 frames at 16 fps (~5s), frames 4n+1. | |
| TI2V-5B: 720P, 1280x704 or 704x1280; multiples of 32, because the | |
| Wan2.2-VAE stride is 4x16x16 and the model card states an additional | |
| patchification layer brings total compression to 4x32x32 β which is why it | |
| is 704, not 720. Length 121 frames at 24 fps (~5s)., lora_strength Official | |
| Lightning/lightx2v 4-step LoRAs: 1.0 on both experts β verified in | |
| ComfyUI's shipped t2v and i2v templates, where both LoraLoaderModelOnly | |
| nodes carry strength 1.0. Character/style/motion LoRAs are commonly run at | |
| 0.6-1.0, and when stacked on top of a 4-step distill LoRA the usual | |
| community move is to pull the high-noise-side strength down (roughly | |
| 0.5-0.8); those subject-LoRA numbers are community practice, not published | |
| by the authors β see "unverified".. | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| ADOPTION & POPULARITY | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| Downloads 851 | |
| Rating 100% positive (62/62) | |
| Comments 1 | |
| Published 202 days ago | |
| Download rate 2.3/day since release | |
| Adoption rank #182 of 200 in this collection (Niche) | |
| Momentum rank #147 of 200 by download rate (Long tail) | |
| Adoption tier is the percentile of total downloads within this collection; | |
| momentum is the percentile of downloads-per-day since release. Momentum is | |
| ranked rather than measured against a fixed rate, because this base model | |
| is itself new and every LoRA looks fast on an absolute scale. Both numbers | |
| are CivitAI's, read on the scrape date at the foot of this file. | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| CHAINING / STACKING | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| Stacks with others not stated | |
| The author gave no chaining guidance. The rules below are the | |
| general Wan 2.2 ones, not anything specific to this LoRA. | |
| General Wan 2.2 rules that apply here (see REFERENCE.txt): | |
| β’ The high/low pair is not a stack. Every A14B LoRA that ships two files | |
| needs the high-noise file on the high-noise UNET chain and the low- | |
| noise file on the low-noise UNET chain β two LoraLoaderModelOnly nodes | |
| on two different models, never both on one model. | |
| β’ Never load a high-noise LoRA onto the low-noise expert or vice versa. | |
| It loads without a single warning because the two checkpoints are | |
| architecturally identical, and the damage shows up only as degraded | |
| output. Filename is the only guard. | |
| β’ A single-file A14B LoRA with no high/low marking (common for Wan2.1-era | |
| LoRAs and for LoRAs extracted from merged models) goes on BOTH experts | |
| at the same strength. Loading it only on the low-noise expert is the | |
| safer half-measure when it fights the composition. | |
| β’ Stack subject/style LoRAs after the accelerator LoRA in the chain: | |
| UNETLoader -> Lightning LoRA -> subject LoRA -> ModelSamplingSD3. Order | |
| within LoraLoaderModelOnly chains is additive so it rarely matters | |
| mathematically, but keeping the accelerator first matches every | |
| reference workflow. | |
| β’ Do not stack two step-distillation LoRAs on the same expert. One | |
| accelerator per expert. | |
| β’ When adding subject LoRAs on top of a 4-step distill LoRA, reduce the | |
| subject LoRA strength rather than the distill LoRA strength β the | |
| distill LoRA is what keeps 4 steps viable at cfg 1.0. | |
| β’ Keep the LoRA strength for the two experts independently tunable. | |
| Because the high-noise expert owns layout and the low-noise expert owns | |
| detail, the community routinely runs a style LoRA weaker on high-noise | |
| and full-strength on low-noise. | |
| β’ TI2V-5B LoRAs chain like any single-model LoRA and can never be mixed | |
| into an A14B chain, or the reverse β the two have entirely different | |
| tensor geometry. | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| PROS | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| + Positively received β 62 thumbs-up with no down-votes. [source: CivitAI | |
| stats] | |
| + Explicit trigger word(s) β herostyle β so the effect can be turned on | |
| and off from the prompt. [source: model metadata] | |
| + Compact at 146 MB β cheap to stack with other LoRAs. [source: file | |
| size] | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| CONS & CAVEATS | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| - Author never states which generation mode it was trained for; test | |
| against your own workflow before relying on it. [source: description] | |
| - Modest adoption (851 downloads) β less community feedback to rely on. | |
| [source: source stats] | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| COMPATIBILITY WARNING | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| Three distinct answers. The structural half of each is verified from the | |
| shipped config.json and safetensors index files; the behavioural severity | |
| is community-reported and flagged in "unverified". (1) A14B <-> TI2V-5B: | |
| INCOMPATIBLE. The 5B model is dim 3072 / 30 layers / 24 heads / ffn 14336 | |
| against the A14B's dim 5120 / 40 layers / 40 heads / ffn 13824 (both | |
| configs fetched and compared). In ComfyUI native this is SILENT | |
| DEGRADATION, not a hard error: keys for blocks 30-39 hit | |
| logging.warning("lora key not loaded: {}") in comfy/lora.py, and keys for | |
| blocks 0-29 match by name but not by shape, so they hit the shape guard | |
| logging.warning("WARNING SHAPE MISMATCH {} WEIGHT NOT MERGED") in | |
| comfy/lora.py or the caught exception logging.error("ERROR ...") in | |
| comfy/weight_adapter/lora.py, which returns the weight unmodified. All | |
| three call sites were read in the current ComfyUI source. The queue | |
| completes, no exception reaches the UI, and the output is effectively that | |
| of the LoRA node bypassed. (2) T2V-A14B <-> I2V-A14B: STRUCTURALLY | |
| COMPATIBLE, behaviourally risky. Verified from the safetensors index files: | |
| both experts carry exactly 1095 tensors with identical key names, and the | |
| only tensor whose shape differs is patch_embedding.weight (in_dim 16 vs | |
| 36), which LoRA trainers normally exclude. So a T2V-A14B LoRA loads onto | |
| I2V-A14B and vice versa with no warnings. What you get is a real but off- | |
| target effect. Treat cross-loading as "loads and does something", never as | |
| "certified". (3) High-noise expert <-> low-noise expert (within one A14B | |
| family): the two checkpoints have identical architecture, so a LoRA trained | |
| on one loads onto the other with zero warnings. This is the most dangerous | |
| case in the family, because there is no signal at all that you mismatched. | |
| Filename is the ONLY discriminator. (4) Wan2.1 14B LoRAs. Verified: | |
| Wan2.1-T2V-14B's config.json is dim 5120 / ffn 13824 / 40 layers / 40 heads | |
| / in_dim 16 β the same block geometry as Wan2.2-A14B, so Wan2.1 T2V LoRAs | |
| load. Wan2.1-I2V-14B is only a partial match: its index has 1303 tensors | |
| including 8 img_emb.* keys and 200 cross_attn k_img/v_img keys for its CLIP | |
| path, which Wan2.2-I2V dropped entirely (0 such keys); those will be | |
| skipped with "lora key not loaded". | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| AUTHOR'S DESCRIPTION | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| Female Superhero Costume LoRA This pair of High and Low noise LoRAs for Wan | |
| 2.2 will clothe your main character in a realistic superhero costume. The | |
| model was trained on about a hundred images of a wide variety of costume | |
| styles with capes, skirts, long jumpsuits, full coverage, cleavage, and so | |
| on. For prompting tips and guidance, see below. In my tests, this LoRA does | |
| not affect the main character's face or the rest of the scene very much at | |
| all, so it should be safe to add to any character or setting you want. I | |
| did not train any topless, nude, or other sexy situations into this model, | |
| so you will need other LoRAs to help with that if it's what you want in | |
| your video. | |
| Important: As always with my models, this LoRA does not represent any real | |
| person in any way. If this model produces a video with any resemblance at | |
| all to a real person, that is purely an accident of AI random generation. | |
| High and Low Noise Models | |
| You can use either one separately or both together for different effects. | |
| I've included some comparison videos in the samples above, and feel free to | |
| experiment to find the look that you like the best. If you use the high | |
| noise model only, Wan will invent some small details in the costume that | |
| may not be what you described in your prompt. If you only use the low | |
| noise, then some of the broad shapes of the costume (like the style of | |
| boots, skirt, and cape) will be more like plain Wan clothing. The end | |
| result is all a matter of taste, so just play around with it and see what | |
| you like. | |
| Prompt Tips and Guidance | |
| - Do NOT use the word "super" or "superhero" in your prompt unless you want | |
| the iconic S logo. | |
| - Trigger word: herostyle | |
| - You can use a very minimal prompt like "she is wearing a hero costume." | |
| and let Wan come up with random stuff for your costume, or you can be very | |
| specific about every piece of the outfit. It's up to you. I'm adding some | |
| sample prompts below to see what kinds of specific attributes I've used in | |
| my testing. | |
| - The training images included some unusual hair colors because it seemed | |
| to fit, so feel free to ask for hair in colors like green, purple, silver, | |
| blue in addition to the regular hair colors. | |
| - The images included some flying and floating scenes which should help Wan | |
| to generated those. Some of these scenes are in the samples above. | |
| - If you can describe a logo or a motif, wan will do its best to help. For | |
| example, "she has a stylized skull logo in the middle of her chest." | |
| - I did not include any masks or helmets in my dataset because I censored | |
| the faces of all the images, which would also hide a mask or helmet. You | |
| should be able to add those back in with prompting, though, especially | |
| since there isn't any face guidance in the dataset. | |
| I hope you enjoy working with this model. Don't forget to post your videos | |
| back to this model page or tag the model when you post your video so I can | |
| see what kinds of creative things you're all coming up with. Enjoy! | |
| Sample prompts: | |
| herostyle. | |
| close-up shot. natural light, warm light, rim light. | |
| a woman is flying through the air above a forest. the wind blows her hair | |
| as she speeds across the landscape a the speed of a jet airplane. | |
| she is wearing a hero costume that includes a red skin-tight stretchy nylon | |
| bodysuit with white accent details and high heel boots. | |
| her bodysuit is full-length covering her legs completely. she has bare arms | |
| and wide straps at the shoulder. the bodysuit neckline is a deeply plunging | |
| v-neck neckline showing her cleavage. translucent mesh panels are placed on | |
| the costume to draw attention to her gorgeous figure. | |
| she wears a belt around her waist. | |
| she has a cape in black with trim colors that match her bodysuit. | |
| the costume is designed in a sci-fi movie style with glowing circuitry and | |
| cybernetic parts. | |
| she is looking to the side with a smiling expression. | |
| she has very long straight purple hair. | |
| she is slim, fit, and athletic with very long slender legs and an extremely | |
| thin narrow waist. | |
| from the front camera angle. | |
| the camera pans and pushes in to highlight the powerful and beautiful hero | |
| woman. the atmosphere is epic and cinematic. herostyle. | |
| close-up shot. natural light, warm light, rim light. | |
| a very muscular woman is getting out of the back door of a sleek black | |
| limousine with tinted windows. she steps out of the limousine and stands up | |
| slowly, then walks towards the camera with a sexy, sensual strut. | |
| she is wearing a hero costume that includes a white skin-tight stretchy | |
| nylon bodysuit with neon orange accent details and high heel boots in a | |
| color matching the trim on her bodysuit. | |
| her bodysuit is full-length covering her legs completely. she has bare arms | |
| and wide straps at the shoulder. the bodysuit neckline is high covering her | |
| chest completely. | |
| she wears a belt around her waist. | |
| the costume is designed in a futuristic military style with glowing | |
| circuitry and cybernetic parts. | |
| she is looking into the distance with a proud expression. | |
| she has long wavy platinum blonde hair. | |
| she is slim, fit, and athletic with very long slender legs and an extremely | |
| thin narrow waist. | |
| from the front camera angle. | |
| the camera pans and pushes in to highlight the powerful and beautiful hero | |
| woman. the atmosphere is epic and cinematic. herostyle. | |
| wide shot. natural light, warm light, rim light. | |
| a very busty woman is drinking a beer at the bar in a pub. | |
| she is wearing a hero costume that includes a lilac-purple skin-tight | |
| stretchy matte carbon fiber bodysuit with black accent details and high | |
| heel boots in a color matching the trim on her bodysuit. | |
| her bodysuit leaves her legs bare. she wears a very short pleated skirt. | |
| she has bare arms and wide straps at the shoulder. the bodysuit neckline is | |
| high covering her chest completely. | |
| she wears a belt around her waist. | |
| the costume is designed in a futuristic military style. | |
| she is looking into the distance with a smiling expression. | |
| she has short wavy brown hair. | |
| she is slim, fit, and athletic with very long slender legs and an extremely | |
| thin narrow waist. | |
| from above, overhead shot, high camera angle, looking down at the woman | |
| from above from the side. | |
| the camera pans and pushes in to highlight the powerful and beautiful hero | |
| woman. the atmosphere is epic and cinematic. Here is a full prompt text | |
| with randomized choices for use with dynamic prompting nodes like the ones | |
| in my workflows. Prompt words that look like __something_here__ are | |
| callouts to separate wildcard files that I have in my comfyui, but it | |
| should be obvious from the prompt context what kinds of words will fill | |
| that space. | |
| herostyle. | |
| {wide|medium|close-up} shot. natural light, warm light, rim light. | |
| a {very busty|extremely muscular|_} woman is { | |
| __all_activities__.| | |
| standing on top of {a tall building in a city with a city skyline and a | |
| blue sky in the background|a tall cliff in a desert with a desert mesa and | |
| a blue sky in the background|a snowy mountain in the himalaya mountains} | |
| with the wind blowing her hair.| | |
| flying through the air above a {city skyline|small town|forest|mountain | |
| range}. the wind blows her hair as she {gently floats down to the | |
| ground|speeds across the landscape a the speed of a jet airplane}. | |
| } | |
| she is wearing a hero costume that includes a {__simple_colors__ skin-tight | |
| stretchy {lycra|nylon|matte carbon fiber} bodysuit with {white|black} | |
| accent details|{white|black|dark grey} skin-tight stretchy | |
| {lycra|nylon|matte carbon fiber} bodysuit with __simple_colors__ accent | |
| details} and high heel boots {_|in a color matching the trim on her | |
| bodysuit}. | |
| her bodysuit {is full-length covering her legs completely|leaves her legs | |
| bare. {the bodysuit is cut very high over her hips|she wears a very short | |
| pleated skirt}}. she has {long sleeves|bare arms and wide straps at the | |
| shoulder}. the bodysuit neckline is {high with a collar|high covering her | |
| chest completely|a deeply plunging v-neck neckline showing her cleavage}. | |
| {_|translucent mesh panels are placed on the costume to draw attention to | |
| her gorgeous figure.} | |
| {_|she wears a belt around her waist.} | |
| {_|she has a {long|_} cape in {white with trim|black with trim|_} colors | |
| that match her bodysuit.} | |
| the costume is designed in a {futuristic military style {_|with glowing | |
| circuitry and cybernetic parts|with areas of armor plating}|sexy | |
| provocative style|old comic book hero style|sci-fi movie style {_|with | |
| glowing circuitry and cybernetic parts|with areas of armor plating}}. | |
| she is looking {at the camera|to the side|into the distance} with a | |
| {proud|smiling|intense dramatic} expression. | |
| she has {short|medium-length|long|very long} {straight|wavy|curly} | |
| {black|blonde|brown|auburn|copper-red|silver|platinum | |
| blonde|pink|blue|green|purple} hair. | |
| she is slim, fit, and athletic with very long slender legs and an extremely | |
| thin narrow waist. | |
| __cameraangle-wan-plainbehind__. | |
| the camera pans and pushes in to highlight the powerful and beautiful hero | |
| woman. the atmosphere is epic and cinematic. | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| EXAMPLE MEDIA | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| 6 preview(s) in ./example_images/ β community results for | |
| this LoRA. Videos keep a matching _poster.jpg still frame. | |
| No prompts are recorded below: CivitAI's public API flags these posts as | |
| having generation metadata but does not return it, so the prompts cannot be | |
| scraped. Open the model page to read them. | |
| 01_video.mp4 (video, 640x960) | |
| still: 01_poster.jpg | |
| 02_video.mp4 (video, 1932x960) | |
| still: 02_poster.jpg | |
| 03_video.mp4 (video, 1024x2066) | |
| still: 03_poster.jpg | |
| 04_video.mp4 (video, 640x960) | |
| still: 04_poster.jpg | |
| 05_video.mp4 (video, 648x1152) | |
| still: 05_poster.jpg | |
| 06_video.mp4 (video, 640x960) | |
| still: 06_poster.jpg | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| HOW TO USE (COMFYUI) | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| 1. Put every LoRA .safetensors in ComfyUI/models/loras/ (subfolders are | |
| fine and are used by reference workflows). Confirmed by the file-layout | |
| tree inside ComfyUI's own video_wan2_2_14B_t2v template note. | |
| 2. Install the base weights alongside them. A14B (from Comfy- | |
| Org/Wan_2.2_ComfyUI_Repackaged, split_files/): | |
| diffusion_models/wan2.2_{t2v,i2v}_high_noise_14B_fp8_scaled.safetensors AND | |
| the matching _low_noise_ file β you need both; | |
| text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors; | |
| vae/wan_2.1_vae.safetensors. TI2V-5B: | |
| diffusion_models/wan2.2_ti2v_5B_fp16.safetensors, the same umt5 text | |
| encoder, and vae/wan2.2_vae.safetensors. | |
| 3. Load the base model with UNETLoader ("Load Diffusion Model"), NOT | |
| CheckpointLoaderSimple β Wan2.2 ships as a bare diffusion model, with the | |
| text encoder loaded separately by CLIPLoader (type "wan") and the VAE by | |
| VAELoader. This is the node set used in every shipped Wan2.2 template. | |
| 4. For A14B, build two parallel chains, one per expert: UNETLoader | |
| (high_noise) -> LoraLoaderModelOnly -> ModelSamplingSD3(shift) -> | |
| KSamplerAdvanced, and separately UNETLoader (low_noise) -> | |
| LoraLoaderModelOnly -> ModelSamplingSD3(shift) -> KSamplerAdvanced. Use | |
| LoraLoaderModelOnly, not LoraLoader β Wan LoRAs patch the DiT only, there | |
| is no CLIP side to patch. | |
| 5. Put the high-noise LoRA file on the high-noise chain and the low-noise | |
| LoRA file on the low-noise chain. If a LoRA ships as a single file with no | |
| high/low marking, loading the same file on both chains is the standard | |
| community handling. | |
| 6. Wire the two KSamplerAdvanced nodes in series on one latent: the first | |
| (high-noise) runs add_noise=enable, start_at_step 0, end_at_step = | |
| boundary, return_with_leftover_noise=enable; the second (low-noise) runs | |
| add_noise=disable, start_at_step = boundary, end_at_step = total, | |
| return_with_leftover_noise=disable. Boundary is step 2 of 4 with a 4-step | |
| Lightning LoRA, step 10 of 20 undistilled. | |
| 7. For TI2V-5B there is only one chain and one KSampler: UNETLoader -> | |
| LoraLoaderModelOnly -> ModelSamplingSD3 -> KSampler. No expert split, no | |
| second sampler. | |
| 8. Fastest correct start: load ComfyUI's built-in templates | |
| video_wan2_2_14B_t2v / video_wan2_2_14B_i2v / video_wan2_2_14B_flf2v / | |
| video_wan2_2_5B_ti2v (Workflow > Browse Templates) and swap the LoRA | |
| filenames β the 14B templates already ship the two-expert plumbing and the | |
| Lightning LoRA nodes wired correctly. Note the 14B t2v/i2v templates are | |
| now packaged as a subgraph. | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| Scraped by lorakit on 2026-09-08 13:29 UTC. | |
| Stats and description are the author's; pros/cons are derived from the | |
| evidence tagged beside each line. | |
| βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |