CivArchive

    🗡️💀 DaSiWa-WAN 2.2 I2V 14B Lightspeed | FP8 Safetensors💀🗡️

    My new flagship model for WAN 2.2 I2V generation - This is the best of the best!

    This is a WAN 2.2 Model: You will need one pair of High + Low.

    Version overview: https://civarchive.com/articles/23495/dasiwa-model-versions-and-timeline


    🔮 Key Features:

    • 🔥 LoRA-Free Generations
      Generate high-quality videos without stacking Wan 2.2 LoRAs (unless you want adding spacial styles/concepts).

    • ☄️Fast: 4 step generation

    • Extreme versatile (more build in concepts)

    • Quality motions (less slowdowns)

    • 🔞 NSFW + SFW:

      • Enhanced anatomy + poses + framing

      • Better understanding of sexual concepts

    • 🪄 Better Prompt Responsiveness

    • 🥺👉👈Better understanding of anime/manga style composition

    • 🪡 FP8/FP8+ precision

    • ⚠️ Read "About this version" details for the version you are using for more information!

    • 🚫 Do not use any extra speed-up (low step) LoRAs, this is baked in already


    🍒Workflow

    Make sure to checkout my easy to use Workflows!


    🍄LoRA's

    Try first without additional LoRAs!

    But: This checkpoint is not meant to replace all LoRAs, it is meant to:

    • Perform better overall at his own

    • As easy as possible to use

    • With LoRAs to be absolutely awesome


    ⚠️ Read the corresponding announcements.

    📢 Make sure to check it out for in-depth information and a complex comparison!


    • Steps: 4

    • CFG: 1

    • Sampler/Scheduler: Euler/Simple or Euler/linear_quadratic

    • Resolution up to 720p (native quality).

    My go to settings:

    • 0.52 - 0.83 MP

    • CFG 1

    • Euler/linear_quadratic

    • 4 steps

    • 16 fps

    • Sigma Shift: 5

    • Add other LoRAs with 0.3-1

    • 16 fps, 81 frames ~ 5s


    Dependencies


    🩻 Known issues

    • Tell me 🫵🫢


    🩺 Fixes & Feedback

    • If you use LoRAs, try to respect the LoRA training triggers and try some versatile descriptions, most LoRAs will work with 0.3-1.2 (start with 0.3)

    • Do not mass add LoRAs, just add 1 or 2 (x2 High+Low)

    • Negative prompting do not work with cfg 1, thats a limitation of speed-ups with cfg 1

    • Low resolution (e.g. 480p) are only for fast samples and will blur fine details, do a higher resolution if you want clear details

    Before posting any questions I suggest reading my guide.

    Update your ComfyUI ❗

    🪧❗ Test your comfyui-backend with this absolute basic test-workflow before asking about errors.


    🖤 Why I Made This

    I was tired of using all these massive list of LoRAs, just to get a remotely good result after 10 generations, consuming hours of time.
    So I can just make my videos with 1 or 2 concept LoRAs without pushing 6 till 10 LoRAs (Low/High) into a generation.


    This checkpoint is also my personal playground.


    Closing words

    🤩 I want to thank all the fantastic other creators who made super nice LoRAs and concepts to play with! Support that awesome creators by using their LoRAs and post to their gallery and share the meta-data!
    ⚠️ I made all this with permissions or open-source resources (the time it is incorporated).

    I share as much insights as I can without compromising my work. I'm doing this for fun as my hobby and just do not want my hobby to be destroyed.

    More details can be obtained in the corresponding announcements!


    If you would like to contribute in my awesome (😉) checkpoint or willing to share resources I'll gladly give credit! Just contact me!

    ✅ All credits / resources are mentioned inside the announcements! - Since different versions may have different resources.


    YOU are responsible for outputs as always! If you make ToS violating content and I get aware I WILL report this.


    Disclaimer

    This models are shared without warranties and with the condition that it is used in a lawful and responsible way. I do not support or take responsibility for illegal, harmful, or harassing uses. By downloading or using it, you accept that you are solely responsible for how it is used.


    Custom License Addendum: Distribution Restriction

    Notice: Notwithstanding the base license selected for this model, the following restrictive terms apply:

    1. No Redistribution: You are not permitted to host, mirror, or redistribute this model (checkpoint, LoRA, or Safetensors files) on any other platform, website, or service (including but not limited to Hugging Face, Tensor.art, or SeaArt) without explicit written permission from the creator.

    2. Attribution & Source: This model is officially maintained only on Civitai or other platforms where I explicitly own the repository. To ensure users receive the correct version, updates, and safety metadata, please point users to the original URL.

    3. Usage: All other rights regarding the use of the model for image generation remain as per the terms and the restrictions provided per model.

    Description

    ✅ Optimisations

    • 🪡 FP8+ (FP16 base) precision

    • 🌟 CFGZeroStar patch (better results and prompt adherence)

    • 🍰 Baked Latest Distillation (r64-1022)

    • 🍀Additional concept optimisation (compared to MidnightFlirt)

    • 🎬 Reward Attention (more realistic movements)

    • ✨ No or extreme low transformation of details with anime/realistic images (lips, eyes, ears, breasts, genitals, ...)

    • 🌠 Refined motions (compared to MidnightFlirt)

    • 💌 It can guesstimate better details out of frame (compared to MidnightFlirt)

    • 💫 As close to RL motion-speeds as I could get with speed-up tech

    • 🍌Even less tries for good results (compared to MidnightFlirt)

    • 🧩Even better compatibility with LoRAs (compared to MidnightFlirt)

    • 🖼️ Usable with your preferred/custom CLIP (if compatible)

    • 😵‍💫Additional reduced hallucinations (compared to MidnightFlirt)

    • 💠 Zero prompt results capable

    • 🚫Excluded CLIP

    🩻 Known issues

    • 🫦 The most delicious sin!

    • Tell me 🫵🫢

    ⚠️ Requirements:

    • Comfyui ≥ 0.4.0

    FAQ

    Comments (155)

    kostyg31337Dec 18, 2025· 2 reactions
    CivitAI

    I have only black screen as output, but if I use other GGUF ver, works fine... Any solution?

    darksidewalkerDec 18, 2025· 1 reaction

    Update comfyui to 0.4.0 or later

    darksidewalkerDec 18, 2025· 1 reaction

    The version requirements are in about this version mentioned

    kostyg31337Dec 18, 2025

    Yes, sorry, my bad, was sure its latest version - fixed, now its fine

    CubeyAIDec 19, 2025· 1 reaction

    is it something different with the mode? I still get black screens despite updating, I'm on wanvideowrapper

    saehara151Dec 19, 2025

    me too, I tried using --use-pytorch-cross-attention but now it says me: unet unexpected: ['blocks.0.self_attn.q.comfy_quant', 'blocks.0.self_attn.q.weight_scale', 'blocks.0.self_attn.k.comfy_quant', 'blocks.0.self_attn.k.weight_scale', 'blocks.0.self_attn.v.comfy_quant', 'blocks.0.self_attn.v.weight_scale', 'blocks.0.self_attn.o.comfy_quant', WARNING: No VAE weights detected, VAE not initalized. no CLIP/text encoder weights in checkpoint, the text encoder model will not be loaded.

    darksidewalkerDec 19, 2025

    @saehara151 This error is something else, you have to set clip and vae.

    CubeyAIDec 19, 2025· 2 reactions
    CivitAI

    I can't get the model to work with WanVideoWrapper, is there a reason it would work with the workflow you provided but I'm not really sure as to why It wouldn't work otherwise, with the only difference being euler simple vs dpm++_sde with beta

    darksidewalkerDec 19, 2025· 1 reaction

    I do not use the WanVideoWrapper, but I assume the problem is this node. You could go to his github and post a bug report why the node produces blackscreen with fp8 mixed quant ops system in comfyui.

    darksidewalkerDec 31, 2025

    BTW - I tested with WanVideoWrapper/Sampler and TastySin-v8.1 - It is compatible with it. It produces just fine videos. The set-up is convoluted, but completely possible.

    cao888888Dec 19, 2025· 1 reaction
    CivitAI

    Very excellent model! Five star rating! When creating videos, the original image can be accurately used as the first frame, and the consistency in the subsequent frames is also excellent. I have used many smooth and remix model workflows shared by experts, but I have been struggling for a long time without achieving the same effect as Dasiwa. This will be the only I2V model I will be using for a long time! Thank you very much, author!

    darksidewalkerDec 19, 2025

    Thank you for the kind words!

    SaxliveDec 20, 2025

    What workflow are you currently using? Would you mind recommending it to me?

    OzzyOsmanDec 19, 2025· 3 reactions
    CivitAI

    Great model. Will you convert it to GGUF quants? Thanks in advance.

    darksidewalkerDec 19, 2025· 1 reaction

    Serious question? 🤔
    Did you not tried even reading the 3rd line after the headline of the front-page or tried searching over civitai search?

    OzzyOsmanDec 19, 2025

    @darksidewalker Nope. I just look at the models at the top. Too many colors and descriptions. I gloss over it. Are we getting GGUF or do I have to wait for Bedovyy to convert it?

    darksidewalkerDec 19, 2025

    @OzzyOsman So... You were looking at his HF page? Right? 🤣

    OzzyOsmanDec 19, 2025

    @darksidewalker I try to quantize the GGUF models myself, but Wan2.2 is a tricky mistress. I like your models, have them all, but prefer how the GGUF works for speed. Anywho, keep up the good work.

    darksidewalkerDec 20, 2025

    @OzzyOsman I was joking around, because there is already a gguf version, pinned under resources, mentioned on the HF page of the guy who made quants of my older model and mentioned on the overview of my models. I really have no clue how I could "advertise" even more XD

    kostyg31337Dec 20, 2025· 1 reaction
    CivitAI

    Any ideas why the video tends to loop? For example, "a girl leaves a room," but in practice, especially if the video is longer than 7-8 seconds, she returns, often backwards.

    The loop setting is disabled, of course — this applies to many wan 2.2 builds, not just this one.

    Kiri0SamaDec 20, 2025· 1 reaction

    In Wan 2.2, generating videos longer than 5 seconds often results in looping

    darksidewalkerDec 20, 2025

    WAN is made for 81 frames, all over this tends to loop back. It is what it is.
    Exception is WAN S2V and WAN-Animate, they can produce virtually endless videos.

    Minase460Dec 20, 2025· 1 reaction

    Check you didn't enable "pingpong" in the node setting. This is what causes this.

    As someone already mentioned is the pingpong parameter what causes this. Check your node parameters, probably in the Video Combine node, if I am not mistaken, there is this parameter named pingpong, you have to set it to False and that will do it

    motimotikunDec 20, 2025· 2 reactions
    CivitAI

    This model is absolutely amazing. I'm looking forward to the next update! Keep up the great work!

    darksidewalkerDec 20, 2025· 1 reaction

    Thank you very much! I planned one, but know it the time to make some art myself and pushing the model to its boundaries ^.^

    InvadedByTheSalaryKingDec 20, 2025
    CivitAI

    Using WanGP2.2 I get results that just look like white noise. Previous versions like Midnight Flirt worked fine, but this one doesn't

    darksidewalkerDec 20, 2025

    So you may ask the WANGP2.2 creator why this might happen with fp8 mixed precision ops. 👍

    YouShouldClickDec 20, 2025· 2 reactions
    CivitAI

    You teased me about this one and I'm not disappointed ! I throw away all my Loras ! It works like a charm ! you did an awesome work once again ! thank you for sharing !

    darksidewalkerDec 20, 2025· 1 reaction

    Thank you! But please throw not all LoRAs away, some may be needed and the awesome creators also need credits! :)

    LimonobatonoDec 21, 2025
    CivitAI

    Where did the t2v version go? Will there be a new one?

    darksidewalkerDec 21, 2025

    Where? - There never was one.
    Will there be a new one? - I doubt that ^^

    LimonobatonoDec 21, 2025

    @darksidewalker Just recently there was something, it seemed to me that it was a t2v model. But it immediately disappeared, I didn’t even have time to download it.

    darksidewalkerDec 21, 2025

    @Limonobatono I never did a t2v model. 🤔

    LimonobatonoDec 21, 2025

    @darksidewalker Perhaps I got something wrong. It's a pity. It would be interesting to see one like this.

    aureliusDec 21, 2025· 6 reactions
    CivitAI

    Black screen.

    darksidewalkerDec 21, 2025· 1 reaction

    Blue screen. 🤣

    darksidewalkerDec 21, 2025

    About this version.

    markharper80266Dec 21, 2025· 2 reactions

    Also getting black screen results. Not seeing anything that suggests why in the About section

    darksidewalkerDec 21, 2025

    @markharper80266 Look closer. There are "requirements".

    SysDeepDec 23, 2025

    Blacksceen because you use WanVideo Sampler ...

    MerciCHJan 3, 2026

    I have the same problem before. So I updating Wanvideo Wrapper. It works for me.

    CorenDec 22, 2025
    CivitAI

    Hey thank you for adding the version numbers. I think i remember a comment around here that most of these seem more like flavors of focus rather than actual incremental improvement but smooth brains like me like ascending order. There are already so many choices to make, this make one easier, even if it isn't necessarily accurate.

    darksidewalkerDec 22, 2025

    👋 I decided to do it because of adding also gguf quants to the list, to give more clear hints what version is what. I'm not against numbers, but I like my version naming, so I'll stick with that too😁

    SysDeepDec 23, 2025· 3 reactions
    CivitAI

    Damn no "WanVideo Sampler" Support ... please write it on your model info page

    KSampler works fine but i hate KSampler :D

    darksidewalkerDec 23, 2025

    You could make a report at his github and ask to support mixed precision, like comfiui native, than it would work, I assume.
    I can not test any custom node, if it works native (and ksampler is) is is ok.
    I won't write it on the model-page, because only v8.1 is mixed precision, so that would not be right for other versions or future versions.

    darksidewalkerDec 31, 2025

    BTW - I tested with WanVideoWrapper/Sampler and TastySin-v8.1 - It is compatible with it. It produces just fine videos. The set-up is convoluted, but completely possible.

    mylo1337Dec 23, 2025· 2 reactions
    CivitAI

    LureNoir and TastySin are both really good, have excellent motion and prompt understanding. It understand concept such as scene changes, camera movement etc very well.

    It would be really nice if there was also a text to video variant of this model, the prompt understanding on this model is incredibly good, most of the t2v models I've tried aren't nearly this good.

    kyonn92Dec 23, 2025· 3 reactions
    CivitAI

    Note: You need to update the comfyui to the latest version, or you will get error when loding the v8-v1 checkpoint

    darksidewalkerDec 23, 2025

    Yes. Min version is stated in "about this version"

    yukiookami_8593027523Dec 24, 2025· 2 reactions
    CivitAI

    this one seems to have build in lighting lora, but when i use lighting lora with it i sometimes get much much more interesting result

    Devilday666Dec 25, 2025· 3 reactions
    CivitAI

    No GGUF?

    qekDec 25, 2025

    You made the same comment for Smooth Mix Wan 2.2 and deleted it after 1 minute
    https://civitai.com/models/2190659
    https://huggingface.co/Bedovyy/dasiwaWAN22I2V14B-GGUF
    enough?

    Devilday666Dec 25, 2025· 1 reaction

    @qek Yeah after reading I saw the link for smooth mix GGUF. Thanks for these links for this. Merry Christmas :)

    IrakTheOracleDec 26, 2025· 1 reaction
    CivitAI

    No "WanVideo Sampler" support means I won't be using this one :(

    darksidewalkerDec 26, 2025

    Well, or you could write the creator of wan video sampler to support mixed precision, like the normal ksampler does. I can not know if any custom node support less than just the standard sampler and I can not change the checkpoint at this moment.

    qekDec 26, 2025

    It's seems Kijai never fixed the problem with loading fp8 models

    darksidewalkerDec 26, 2025· 3 reactions

    I'll probably never test/support it officially.
    Because this custom node is a mess with 1.1K open issues, more VRAM consumption and a bunch of unnecessary extra nodes needed.

    darksidewalkerDec 31, 2025

    BTW - I tested with WanVideoWrapper/Sampler and TastySin-v8.1 - It is compatible with it. It produces just fine videos. The set-up is convoluted, but completely possible.

    vredbat02664Dec 27, 2025· 2 reactions
    CivitAI

    I have tried the TastySin GGUF, MidnightFlirt and LureNoir and I can say the best camera movements came from LureNoir and most facial consistency from Tasty Sin GGUF, given the same seed and prompts. Given a chance, I'd want to keep LureNoir and TastySin and I'd probably skip Midnightfirt to save SSD space.

    Looking forward to the Tasty Sin V8-1 (FP8) to see if its camera movements better LureNoir.

    Excellent work DaSiWa, I do think NSFW Enhanced Q8 and your checkpoints are true rivals.

    bbx749Dec 27, 2025· 8 reactions
    CivitAI

    Thanks for the great model! Is it possible to get an SFW version?

    darksidewalkerDec 28, 2025· 4 reactions

    You can just use it for SFW content no problem, nobody is forcing you to do NSFW content.

    bbx749Dec 30, 2025

    @darksidewalker In my opinion, different NSFW Loras that were merged into the model will harm SWF requests. Correct me if I'm wrong.

    darksidewalkerDec 30, 2025· 1 reaction

    @bbx749 I don't think so.

    bbx749Dec 30, 2025

    @darksidewalker I would gladly agree if my sfw characters didn't have semen dripping from their mouths so often.

    darksidewalkerDec 30, 2025

    @bbx749 I do not right get your point. If it does not work for you why you are not just using basic WAN22 or any other checkpoint that fits your needs? I have no problem to get SFW with correct prompting.

    bbx749Dec 31, 2025

    @darksidewalker For some reason, your model works much better than anything I've tried before! That's why I'm your fan. Perhaps you could publish a list of the sfw loras you used in it?

    darksidewalkerDec 31, 2025

    @bbx749 I did publish a complete resource list inside the announcement.

    lalouwuDec 29, 2025
    CivitAI

    Hello!

    This is the second time I've commented on one of your posts with my beginner's questions.

    I've been using your MidnightFlirt model and I love the results I've achieved. So I'm very grateful to you for that work.

    I have a question. What are the differences in terms of the results generated between this model in .Safetensor format and the other model in .GGUF format? Or do they only differ in terms of configuration?

    I read something in your Usage Guide - Definitive Edition, but it wasn't very clear to me.

    Thanks!

    darksidewalkerDec 29, 2025

    You could read my announcement made for this model, too.https://civitai.com/articles/23271/release-of-dasiwa-wan-22-i2v-tastysin-lightspeed-or-gguf-or-safetensors
    Safetensor is just the better format, faster, better memory management and in this case a bit smarter.
    The results should not differ very much.

    littlemyaiDec 31, 2025· 1 reaction
    CivitAI

    Would you consider making a T2V version?

    darksidewalkerDec 31, 2025

    Not planned atm.

    Reckon you would boss it.

    darksidewalkerJan 1, 2026· 1 reaction

    @smokeymcpott158455 There are other things to come, I have limited time^^

    kronos1959777Jan 3, 2026

    If you use wan2gp, any wan model you can use no input image and it defaults a black frame first frame and generates video same as if t2v.

    senduJan 2, 2026· 1 reaction
    CivitAI

    FYI, on 32GB VRAM (5090), 64GB RAM, 4 steps, cfg 1, 81 frames, torch compile, sage attention and native nodes it does 1920x1080 perfectly fine. Indeed, I don't consider 720p or less to be usable with any wan model due to glitches in fine details like eyes.

    Walltime from clicking run to getting a video at 1080p is only 6mins from a new prompt or 5mins for a new seed.

    RedditUser981Jan 2, 2026· 2 reactions
    CivitAI

    release q8 gguf please

    darksidewalkerJan 2, 2026· 1 reaction

    It is, you can look that up on my collection, the civitai search or by just reading the front page properly. 👍

    darksidewalkerJan 2, 2026· 1 reaction

    Oh I forgot to mention "Suggested resources" ... also pinned there!

    jasonbourne6699Jan 2, 2026
    CivitAI

    Thanks for your work

    My feedback :
    +Great face consistency, the best i've seen
    - body / action are quite "generic"

    I'll use this one for specific scene, and the other model for sex scene

    darksidewalkerJan 2, 2026

    If it is "generic" for you this might just be your prompt, it will do what you prompt for 😊
    Actions can not be generic or non generic on its own. ✌️

    gboy88Jan 3, 2026
    CivitAI

    Hi, for some reason my outputs all have like a grainy look, its especially apparent in water scenes or with vegetation and hair. So far i only tried lower resolutions like 640x640.
    Im using the recommended workflow, euler/simple etc. im not sure what im doing wrong.

    darksidewalkerJan 3, 2026

    Well, low resolution will result is pixelation. Less pixel, less sharpness.

    gboy88Jan 3, 2026

    @darksidewalker Thank you for the reply. I just tried one with the native resolution "848 × 1088" with 6/3 steps and yes it is less pronounced but its still noticeable and its also not the "normal" pixelation effect you have with low resolution its more like aliasing.
    The thing is i can't see this effect in other example creations i see here, so i assume there is some mistake on my end.
    So if this isn't a common issue i probably have to find out myself what is off with my setup.

    gboy88Jan 3, 2026

    About the model itself, the prompt adherence is actually pretty good, if you are taking in effort with your prompts, it can create a lot of different scenes.

    darksidewalkerJan 3, 2026

    I do not have a clue, but did you check if you really use high+low in the right checkpoint loaders?

    AlphaWhiskeyPapaJan 8, 2026

    I'm having this same issue too. I'm not using low res and everything is set up correctly.

    OrneryFansJan 3, 2026
    CivitAI

    Thank you for all the hard work. This is great.

    How would you adjust your recommended settings for a 5090?

    Cheers.

    darksidewalkerJan 3, 2026· 1 reaction

    Hi!
    just raise your resolution till you are satisfied and still can stand the generation time.
    Maybe 6 (3+3) steps.

    abggsop586Jan 3, 2026· 1 reaction
    CivitAI

    This one is absolutely insane. The best one so far.

    SwallowGumJan 3, 2026· 4 reactions
    CivitAI

    Thanks for your great work! I am a fan of you models but with TasatySin it only generates Black Screens. The earlier ones still work fine with your Workflows. Is there something I need to add to my ComfyUI? I can run your Workflows with the older models just fine.

    darksidewalkerJan 3, 2026· 3 reactions

    Update your comfyui.

    SwallowGumJan 3, 2026

    @darksidewalker thanks. Should have thought of that myself .....

    stableskynet171Jan 5, 2026

    @SwallowGum I'm having the same problem when testing the new model out. I've updated everything but still getting black generation. Like you the older Midnight model works fine. Did you use a different node to load the model?

    darksidewalkerJan 5, 2026· 1 reaction

    @stableskynet171 If you get a black screen you are NOT up2date. For sure.

    HornySith31Jan 3, 2026· 1 reaction
    CivitAI

    can someone help me set this thing up in comfyui? its my first time using it and i cant figure out where this is supposed to go

    darksidewalkerJan 3, 2026· 1 reaction

    You could just use my workflow. There is everything with a description.

    CiRenderJan 3, 2026· 2 reactions

    Definitely agree with darksidewalker, his workflows are fantastic. Basically, go here: https://civitai.com/models/1823089/dasiwa-wan22-workflows, download a workflow, unzip it, launch ComfyUI, drag the .json file you unzipped into ComfyUI, and the workflow will pop up. You will likely get a message that some nodes aren't installed but you can ask ComfyUI to automatically install them if you have ComfyUI Manager (https://docs.comfy.org/manager). If not, check the left side of the workflow in ComfyUI. That's where all of the requirements are listed, so you can install them by following instructions on their linked GitHub pages. Once you have the requirements (either installed via ComfyUI Manager or manually via GitHub) you then download the models in the "Model Links" section of DaSiWa's workflow and refer to the "Model Storage Location" section so you know where to put them. As the notes say, you don't need both the .safetensors and the .gguf models. (Use .gguf when you want fast, lower-VRAM, practical generation, and don’t care about fine-grained tinkering. Use .safetensors when you want full fidelity, training, LoRAs, and maximum control, at the cost of more VRAM.) Note that you need to download both High and Low models for whichever version you choose, and they go in the same folder together. If the workflow doesn't run after doing all this, try updating ComfyUI--that fixes a lot of issues.

    luisvivartJan 3, 2026· 1 reaction
    CivitAI

    This looks great. Would it be too much to ask you to release one with fp16 precision?

    darksidewalkerJan 3, 2026· 1 reaction

    Thx. 😊 But why? Fp16 won't run on consumer hardware anyways and this is for users to make awesome art 🎭.

    clevnumbJan 4, 2026

    Yes please do

    clevnumbJan 4, 2026

    @darksidewalker Us AMD Strix Halo users can use FP16 and BF16 models easily, please release...future proof it?

    darksidewalkerJan 4, 2026

    @clevnumb I'm not convinced yet.

    clevnumbJan 7, 2026· 1 reaction

    @darksidewalker Whatever. I'll just use larger models elsewhere.

    slota145Jan 4, 2026· 4 reactions
    CivitAI
    Hey :D Sorry if this is a stupid question, but I'm new to all this and a bit lost. Here's the high noise version for download, and where can I find the low version?

    1165966497192Jan 4, 2026

    在它的旁边,上面切换就行

    darksidewalkerJan 4, 2026

    It is directly under the headline

    slota145Jan 4, 2026· 3 reactions

    @darksidewalker ohhh shit... I'm stupid after all xD ty u very much! <3 I thought these were tags xD

    oldthrashbarJan 4, 2026· 3 reactions
    CivitAI

    Honestly you guys would be surprised at how well the regular wan 2.2 checkpoints work if you just go back to them and use loras.

    SillyfoolsJan 4, 2026

    Unfortunately, in my case, some lora keeps altering faces to a realistic style and adds suction lips, which I'm not interested in. This checkpoint, however, works well for me (albeit on the softer side of motion).

    I'd be grateful for any tips to fix those problems effectively.

    darksidewalkerJan 4, 2026· 6 reactions

    Well the whole point of the checkpoint is an optimized experience without stacking Loras. Sure you could achieve comparable results with stacking Loras, adding the following downsides:

    - much more VRAM consumption bc each Lora must be loaded in addition

    - slower speeds, each added Lora will introduce more computing time

    - manually finding the right weights for many combinations that already are optimized

    - loosing enhanced understanding and techniques that wan 2.2 has not baked in

    - ... even more

    But everyone is just free in using what fits them best.

    It is always wierd if somebody is claiming such things without a single contribution. And than the question arise, why are you even here if the checkpoint is not adding value for you? Just to blame it? I do not get it.

    oldthrashbarJan 4, 2026· 2 reactions

    @darksidewalker don't get me wrong, I like your checkpoints and find them useful. This is just a reminder that the normal stuff is still useful. A lot of people get used to using only these merges and end up fighting with them for good results on some things.

    AjaxdiffusionJan 4, 2026· 2 reactions

    what a random comment under a CUSTOM MERGE.

    huh

    zzozzJan 4, 2026
    CivitAI

    how can make the movement girl only?

    when making cowgirl the man bottom always moves. I don't want male move.

    I tried more cfg to high and low, tried negative prompt to 'male move', tried positive prompt to 'do not move'.

    everything was useless.

    how can I do?

    jenhsunJan 6, 2026

    try this maybe?

    girl is doing all the work moving her hips up and down while the man lying on his back doing nothing

    QwiziRAMJan 5, 2026· 3 reactions
    CivitAI

    The “midnight” version works perfectly for me. There’s slightly insufficient motion, but that’s likely because I haven’t fully figured out the process yet. However, the “TastySin” version produces completely black frames instead of video. What could be causing this issue?

    vixelfoxJan 5, 2026· 1 reaction

    same here, completely black frames

    darksidewalkerJan 5, 2026· 1 reaction

    What could be causing this issue? ~
    1# outdated comfyui installation
    2# outdated custom nodes

    Update everything.

    stableskynet171Jan 5, 2026

    Same problem. Updated everything and using your latest workflow as well.

    darksidewalkerJan 5, 2026· 1 reaction

    @stableskynet171 Double check, blackscreen is always outdated nodes/comfy.

    TOF_enjoyerJan 5, 2026· 1 reaction

    I update my comfyui and everything to latest and it worked for me, make sure you update comfyui and dependency.

    stableskynet171Jan 5, 2026· 1 reaction

    @darksidewalker It seem my comfyui download the update but didn't automatically switch to the new version. I had to manually switch it to 0.7 version and it works. Thank you for the model and helps.

    JiankaJan 5, 2026· 2 reactions
    CivitAI

    Works great with SVI. ❤️

    Just_hobbyJan 6, 2026
    CivitAI

    Will the real person be worse than expected?~~

    darksidewalkerJan 6, 2026

    What does that question mean?

    Just_hobbyJan 6, 2026

    @darksidewalker I'm sorry, I feel a bit stupid. I used translation but didn't check the content carefully. I want to know how the realistic human character effect turns out—will it still end up looking like a cartoon style, or have color distortion, or something like that?

    darksidewalkerJan 6, 2026· 2 reactions

    It will use whatever you set as initial image.

    MekalonikaJan 6, 2026
    CivitAI

    Does it work with 12GB VRAM?

    darksidewalkerJan 6, 2026· 1 reaction

    Someone wrote it work if you have enough RAM and or swap/pagefile.

    gambikules858Jan 6, 2026· 1 reaction

    32 ram + 80gb on your ssd

    Just_hobbyJan 6, 2026· 1 reaction

    我刚跑了下 12G显存 48G内存 560*720 81帧 只需要410秒,使用sage只需要336秒,但是我无法生成正常的阴茎

    dzst96Jan 6, 2026· 1 reaction

    rtx 3080 10gb vram+64gb ram and it works 768x528 resolution fine for even 7 sec long video, generates in about 10 minutes, i havent tried higher resolutions yet,

    MekalonikaJan 7, 2026

    Thanks guys, i have 32GB RAM and plenty GBs of SSD. I'll try tonight.

    dzst96Jan 8, 2026

    @Mekalonika generation will be much quicker (at least in my case) if you add sage attention

    MekalonikaJan 23, 2026

    @dzst96 thanks, never use sage but ill try.

    Just_hobbyJan 6, 2026
    CivitAI

    I've been getting errors constantly when using the latest development version, but it works fine as soon as I switch to the latest stable version.

    aidrochJan 7, 2026· 8 reactions
    CivitAI

    Nice model! Compared to SmoothMix, it has much smoother animation, less detail loss, better anime anatomy, faster generation time by about 5%, and uses less VRAM. Perfect WAN fine-tune. Looking forward to the next version.

    YoureMyPayPigJan 7, 2026
    CivitAI

    LOVE THIS MODEL: Does anyone know how to keep mouth from moving/talking? No matter how many prompts I do, they still get chatterbox syndrome

    darksidewalkerJan 8, 2026· 2 reactions

    Could be your sampler, res_multi tends to exaggerate such motions. Or you have to prompt something for face details. I personally don't have this problem and most likely it is the sampler or just prompting issue.

    g1263495582Jan 11, 2026· 1 reaction

    Add live wallpaper LoRA + live2d LoRA at low noise, it will help. Also, use an image where the character's mouth is closed.

    YoureMyPayPigFeb 11, 2026

    @darksidewalker thank you, dev.

    YoureMyPayPigFeb 11, 2026

    @g1263495582 thank you, I'll try it now.

    g1263495582Feb 11, 2026

    @YoureMyPayPig Recommend using 3k sampler + try to use images where the character’s mouth is closed + the LoRA mentioned earlier.

    stringshJan 8, 2026· 1 reaction
    CivitAI

    Amazing model, thank you!

    abggsop586Jan 9, 2026· 3 reactions
    CivitAI

    the quality and the motion of the videos have been great but I have a question.

    How do I keep the style consistent to the original image without it changing the eye color of the character?


    midnightflirt was more consistent with the original style but had worse motion output while this one has good motion output but weak consistency

    darksidewalkerJan 9, 2026

    This must be a prompting issue or a lora you are using. Maybe you should prompt the eye color. On my gens the eye color never changed anyhow.

    fewe85471Jan 9, 2026

    Also consider SVIpro 2 and/or VACE

    abggsop586Jan 11, 2026

    @darksidewalker so it was due to a lora after all. I am using the walk lora to generate walking motion videos as I am unable to get decent walking motion results with just the base checkpoint. Do you have any prompting suggestions to avoid using loras and still get decent motion output?

    PillowbiteJan 9, 2026· 1 reaction
    CivitAI

    Does paying the 2k buzz for early access allow you to download both High and Low? Or is it 2k buzz each(4k total)?

    AlberistJan 9, 2026

    2k each.

    darksidewalkerJan 10, 2026

    It is totally up to you if you like to support me or wait :)

    grasshopper85116Jan 9, 2026
    CivitAI

    Is this mostly for stylised concepts, or does it work well for realism too?

    darksidewalkerJan 10, 2026

    As you can see on my announcement/description and on the front page and also my examples... all.

    ql_weepingwind036Jan 10, 2026
    CivitAI

    I noticed that a new version named TrueVision begin EA,Except for not baking distillation in,it‘s the same with the lightspeed V9,right?

    darksidewalkerJan 10, 2026

    All speed-enhancements are excluded. Otherwise the same.

    Also this way are all downsides of distillation not there, so it is a more pure checkpoint for mixing with other technologies.