❇️💀 DaSiWa-WAN 2.2 I2V 14B Lightspeed | NVFP4 | Safetensors 💀❇️
NVFP4 Version of the model.
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
🪡 NVFP4 precision (FP16 base)
⚠️ 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!
🛠️ Recommended Settings
Steps: 4
CFG: 1
Sampler/Scheduler: Euler/Simple or Euler/linear_quadratic
Resolution up to 720p (native quality).
My go to settings:
Aspect ratio 3:4 (560:720) or 9:16 (608:1072)
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 🫵🫢
Approximate expected quality from quantization
Refer to this comparison article.
This are my tests compared to a full fp16 safetensor checkpoint taking prompt- and visual satisfaction into account on my DaSiWa checkpoints.
⚠️ Do not compare this with the unofficial quants of my checkpoint made by others, they are based on FP8 and not FP16 like my quants!
🩺 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:
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.
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.
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
See FP8
FAQ
Comments (32)
Will you also be releasing GGUF versions when the time comes?
Yeah ... my upload capabilities are just limited
there is application called ggufy to quantize models into gguf version. https://github.com/qskousen/ggufy
@opie1 There is a gguf version already ... guys...
@Darksidewalker oh i was just sharing that software so that people dont have to annoy/ask creators often. thank you for uploading the gguf version! (salute emoji).
@opie1 I noticed that you share that link, but without knowing what you do, you may produce trash gguf version or wrong things that are inferior, harming the creators. Even worse if this bad q-quants are shared, sometimes without permission... A average person cannot know if the quant is ok or not...
@Darksidewalker interesting , this gets me curious. "you may produce trash gguf version or wrong things" please elaborate and how can we fix it or something. i think this is also a good idea for article writing.
so far all the quants worked quite good (tried it with over 120 quants now).
-instead of using comfyui-gguf city96 +llama.cpp cli method (its too annoying to put in command and the anxiety/possibility of quantizing with above method might crash pc every god damn time).
-i just use this since it allows to do the same with few clicks and WITHOUT using huge ram.
( maybe i should add in my text that stick to default settings or do not try to decrease the size of quants ).
-audience/users/people should always first match SHA256 or similar before talking shit about creator's models.
@opie1 no average user is comparing hashes
I would like to buy this model even if it cost a lot, but Civitai kill most basic payment method so i need to wait 15 days :/
By the way in workflow from this creator i would like to have easy method to change seed generation method from random to fixed because i can't find it and i would like to have workflow for long generation by few video one by one like that SVI.
Boundbite lightspeed appears to OOM my generations. Even though Synthseduction lightspeed generates just fine with 0.83-mp HD and 1080 resolution preset. Is boundbite actually heavier
Maybe you cannot use NVFP4 and need to use FP8?
@Darksidewalker
Ooh. Oof I see now. Sorry I didn't know that was a thing.
@Dasbestos Sorry for the confusion, but I have no other way to organize this. There are fp8, gguf, nvfp4 and truevision already.
I usually end up with a low movement results and donw know why. tried moving shift up and down and even "forcing" by prompt but its always the same.
I have tried multiple models from you,
and in some cases, the video generated by v8 tastysin is even better ：the movements won't slow down ( issue in v10 and v11), and the camera won't move uncontrollably (issue in v9 ). Could you please make NVFP4 version for v8 tastysin？Thank you !
What quant is your v8? It sounds like you're comparing apples (less quantization) and oranges (more quantization). Are you sure the quantization isn't the reason for the issues you're seeing. Throwing away data (quantization) can have consequences.
i get an error: [INFO] Prompt executed in 0.51 seconds
[INFO] got prompt
[ERROR] Failed to validate prompt for output 2503:
[ERROR] * (prompt):
[ERROR] - Required input is missing: images
[ERROR] * SaveImage 2503:
[ERROR] - Required input is missing: images
[ERROR] Output will be ignored
[ERROR] Failed to validate prompt for output 2502:
[ERROR] * (prompt):
[ERROR] - Required input is missing: images
[ERROR] * VHS_VideoCombine 2502:
[ERROR] - Required input is missing: images
[ERROR] Output will be ignored
same here
The model file size is much more optimized now. Using the creator’s latest workflow, generating a 6-second video (0.65 MP, 720p, 16 FPS) took 209 seconds on the first run. Even with simple prompts, the results are already quite good. With more advanced “director-level” prompting, the output could likely be pushed even further.
My setup: RTX 5070 Ti + 32 GB DDR5.
I recommend anyone with an RTX 50-series GPU give it a try.
Thank you for publishing good model. I have posted animation generated with this model. This model is quite good with low spec GPUs.
Took only 150 sec for 81 frames with my workflow.
Do you know how excited I get for new Dasiwa models? Amazing!
I'm not quite sure if my settings are wrong.
The videos I generate in ForgeNEO with this version all look like 0.5x slow motion. I tried adjusting resolution, shift, and frame rate, But none of it works
I cannot help, I don't use forgeNeo
@Darksidewalker I've found the problem. When I set the frame rate to 16 and total frames to 96, the video playback speed is normal. But once I try to generate more frames, it stretches the actions understood by WAN2.2, rather than adding more action details at normal speed to make up the duration.
Do you recommend any optimal configuration to run this model on a laptop with a 5070 8gb vram?
My computer is equipped with a GTX 5070 Ti 16GB GPU and 96GB of DDR6 memory. I can run the FP8 version of DaSiWa-WAN 2.2 I2V 14B very quickly using the Kijia workflow. However, when I run the NVFP4 version, ComfyUI throws the error torch.OutOfMemoryError: Allocation on device. This error means that the GPU ran out of memory. Why is this happening? Thank you.
我使用的5090，这玩意直接占用了我24g+的显存
I have a 3060 12GB. should I use this one or the Q5 GGUF one?
nvfp4 is a format for the 5000 series; for earlier versions, it will actually slow things down
amazing to use if you have a 5000 series gpu <3 it works very quickly on my own setup.
does 4 step mean 2 high and 2 low or 4+4?
2+2