💀💋 DaSiWa-WAN 2.2 I2V 14B | Lightspeed | GGUF 💋💀
GGUF 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
💫Quality motions (less slowdowns, no pixelated hyper-motion)
🔞 NSFW and SFW + Extreme versatile (more build in concepts):
Enhanced anatomy + poses + framing
Better understanding of sexual concepts
🪄 Better prompt responsiveness
👘 Better understanding of anime/manga style composition
🪡 Q8 (FP16 base) precision
🚫 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
🪧Announcement
⚠️ Read the corresponding announcement.
📢 Make sure to check it out for in-depth information and a complex comparison!
🆕 New to WAN 2.2 I2V? - Check out my guide.
🛠️ Recommended Settings
Steps: 4
CFG: 1
Sampler/Scheduler: Euler/Simple, UniPC_BH2/Simple
Resolution up to 720p (native quality).
Add other LoRAs with 0.3 - 0.6 at first
16 or 24 fps, 81 or 97 frames ~ 5s
Dependencies
🩻 Known issues
Tell me 🫵🫢
Approximate expected quality from quantization
This are my tests compared to a full fp16 safetensor checkpoint taking prompt- and visual satisfaction into account on my DaSiWa checkpoints.
Quantization - Quality estimation
Q8 ⭐⭐⭐⭐⭐ ~ like FP16/FP8+, excellent results
Q6 ⭐⭐⭐⭐ ~ like FP8, very good results
Q5 ⭐⭐⭐ ~ good to very good results
Q4 ⭐⭐ ~ medium-good results
Q2 ⭐ ~ poor results, only use if you have to
⚠️ 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 0.3-0.6 (start with 0.3)
Raise LoRAs in little steps +0.1
Do not mass add LoRAs, just add 1 or 2 (x2 High+Low)
Negative prompting do not work with cfg 1, that's a limitation of speed-ups with cfg 1 (except you use NAG)
Low resolution (e.g.384x576) are for fast samples and will blur fine details, do a higher resolution if you want clear details
🪧❗ 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
FAQ
Comments (18)
I assume it has a Lightning lora baked in, which is a downside in my opinion - less control for motion. Did some initial tests and movement feels very stiff compared to other i2v GGUF models that I use. Will fiddle with it a bit more, but at this point I regret spending buzz on the model :<
Okay ... But you seem surprised about lightning, when even the name is lightning. What did you expect? Also you could have get the non-lightning model if this bothers you.
At this point such a comment bothers me. Since the motion is fine, you did just not read and are angry now?
Also you are sure you did dl the correct quant from the download dropdown?
There is a BUG on civitai, that it will always DL the 9GB file. It is already reported and I made a ticket. They will look into it.
You will not lose any buzz, just dl again when possible.
Thank you for your support!
The civitai-dl bug got fixed by the support.
I dl'd the 9gb and then dl'd the 18gb true vision and got really confused, very unfortunate timing
light speed tensor fp8 v11 vs gguf q8 v11?
When I use both, fp8 is faster by 3 sec for 4 steps in hd++ for 5080. But I see some movements are more natural on q8 like finger doesn't go into the body, like the stomach or chest, but you put in comparison that FP8 has better quality, so I am not sure. What do you recommend?
btw I really love your work. Thanks a lot. You saved my life.
In general a proper made Q8 should be beating out FP8 on paper for quality, speed will favor FP8 though. Quick edit to say as long as your card supports native FP8, rtx 4000 series and up its faster
@FloatsYourStoat Not true, this only counts for fp8 simple scaling. Since fp8 mixed and learned exists it is just superior.
@Darksidewalker I see, I assumed it was just a simple fp8 conversion as most people seem to make.
@FloatsYourStoat I scale them as advanced as I can. Calculating learned math and sota schedulers.
Overall my fp8 mixed checkpoints are more advanced than gguf, but there is always the "bad seed" factor.
@Darksidewalker I take it you put that same level of effort into your NVFP4 version of this model too then? because that sucker is seemingly far better than it has any right to be
@FloatsYourStoat Yes, my NVFP4 are also not just simple scales, advanced processing is made there too
@Darksidewalker Is Civit showing people what FP8 format they're getting yet or are all the downloads only labeled FP8? If these new FP8 formats are so much better then people should be able to know what they're getting. They would also potentially negate the need for Q8_0 GGUF — which the regular FP8 can't do because it has quite a bit higher perplexity.
@Darksidewalker Is NVFP4 better for nvidia cards in terms of quality or would you recommend the FP8 or GGUF?
@ss9999 Civitai is not showing it
@DaddyWolfgang More compression = less quality.
I'm new to this. May I ask what is "Chunking"? You wrote in the workflow that it helps with longer video. Meaning we can use it to generate longer than 5s? Thanks.
Chunking is not for longer vids, it is for VRAM savings.
WAN22 is 5s native.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.