💠💀 DaSiWa-WAN 2.2 I2V 14B | TrueVision | Safetensors 💠💀
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
This are my 💠 TrueVision - Series.
This is meant to be a checkpoint that is not altered by technologies or merges that reduce quality by any means.
Only enhancements are made that are tested by me to be stable and enhance visual quality and add features like NSFW.
This naturally excludes most or all distillation and sampling speed-up enhancements.
Do not expect this to be much faster as basic WAN 2.2 iteration speeds.
🔮 Key Features:
🔥 LoRA-Free Generations
Generate high-quality videos without stacking Wan 2.2 LoRAs (unless you want adding spacial styles/concepts).💎 Pure Quality: 20-30 step generation
💫 Quality motions
🔞 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
🪡 FP8+ (FP16 base) precision
💠 Highest alternate LoRA compatibility
🚫 Speed-Up LoRA's excluded (use as you like)
🚫 CLIP excluded (lower size, choose your preference)
🧩 Mix and match with alternate technologies
🍒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!
🆕 New to WAN 2.2 I2V? - Check out my guide.
🛠️ Recommended Settings
Steps: 20-30
CFG: 3.5
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
🫦 This Synth is adorable!
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 0.3-0.6 (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. Also was not satisfied with the results other checkpoints can achieve across the board.
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 (35)
Ah, what's the difference between the 8 gb and 18gb file for the v11?
Nvfp4 vs fp8 precision
How optimized is the distilled version of this? I have been using the truvision version with varying weights on the speed loras for months now, so I can use negative prompts and get more precise motions. Would you say this version is better with the lightspeed version overall? I want to know which one I should get so it's important for me to know what you recommend. If I do get this version that isn't distilled would you say the newer speed loras from kijai are different than the 1022 and 1030 loras from lightx2v? Oh and if we get the new kijai loras for this which size?
I made good samples with the lightspeed version and did only few tests with the True Vision because of limited time.
I prefer the lightspeed version, but it is hard to say if this is what YOU will want, since this depends on your setup and goal.
If you want full control and setup your own strength for the loras you may have to use use TrueVision, but will need more time and VRAM.
I always use the highest possible quality versions.
Hi - just got Snatchkiss. Is there a reason it is 5GB larger than BoundBite?
Thanks
Contains more high precision layers
anyone know where all the other nsfw models went? Luckily I have them all downloaded but I kinda expected them to be back up on red by now.
All my models are available on red
Normally there is no ghosting and there can be many reasons from settings to faulty installs.
I've been experimenting with this for a full day (by starting with DaSiWa's awesome recommended workflow that uses SnatchKiss as a default) and the starting steps_total at 6 is much too low for me... I'm currently testing at steps_total 10 with refiner_step 5 and cfg 1.2 and things are much better!
Q: What settings for this awesome checkpoint has the community settled on for 'really good quality without throwing performance away'?
So this is going to be about motion, but bear with me in case this has been covered. Going from smoothmix (I know I know) on High pass and v10 on Low pass - generally I found simpler motions to be more dynamic out of the box - with it shitting the bed as soon as complex motions/genitals at nonstandard angles are introduced (of course). From looking at your comparisons that all seems to add up - yours is focusing on stability and visual fidelity first and foremost - my question would be - since prompt adherence here seems to be better - if there’s anything specific I can prompt for to give it a kick up the ass a little bit? I am not necessarily looking to go to smoothmix levels of dynamism ( would be nice if I have the option reqardless), but anything to make it energetic? The results I get seem to be basically what I used to get with base WAN2.2 + lightning loras - lethargic motions, but now with way more visual fidelity (a bonus!).
Using your FP8 Truevisions, Generating at 16fps then interpolating to 32fps if that makes any difference.
Thanks a bunch in any case, I love how clean my results are so far with FP8’s are.
look up painter i2v
@pocketpie already have, and it’s dogshit. degrades results without doing what I want anyway to boot. I simply use Q8 Smoothmix on High pass when I need more motion.
After reading your comment I added a third sampler and loaded up Smoothmix first with it's own dedicated path. I put it at cfg2 and it does a good job of adding those little details.
bro can i request you to make on wan 2.2_Bernini_model
Is the True Vision version recommended for a 4090 (24GB VRAM)? My objective is A++ spiciness with A- quality minimum and A-level performance. (I am ok with a small quality loss as long as I don't sacrifice the animation quality or throw away performance)
It's less creative and more precise compared to older models. Good if you want animation to follow precisely your first/last images to make animations.
@naxo556x Thank you. Precision is probably favored here so will stick with it for now. I notice that there is a GGUF version but the quantization is low-quality so I think this version is best for my 4090. Which variant do you use the most?
You can also use True Vision only for High noise and use the distilled Low noise. I get pretty good results with less artifacts especially with BoundBite Low noise which is for me the best ever he made. Even SnatchKiss Low noise tends to generate artifacts.
Would it be possible to create an INT8 ConvRot + additional layers version from the FP16 model?
Lots of ghosting in movement for Snatchkiss-Truevision. I have to set refiner_step to 4 and steps_total to 8 to make it work. This model probably won't be everyone's cup of tea, it works well for first/last image-to-video since it follows images very precisely. It's good if you are looking for precision, but it can be a bit boring if you want the model to be creative.
Not a bad model. I switch between snatckiss-true vision and boundbite depending on what I want.
If only first/middle/last image to video worked great, this model would make very precise animation loops without relying on third party software or additional tricks.
This is not a distilled version 8 steps are way too less for non-distilled. You maybe confused this. It would need like 20-30 steps total.
@Darksidewalker gotcha. Your workflow by default was set to refine 2/total steps 4 with notes recommending that as well, so I asumed I should stick to that.
@naxo556x Well, it is because I use my lightspeed quants 😸
@Darksidewalker right... I need to redownload and reorganize my stuff. I used gguf, then have mixed lightspeed with truevision thinking they're the same so had different performances. Thank you for your insight. I did not notice True Vision/Light speed were vastly different.
@Darksidewalker Did you personally noticed any significant quality difference between Truevision and Lightspeed models? Especially for first/last image workflows? Probably people ask that a lot.
@naxo556x I can't tell since I don't use TV personally, I created in on demand
bro help me i am gettings little blur quality as compare to v10 even i used lighting lora and can you please inbuild there a armpits licking lora inside or any that can understand what armpits is
No armpit lora baking 😆
make sure to use non-distilled settings with "TrueVision" or use the lightspeed version.
@Darksidewalker 😆😆
I wanted a version of Dasiwa that supports CFG > 1 so that I can add a negative prompt, and this looks like the only version without the lightning LORA based in. Is that correct?
Problem is that these model files are larger than your others @ 18GB. So I may have trouble running them.
TrueVision are the only without baked lightspeed from me.
I'll not maintain low quants for this, sorry.
@Darksidewalker Why is it the model file larger than the other (lightning) versions?
This model is fantastic and I hope for even more advancements to it.
Thank you!