@c0ur4ge I wanna encourage people to even not use your examples or even your trigger words. The lora is so good, it doesn't need them. There are so many creative wording and phrasings you can use and the lora just makes it work lol. Genuinely impressive. Good job!

You will drive yourself crazy if you focus on more than one parameter at a time. As me how I know... There are too many permutations. If the quality of the image is ok, don't mess around with the ksampler. If you have a high shift with i2v you can have a very static output. Leave it around 5-6 for both. Playing with shift is much more useful with t2v. Think of image quality and temporal consistency and one problem and conditioning as a separate one. Parse your conditioning clearly, use parentheses and weights. Use default negatives. Start really simple. One clear action with concrete detail, no ambiguities or abstract concepts. If it's ignoring your input, then you can lower the strength of the models. Otherwise, it's about prompting. A concise three sentence prompt can give you way better output than a jumble on conflicting copy/paste junk. Don't try and iterate 10 settings at once, you will learn nothing. That's my I2V wisdom as it applies to a general-purpose model like this one, with respect to what I assume you mean by 'more assertive.' If your output is crappy, that's a different problem that you need to fix first.

YMMV, but a good starting point (as the first thing in your prompt) is: " Immediately cut to a new scene where the exact same subject is now ..."

Did you try it?

@HappyBot not yet.

Bro, have you solved? I'm no expert but I can share what I know! I know the feeling of being lost haha write me a messagge!

This model is a LoRA model that requires the Wan 2.2 video base models (low and high) and software like comfyui (little else compares, but there are options) to run it. You can download the models and download workflows here on civit and all of the software is free. ComfyUI can be installed in a variety of ways and alternative GUI front ends exist for it. To use this LoRA you need to load the LoRAs and the models into a workflow and set up the parameters to your liking (and adjusted for your hardware) and run the workflow.

This is LoRA, you need the Wan 2.2 14B model . LoRAs are added after the models in the workflow.

you need comfyUI or wan2GP and you need to watch a lot of tutorials on YouTube, there are many 100s, Thats what I did, or pay someone on patreon to help you , thats what i also did, the try and try again, to personally tutor you would take many hours of work my friend

@robyc Please help me. I have wan 2.2 safetensor. I also have LORAs. I just need a workflow for image to video.

@onelanone1 - https://civitai.com/search/models?baseModel=Wan%20Video%202.2%20I2V-A14B&modelType=Workflows&sortBy=models_v9%3AcreatedAt%3Adesc

just another note since this might get a few eyes on it, if you are using any speed LoRA like lightx2v I would recommend connecting or loading the speed LoRA AFTER all other LoRA or at the bottom of the list if you are using a stacked LoRA loader, this way the motion priors and everything else get applied before the speed LoRA goes in and stomps on everything. it helps to keep things more uniform as the speed LoRA aggressively denoises. not strictly necessary but something in my experience that helps in general usage

Thank you so very damn much! So glad you posted this.

@cruedtattoo606 I read someone else's poorly explained reply here that helped me figure it out but took a while. Thought i would save others some time, seems like a common issue.

@c0ur4ge I didn't know it made the results worse, it was the default for the workflow i had. What Lighting Lora do you recommend? my issue was i replaced this lora with the lighting lora so i had no lighting lora. I thought Loras were just to modify the base model, i didn't think one was needed to make it work at all.

@Destromite  us these wan lora's for 2.2 wan2.2_i2v_lightx2v_4steps_lora_v1_high_noise and wan2.2_i2v_lightx2v_4steps_lora_v1_low_noise. Use the Dr34ml4y the same way. High first and then low. Just make sure to match wan2.2_i2v_lightx2v_4steps_lora_v1_high_noise with dr34ml4y high noise and wan2.2_i2v_lightx2v_4steps_lora_v1_low_noise with dr3aml4y low noise. You need to stay with wan lora's if using wan. It's a complicated system. You have triple lora loaders, power loaders, etc. What i have found best is use the factory workflows, ( I am using aorus master 5090 and stock workflows are faster then custom) just change the models and vaes etc and results are great. Custom workflows are great and fancy and will do things built in workflows won't but they also can overheat your video card will break your system if you are not careful. I have 5 versions of comfyUI because I have three 4TB drives and I want to see what works best. ComfyUI portable is the best and to be honest, use Venice AI to hjelp you input commands for Powershell. Literally just say make a command for installing a wheel triton or sageattention for python 3.13 and it will make the code and you run it. I learned coding from grok and Venice and now Venice just added text to video and image to video and its annonymous. Python 3.11 3.12 3.13 3.14r1 and 3.14 stable. 3.14 is the fastest BUT transformers will not work past 3.12 and sageattention and triton has to keep being reinstalled. I literally think DR34ML4Y is one of the best lora's out there for the way I like to type. Sometimes I make very shitty lazy prompts and this lora just does them. I found being very very detailed using this lora ruined the style I like. Other lora's are different though. They must be worded strongly or you will have to bump the cfg up. I know this was long and sorry but no one helped me at all. I learned it myself and the one comment I rad of yours. @c0ur4ge This lora is the Onyx Tier. Nice job!

@clzpetn804 As I understand it, LoRAs apply updates to the weights additively. So the order they are applied should make zero difference. I tested this myself by switching the order with the same seed and got identical generations.

like this? https://i.imgur.com/GOAo4SR.png Image To Video Comfy Default WAN 2.2 Workflow. Videos are still blurry and don't do what I ask for. I changed to 8 and 16 stepsfor both KSamplers.

working good, in the default confyui wan2.2 i2v workflow I just added another 'LoraLoaderModelOnly' nodes in series to current loaders and kept their values same (1), it works with other cfg values (tried between 1-3)