I managed to re-do my little Halloween music video with the Muse V2 workflow. In working with Andy, he made some tweaks to the workflow to help improve the use of cuts within a chunk. Note that I'm learning as I go and there are a few things I’d still like to go back and tweak, but as Shrek says, that will do donkey. This was done with the 5070ti and 64G using 10sec clips at 0.8 megapixel. Thanks to the tweaks by Andy, the barrier of changing camera views at cuts is gone, leaving the option to start a new scene or continue from the previous cut or chunk. The blending between each chunk and cut is flawless from what I can see. No chirps in the audio or visual burps at the transitions. I could have easily made the entire video with one fluid seam, moving the camera around. The biggest hurdle for me was prompting and making sure each image was analyzed with a decent description. The stream feature is powerful in that it let me work each chunk until I got the results I was looking for. Gone are the days of having to cross my fingers by doing full renders over and over. I did follow Andy’s earlier package demo by using the source audio file and then voice only to automate the creation of required chunks. It was not perfect in creating the cuts, but that's easy to take care of with the add/remove tools. This time I did not use the LLM feature for getting prompts, but rather re-used some of the prompting from the other workflow. Given the length of the video, I’m not sure if the LLM route for automated prompting would have been practical, given the detail needed for each chunk and cut. Prompting was the biggest hurdle as the tool for creating worked flawless IMO. What is great, I now have the ability to upscale and simply re-render the entire video overnight with a single click.

I have now got the character sheet V2 to take an image as character, a body shape and clothing and combine into one turnaround for custom character sheets.

Advanced subscribe - how to download MiniMax Director 2?

I thought I’d share my first real output with the new V2 workflow and some highlights. I was not looking to do anything graphically stellar as it was more about getting use to the WF. This was done with the 5070ti, the Singularity_ref2va_Pruned_v1.3_int8 which seemed to work decent with the 8 step lora, albeit, I have not tried the 4 step yet. The non pruned version threw errors and some of the others resulted in unwanted audio, even with the turbo off. First, I opted to take advantage of the external-LLM Planner, providing two reference images and a detailed outline of what I wanted to happen into Grok. With the generated json results I was able to import it back in and voila, all the chunks populated. This was so quick and easy getting the entire project with all the chunks in place. I also used the streamed feature as it literally let me work chunk by chunk to vet each result along the way. Overall, I would have been done much faster, had I taken time to actually vet the prompts a little more carefully. Also, several chunks had unnecessary cuts causing a jerk in the transition as there was no change in the camera. The mechanics under the hood of binding chunks together without jerks worked flawless from what I could tell. This was one of my main concerns. I found it important to work down the chunks and get the results you want before going onto the next. In my case I had 6 chunks @10sec. Keep in mind that I’m learning along the way and I realized after that I should have checked to make sure that all LLM generated prompts actually flowed and made sense. In this case, Franky was going up the stairs twice. If you are working on Chunk 5 and discover that chunk 3 needed a tweak, then you must go back and re-do the next chunk (4-5 etc.) Makes perfect sense as you just altered 3. Fortunately, the WF does it for you if you just decide to hit RUN or Render up to Here. In the end, I’m feeling so much more comfortable with this WF, albeit I’m still learning. Hoping that some others will post their results and share what they discovered etc.

Ok everyone. I hope those of you that have the Director V2 node are enjoying it so far. As promised, I am going to get a load of video tutorials recorded. Are there any requests for what would be more useful first? (runpod installation is coming 1st). Let me know in the comments 👇