Recently, Sheep and I have been working together in a company to create a dynamic comic bidding video, and also initially follow some of the workflow of film and television production companies to formalize the production, to film and television standards to carry out the project in its entirety.
Confirmed episodes
First of all, it is to determine the episodes we want to produce, here the A party gave us a lot of choices, after some comparisons and analysis down the road, we chose the strongest immortal Chen Beixuan season 4, episode 2 of the second episode of the fight between the two men and the gorilla about 50 seconds of footage.
Once we had decided on the content, we analyzed the subplots and met with the team to align them with the information available to us and to the client.
Discussing what to focus on and what to gloss over in those fifty seconds will ultimately make for a cinematic experience, and will also demonstrate to the outside world the strength of our professional film and television production.
Producing split-screen, pixel-by-pixel alignment
We'll get down to a precision level like 0.5s, unfolding the 50s of Dynamic Manga to about 67.5s, with emphasis on the intense fight sequences that follow.
These are the results of a day and night of discussion by a team of about seven people, who went to every shot to find a benchmark video.
Professionals and amateurs, there is a big difference in terms of the amount of films read, if we want to follow up with a deeper plowingAI VideoMovie and TV production, that still requires going to see more great work and building a library of benchmark material.
When the time comes to encounter a similar shot will be able to find out the benchmark, so that the team can not understand the content of this subplot within the team, and lead to the end of the actual effect of the work done and the ideal gap is too large.
Then we simultaneously move forward with characterization, compositions, drawing styles and scene subplots, etc., pre-drawing similar compositions based on our subplot sheets going on midjourney.
When the ambient group is done, we use --sref to unify, so that we can advance the project at the same time in the fastest time and maximize the use of manpower.
This is the subplot that Xiaoshi is specifically responsible for in this project, just three subplots, but against two videos, "Douluo Continent" and other excellent works.
It is possible that many non-design majors are not sensitive to this aspect of composition, and it is even more difficult to describe it with precise cue words to generate the desired image.
Three Godly Ways
So there are these ways up front:
1. mj's /describe to push back against the standard picture, and then translate it into Chinese, and then go to the inside of the translation, change the Chinese, and finally paste the English cue words to the mj inside, but here we need to pay attention to is that our mj cue words to be structured, accurately describes the style of the painting, the main body, the composition, the sense of the atmosphere and so on!
2. Another way is to use joy caption two to reverse the cue words, note that if you need to reverse the composition and light and shadow, then you need to check these two boxes
3. Backpropagation using other models such as ollama or ChatGPT is the same in principle and will not be repeated.
Optimize Cue Words
But to the actual process will often be faced with the complexity of the picture need to choose, at this time, the composition style is almost the same case will need to consider more "sense of atmosphere" and before and after the consistency of the split-screen and the degree of articulation.
And that would really require a needle in a haystack, a step-by-step revision that requires a high level of ability with cue words.
There are too many images to intuitively go to a convenient choice, here we use PureRef, a professional tool to compare, so that the "sense of atmosphere" between the split-screen more unified
Conclusion
If you need to be more refined, this is when the advantages of design professionals become even more obvious!
Use PS to process the image even further, with subject details, color tones, and sense of atmosphere at a higher level.
And to make a good work, just rely on these previous is not enough, especially in the characterization of the above need to spend a lot of effort, not only need to pay attention to the character's style, and the similarity of the original, but also need to pay attention to the degree of recognition, in this regard, you need to communicate with the party frequently.
